Category Archives: Environmental Health

An Integrated Chemical Environment to Support 21st-Century Toxicology

Author Affiliations open
1Integrated Laboratory Systems, Inc. (ILS), Research Triangle Park, North Carolina, USA; 2Sciome, Research Triangle Park, North Carolina, USA; 3Program Operations Branch, National Toxicology Program (NTP), National Institute of Environmental Health Sciences (NIEHS), National Institutes of Health (NIH), Department of Health and Human Services (DHHS), Research Triangle Park, North Carolina, USA; 4NTP Interagency Center for the Evaluation of Alternative Toxicological Methods, NTP, NIEHS, NIH, DHHS, Research Triangle Park, North Carolina, USA

PDF icon PDF Version (277 KB)

  • Summary: Access to high-quality reference data is essential for the development, validation, and implementation of in vitro and in silico approaches that reduce and replace the use of animals in toxicity testing. Currently, these data must often be pooled from a variety of disparate sources to efficiently link a set of assay responses and model predictions to an outcome or hazard classification. To provide a central access point for these purposes, the National Toxicology Program Interagency Center for the Evaluation of Alternative Toxicological Methods developed the Integrated Chemical Environment (ICE) web resource. The ICE data integrator allows users to retrieve and combine data sets and to develop hypotheses through data exploration. Open-source computational workflows and models will be available for download and application to local data. ICE currently includes curated in vivo test data, reference chemical information, in vitro assay data (including Tox21<img src="" alt="™” class=”wp-smiley” style=”height: 1em; max-height: 1em;” />/ToxCast<img src="" alt="™” class=”wp-smiley” style=”height: 1em; max-height: 1em;” /> high-throughput screening data), and in silico model predictions. Users can query these data collections focusing on end points of interest such as acute systemic toxicity, endocrine disruption, skin sensitization, and many others. ICE is publicly accessible at

  • Received: 10 February 2017
    Revised: 10 February 2017
    Accepted: 12 February 2017
    Published: 24 May 2017

    Address correspondence to N. Kleinstreuer, NIEHS, 530 Davis Dr., Morrisville, NC 27560, Telephone: (919) 541-7997, E-mail:

    S.M.B., C.S., S.Q.M. and D.A. are employed by Integrated Laboratory Systems; J.P., A.S., A.T. and R.S. are employed by Sciome; both are consulting companies.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact Our staff will work with you to assess and meet your accessibility needs within 3 working days.


Traditional tests to evaluate chemicals for their potential impact on human and environmental health are performed using animal models, with varying degrees of success in accurately identifying hazards. Advances in science and technology offer the potential to develop more effective approaches based on combining higher-throughput testing methods, human cell–based systems, small model organisms, and computational models. Development and acceptance of new approaches that reduce and replace animal use for better predictivity of human health outcomes require that regulators, test method developers, and computational modelers have access to high-quality toxicity data. Ideally, these data and relevant analysis tools would be available in a user-friendly format that facilitates data integration and exploration, encourages hypothesis generation, and allows for computational processing and modeling. In reality, progress in developing and evaluating new approaches is often hindered by the lack of open-source central access points for curated data on which models can be built or tested. In addition, the data available are often of unknown quality; constructing models or validating new methodologies on such data produces models and test systems with high uncertainty and suboptimal performance. Ultimately, locating and accessing the types of data needed<img src="×72/2014.png" alt="—” class=”wp-smiley” style=”height: 1em; max-height: 1em;” />data in a structured format with metadata or annotation sufficient to gauge their quality and applicability<img src="×72/2014.png" alt="—” class=”wp-smiley” style=”height: 1em; max-height: 1em;” />becomes a huge time sink for researchers and a barrier to innovation.

Existing sources of toxicity test data have advantages and limitations. The European Chemical Agency’s registered substances list (ECHA 2017) contains detailed information and test results on thousands of substances obtained from chemical dossiers submitted by manufacturers. Although this resource has depth of information, the data contained in it do not fully support computational analysis owing to the database’s restrictions on data use, to its limited batch query capability, and to formatting variations of the data fields. The U.S. Environmental Protection Agency (EPA) has several web-based resources providing information on chemical properties and toxicity data available in a computational-friendly format without restrictions on use. These include ToxCast™ data available through the tcpl package (Filer et al. 2016), Aggregated Computational Toxicology Resource (ACToR) (Judson et al. 2008, 2012), and the Chemistry Dashboard (U.S. EPA 2017). These U.S. EPA resources are extremely valuable because they provide detailed information on chemical properties and activities from various sources. However, building and evaluating new approaches require different types of data to be combined in a structured format that is informed by relevant biology or toxicity end points. This task is best facilitated by having the data organized by test guidelines or regulatory end points, a feature that is not currently provided by any of these U.S. EPA resources. Similar challenges with respect to organization exist with other resources such as PubChem (Kim et al. 2016; Wang et al. 2014) and the Toxicological Data Network from the U.S. National Library of Medicine (NLM 2016), as well as the Chemical Effects in Biological Systems (CEBS) database from the National Institute of Environmental Health Sciences (NIEHS) (Lea et al. 2017), all of which provide rich content but may not meet data structure needs.

The National Toxicology Program’s Integrated Chemical Environment (ICE) is a web-based resource ( that provides access to curated data and tools that can be used in the development, validation, and implementation of in vitro and in silico approaches that reduce and replace the use of animals in toxicity testing. The ICE data integrator function allows users to query high-quality in vivo and in vitro test results by chemical and by end point. ICE also includes reference chemical lists and supporting data sets, as well as computational predictions for properties such as physicochemical parameters and clearance rates, which are often needed in model development.

ICE Overview

The current release of ICE features in vivo and in vitro data sets covering regulatory test end points such as acute oral and dermal toxicity, skin irritation and sensitization, eye corrosion, and endocrine disruption as well as in silico predictions of test end points and chemical properties (Table 1). These data sets include data assembled and curated by the National Toxicology Program Interagency Center for Evaluation of Alternative Toxicological Methods (NICEATM) during past evaluations of new test methods and reference chemical lists generated from these efforts. ICE also includes a curated version of the high-throughput screening (HTS) data from Tox21TM and ToxCast™ (Huang 2016; Kavlock et al. 2012; Richard et al. 2016; Tice et al. 2013), with data filtered on the basis of chemical quality control information from the National Center for Advancing Translational Sciences and flags generated by the U.S. EPA’s tcpl pipeline (Filer et al. 2016; U.S. EPA 2016;). Additional data sets will be added with future releases.

Table 1. Data types included in 2017 ICE releases and example end points.

Data types Availability Type End point examples
Acute dermal toxicity October 2017 (tentative) In vivo Rodent LD50
Acute inhalation toxicity October 2017 (tentative) In vivo Rodent LC50
Acute oral toxicity March 2017 In vivo Rodent LD50
Acute oral toxicity March 2017 In vitroa Basal cytotoxicity IC50
Androgenic activity March 2017 In vitro Androgen receptor binding and transactivation (agonist and antagonist activity)
Androgenic activity July 2017 (tentative) In vivo Lowest effect level in the rodent Hershberger assay
Androgenic activity March 2017 In silico Androgen receptor pathway model scores
Curated HTS March 2017 In vitro Assay ACC, AC50
Dermal irritation March 2017 In vivo Skin irritation/corrosion classification categories
Dermal sensitization March 2017 In vivo Mouse LLNA EC3 and human patch test lowest effect level
Dermal sensitization March 2017 In vitro KeratinoSens™, DPRA, hCLAT assay results
Dermal sensitization July 2017 (tentative) In silico Binary sensitizer/nonsensitizer call
Estrogenic activity March 2017 In vivo Lowest effect level in the rodent uterotrophic assay
Estrogenic activity March 2017 In silico Estrogen receptor pathway model scores
Ocular irritation March 2017 In vivo Eye irritation/corrosion classification categories
Physicochemical property predictions March 2017 In silico LogP, logVP, logBCF, logS, melting point, boiling point

Notes: AC50, concentration that increases activity by 50%; ACC, activity concentration at cutoff, a measure of the activity threshold for an assay response based on curve-fitting models; EC3, in the LLNA, a test chemical concentration that produces a stimulation index of 3; hCLAT, human cell line activation test; IC50, concentration that inhibits activity (in this context, decreases cell viability) by 50%; LC50, inhalation concentration expected to produce lethality in 50% of animals tested; LD50, dose expected to produce lethality in 50% of animals tested; LLNA, local lymph node assay; physicochemical properties characterized as log values are log 10; logBCF, log of the bioconcentration factor; logP, octanol-water partition coefficient; logVP, the vapor pressure; logS, log of the solubility in water.

aIn vitro data were used to develop a nonanimal method for setting starting doses for in vivo acute oral toxicity studies.

The query function of ICE, the integrator, facilitates combining the different data sets within ICE, allowing users to merge data points from the different data sets to simultaneously examine end points based on all available substance data or by using an input list of chemicals. Upon submission of the query, the user is provided with an overview of the query results along with several exportable data views, supporting both computational manipulation and manual interaction.

To assist the user with developing and interpreting queries, ICE provides a “Help” page that walks the user through building a search and understanding search output. Other user support resources include a “Home” page where updates and other news will be announced, an “About” page with background information, and a “Data Sets” page with information about the sources of the data in ICE. This page includes links that lead to more detailed information about each data set, including references and metadata for context about the data end points. Some information about the studies available in the source data sets is not available in ICE (for example, detailed clinical observations or some study protocol details), but it is available via the provided links and will also be accessible via CEBS (Lea et al. 2017) to facilitate integration with other data resources from the NIEHS.

User Stories

We developed ICE to address needs that have been expressed frequently by NICEATM stakeholders. As such, we tailored the design of ICE around user stories that developed from those needs. Here, we describe stories from three different user groups (method developer, chemical producer, risk assessor) and how ICE can help meet these users’ needs (Figure 1).

Schematic diagram depicting ICE users on the left panel; resources, namely high-quality data, reference chemicals, and computational tools in the center panel; and outcomes, namely identify opportunities to develop new methods, compare method performance, identify data gaps, obtain and examine toxicity and chemical data, and develop testing strategies in the right panel.

Figure 1. Integrated Chemical Environment (ICE) users, resources, and outcomes. ICE was developed to support three main user roles: method developers, chemical producers, and risk assessors. The center panel lists the resources ICE provides to help these user groups complete some of the major tasks listed in the right panel.

Method Developer

Method developers are developing new in vitro or in silico methods to prioritize substances for further toxicity testing, or to test a substance for potential health impacts. The availability of data anchored to regulatory end points of interest from ICE facilitates exploration of the current state of methods, and such curated end point data can be compared with user-generated test data. These capabilities help method developers improve the performance of existing methods or target new method development to areas where the currently available test methods leave room for improvement. Furthermore, by having in vivo data organized with the in vitro and in silico data, it is easier to identify data gaps that may be targeted using in silico approaches.

In vitro test method developers often need reference chemicals, which are chemicals that cause specific, well-characterized biological effects and can therefore be used to assess the performance of a test method designed to measure that effect. These reference chemicals, along with the supporting data, are helpful in both in vitro method development and validation efforts. The reference chemical lists included in ICE provide a starting place to identify test materials and associated data.

For those developing in silico approaches, good quality data are important for training and testing predictive models. Method developers will find that the data in ICE are well annotated, cleaned, and formatted to support in silico modeling, allowing them to focus on method optimization versus data preparation. ICE currently provides many of the parameters needed as input for these models, and open-source workflows for in vitro to in vivo extrapolation and chemical property prediction will be added to ICE later in 2017.

Chemical Producer

For developers of new products for which toxicity testing will be required, typical testing needs include prioritizing substances to move forward through the development process and generating data needed to meet regulatory requirements. The ICE integrator allows the chemical producer to input a list of chemicals and obtain available testing data, which may help with evaluating the potential for adverse impacts during lead agent prioritization.

For chemical producers seeking to replace animals required for testing, the ICE integrator allows end point–oriented searches to compare data sets across all available substances. This capability allows the chemical producer to compare the results coming from different test methods relating to a specific end point. A query that returns in vitro and in silico test results paired with in vivo animal data may allow the chemical producer to better identify which nonanimal methods will best meet their information needs. This information may provide the chemical producer with the confidence in those nonanimal methods that is needed to change in-house testing approaches for ones that use fewer animals, save money, and better predict the end point of interest.

Risk Assessor

Similar to the chemical producer, the risk assessor may want to obtain available animal or human test data on a list of chemicals or review available data from nonanimal methods. Resources currently in ICE can help risk assessors with prioritizing chemicals for further toxicity testing and with identifying the most informative tests for that purpose in addition to providing data on the likely health impacts of a substance. The ICE data explorer view provides user-friendly data interaction and an easy-to-use snapshot of the types of data and end points. This information can be useful in prioritization or in identification of data gaps. The risk assessor may also be interested in comparing the performance of nonanimal test data with existing guideline animal test results across a wide range of chemicals, which is easily done using the ICE integrator.

ICE supports the exporting of data, which facilitates more detailed comparisons of end point variability. Exported data are preformatted and ready for analysis, enabling the risk assessor to move easily from query output into the subsequent analysis workflows. Transparency and reproducibility of data and process are important to the risk assessor; therefore, ICE provides reference details on the Data Sets page and clear referencing from the query output.

Next Steps

Computational tools will be included in ICE by the fall of 2017. These programs and workflows developed by NICEATM staff and external partners will enable researchers to conduct analyses locally using either their own data or data downloaded from ICE. Currently in the development queue are machine learning models for six physicochemical properties (Zang et al. 2017), predictive signatures based on HTS data such as embryonic vascular disruption leading to adverse prenatal outcomes (Knudsen and Kleinstreuer 2011), and in vivo<img src="×72/2013.png" alt="–” class=”wp-smiley” style=”height: 1em; max-height: 1em;” />to<img src="×72/2013.png" alt="–” class=”wp-smiley” style=”height: 1em; max-height: 1em;” />in vitro extrapolation (IVIVE) workflows (Chang et al. 2015; Wetmore et al. 2012). In addition to downloadable predictions and workflows, interactive tools such as those that facilitate biological pathway–informed IVIVE are needed to help link in vitro activity concentrations to relevant in vivo exposures and outcomes.

Future versions of ICE will also include tutorials aimed at helping users better understand the tools available in ICE and their applications in regulatory safety testing. These tutorials will address questions that frequently arise about appropriate use of computational tools and models on topics such as characterizing the domain of applicability and model uncertainty. Development of these resources is currently underway with a tentative 2018 launch date.


This project was funded in part with federal funds from the National Institute of Environmental Health Sciences, National Institutes of Health under contract no. HHSN273201500010C to Integrated Laboratory Systems (ILS) and its subcontractor, Sciome, in support of the National Toxicology Program Interagency Center for the Evaluation of Alternative Toxicological Methods (NICEATM). The views expressed in this manuscript do not necessarily represent the official positions of any federal agency.


Chang X, Kleinstreuer N, Ceger P, Hsieh J-H, Allen D, Casey W, 2015. Application of reverse dosimetry to compare in vitro and in vivo estrogen receptor activity. Appl Vitro Toxicol 1:33–44, 10.1089/aivt.2014.0005.

ECHA (European Chemicals Agency). Registered Substances database. Available: [accessed 23 January 2017].

Filer DL, Kothiya P, Setzer RW, Judson RS, Martin MT. 2016. tcpl: the ToxCast pipeline for high-throughput screening data. Bioinformatics, 10.1093/bioinformatics/btw680.

Huang R. 2016. A quantitative high-throughput screening data analysis pipeline for activity profiling. Methods Mol Bio l1473:111–122, PMID: 27518629, 10.1007/978-1-4939-6346-1_12.

Judson R, Richard A, Dix D, Houck K, F, Martin M, et al. 2008. ACToR<img src="×72/2013.png" alt="–” class=”wp-smiley” style=”height: 1em; max-height: 1em;” />Aggregated Computational Toxicology Resource. Toxicol Appl Pharmacol 233:7<img src="×72/2013.png" alt="–” class=”wp-smiley” style=”height: 1em; max-height: 1em;” />13, PMID: 18671997, 10.1016/j.taap.2007.12.037.

Judson RS, Martin MT, Egeghy P, Gangwal S, Reif DM, Kothiya P, et al. 2012. Aggregating data for computational toxicology applications: The U.S. Environmental Protection Agency (EPA) Aggregated Computational Toxicology Resource (ACToR) system. Int J Mol Sci 13:1805–1831, PMID: 22408426, 10.3390/ijms13021805.

Kavlock R, Chandler K, Houck K, Hunter S, Judson R, Kleinstreuer N, et al. 2012. Update on EPA’s ToxCast program: Providing high throughput decision support tools for chemical risk management. Chem Res Toxicol 25:1287–1302, PMID: 22519603, 10.1021/tx3000939.

Kim S, Thiessen PA, Bolton EE, Chen J, Fu G, Gindulyte A, et al. 2016. PubChem substance and compound databases. Nucleic Acids Res 44:D1202–D1213, PMID: 26400175, 10.1093/nar/gkv951.

Knudsen TB, Kleinstreuer NC. 2011. Disruption of embryonic vascular development in predictive toxicology. Birth Defects Res C Embryo Today 93:312–323, PMID: 22271680, 10.1002/bdrc.20223.

Lea IA, Gong H, Paleja A, Rashid A, Fostel J. 2017. CEBS: A comprehensive annotated database of toxicological data. Nucleic Acids Res 45:D964–D971, PMID: 27899660, 10.1093/nar/gkw1077.

NLM (National Library of Medicine). 2016. Fact Sheet. TOXNET®: Toxicology Data Network. Available: [accessed 23 January 2017].

Richard AM, Judson RS, Houck KA, Grulke CM, Volarath P, Thillainadarajah I, et al. 2016. ToxCast chemical landscape: Paving the road to 21st century toxicology. Chem Res Toxicol 29:1225–1251, PMID: 27367298, 10.1021/acs.chemrestox.6b00135.

Tice RR, Austin CP, Kavlock RJ, Bucher JR. 2013. Improving the human hazard characterization of chemicals: a Tox21 update. Environ Health Perspect 121:756–765, PMID: 23603828, 10.1289/ehp.1205784.

U.S. EPA (U.S. Environmental Protection Agency). 2016. ToxCast Chemicals: Data Management and Quality Considerations Overview. Available: [accessed 18 January 2017].

U.S. EPA (U.S. Environmental Protection Agency). Chemistry Dashboard. Available: [accessed 23 January 2017].

Wang Y, Suzek T, Zhang J, Wang J, He S, Cheng T, et al. 2014. PubChem BioAssay: 2014 update. Nucleic Acids Res 42:D1075–D1082, PMID: 24198245, 10.1093/nar/gkt978.

Wetmore BA, Wambaugh JF, Ferguson SS, Sochaski MA, Rotroff DM, Freeman K, et al. 2012. Integration of dosimetry, exposure, and high-throughput screening data in chemical toxicity assessment. Toxicol Sci 125:157–174, PMID: 21948869, 10.1093/toxsci/kfr254.

Zang Q, Mansouri K, Williams AJ, Judson RS, Allen DG, Casey WM, et al. 2017. In silico prediction of physicochemical properties of environmental chemicals using molecular fingerprints and machine learning. J Chem Inf Model 57:36–49, PMID: 28006899, 10.1021/acs.jcim.6b00625.

Retraction: “A Case-Cohort Study of Cadmium Body Burden and Gestational Diabetes Mellitus in American Women”

About This Article open

Citation: Romano ME, Enquobahrie DA, Simpson CD, Checkoway H, Williams MA. 2017. Retraction: “A Case-Cohort Study of Cadmium Body Burden and Gestational Diabetes Mellitus in American Women.” Environ Health Perspect 125:A64;

Final Publication: 31 March 2017

PDF icon PDF Version (56 KB)

Related EHP Article

Retracted: A Case-Cohort Study of Cadmium Body Burden and Gestational Diabetes Mellitus in American Women

Megan E. Romano, Daniel A. Enquobahrie, Christopher D. Simpson, Harvey Checkoway, and Michelle A. Williams

Environ Health Perspect 123(10):993–998 (2015),

This article is being retracted at the request of the authors because of inadvertent errors in the statistical code that resulted in the exclusion of 12 gestational diabetes mellitus cases with urinary cadmium below the limit of detection. The coding errors do not impact any other published studies of this population.

In Table 1, the corrected geometric mean for the gestational diabetes mellitus cases was 0.33 μg/g Cr [95% confidence interval (CI): 0.30, 0.37].

In Table 2, the magnitude of the effect estimates reported was attenuated and the p-trend was no longer statistically significant, such that the odds ratios for gestational diabetes mellitus with increasing urinary cadmium tertile were 1.12 (95% CI: 0.64, 1.98) for middle versus low tertile and 1.34 (95% CI: 0.78, 2.29) for high versus low tertile; p-trend = 0.28.

The magnitude of the effect estimates reported in Table 3 and the effect estimates for additional sensitivity analyses reported in the original manuscript were also generally attenuated.

The authors regret any inconvenience to the scientific community.

Erratum: “Telomere Length, Long-Term Black Carbon Exposure, and Cognitive Function in a Cohort of Older Men: The VA Normative Aging Study”

About This Article open

Citation: Colicino E, Wilson A, Frisardi MC, Prada D, Power MC, Hoxha M, Dioni L, Spiro A, Vokonas PS, Weisskopf MG, Schwartz JD, Baccarelli AA. 2017. Erratum: “Telomere Length, Long-Term Black Carbon Exposure, and Cognitive Function in a Cohort of Older Men: The VA Normative Aging Study.” Environ Health Perspect 125:A63;

Final Publication: 31 March 2017

PDF icon PDF Version (58 KB)

Related EHP Article

Telomere Length, Long-Term Black Carbon Exposure, and Cognitive Function in a Cohort of Older Men: The VA Normative Aging Study

Elena Colicino, Ander Wilson, Maria Chiara Frisardi, Diddier Prada, Melinda C. Power, Mirjam Hoxha, Laura Dioni, Avron Spiro III, Pantel S. Vokonas, Marc G. Weisskopf, Joel D. Schwartz, and Andrea A. Baccarelli

Environ Health Perspect 125(1):76–81 (2017),

In this erratum, the authors correct the information that appeared in their original article about the funding sources, including the agencies that provided the funding and the grant numbers:

This work was supported by the National Institute of Environmental Health Sciences (R01ES021733, R01ES015172, T32ES007142), the U.S. Environmental Protection Agency (RD83479801, RD832416, RD83587201), and the Agricultural Research Service of the U.S. Department of Agriculture (contract 53-K06-510). D. Prada received support from CONACYT and Fundación México en Harvard; M. Power received funding from the National Institute of Aging (T32 AG027668); and A. Spiro received support from the Clinical Science Research and Development Service of the U.S. Department of Veterans Affairs. The U.S. Department of Veterans Affairs (VA) Normative Aging Study (NAS) is supported by the Cooperative Studies Program and the Epidemiologic Research and Information Center (ERIC), which is a research component of the Massachusetts Veterans Epidemiology Research and Information Center (MAVERIC).

The views expressed in this paper are those of the authors and do not necessarily represent the views of the U.S. Department of VA.

The authors apologize for the missing information.

Erratum: “Microcystis Rising: Why Phosphorus Reduction Isn’t Enough to Stop CyanoHABs”

About This Article open

Citation: Levy S. 2017. Erratum: “Microcystis Rising: Why Phosphorus Reduction Isn’t Enough to Stop CyanoHABs.” Environ Health Perspect 125:A62;

Final Publication: 31 March 2017

PDF icon PDF Version (50 KB)

Environ Health Perspect 125(2):A34–A39 (2017),

One of the researchers quoted in this article was not fully identified. Timothy Davis is a research scientist at the Great Lakes Environmental Research Laboratory of the National Oceanic and Atmospheric Administration.

EHP regrets the error.

The Forest and the Trees: How Population-Level Health Protections Sometimes Fail the Individual

Photo of a crowd of people on a plaza in Shanghai, China

Photo of a crowd of people on a plaza in Shanghai, China

Burdening individuals with the responsibility of reducing their own environmental exposures is not only unreliable in terms of health protection but also a contributor to environmental injustice.
© Raga/Getty Images

Side-by-side photos showing a Beijing skyscraper on smoggy and clear days

Side-by-side photos showing a Beijing skyscraper on smoggy and clear days

These photos show the air quality in Beijing on 1 January 2017 (top) in the midst of an “airpocalypse” smog emergency and on a clear day the week before. The notoriously bad air in many Chinese cities is slowly improving overall, but long-term exposures are still many times higher than World Health Organization recommendations.
© Greg Baker/AFP/Getty Images 

Photo of a Chinese man and his son flying a kite while wearing face masks

Photo of a Chinese man and his son flying a kite while wearing face masks

This father and son in Shanghai wore masks on a day in 2015 when the city’s Air Quality Index was 2.3 times the limit considered healthy. Chinese residents often turn to breathing masks to protect themselves from dangerous air. Tight-fitting particulate respirators can effectively block airborne pollutants, but the more commonly used surgical masks are mostly useless.
© VCG/VCG via Getty Images

Photo of Flint residents loading cases of bottled water into a car

Photo of Flint residents loading cases of bottled water into a car

Volunteers loaded cases of free water into waiting vehicles in Flint, Michigan, as part of an effort to provide potable water to residents affected by lead contamination. Programs like Flint’s that use public funds to subsidize stopgap measures during an emergency appear to be on solid ground ethically—provided all residents have equal access.
© Geoff Robins/AFP/Getty Images

Photo of Flint residents receiving instructions on using their new faucet-mounted water filter

Photo of Flint residents receiving instructions on using their new faucet-mounted water filter

Residents in Flint learned about their new faucet-mounted filter from the handyman who installed it. The Flint crisis has illustrated the financial and logistical challenges of distributed treatment on a large scale. However, some experts believe point-of-use treatment strategies can be an affordable and efficient option for small water systems.
© Sarah Rice/Getty Images

Background image: © Alija/iStockphoto

Nate Seltenrich covers science and the environment from Petaluma, CA. His work has appeared in High Country News, Sierra, Yale Environment 360, Earth Island Journal, and other regional and national publications.

About This Article open

Citation: Seltenrich N. 2017. The forest and the trees: how population-level protections sometimes fail the individual. Environ Health Perspect 125:A65–A70;

Published: 31 March 2017

PDF icon PDF Version (5.5 MB)

Beijing residents wearing face masks to protect themselves against extreme air pollution.1 Residents of small towns such as Mooringsport, Louisiana,2 and Coal Mountain, Virginia,3 buying bottled water or filters to avoid unsafe water. For epidemiologist Barbara Hoffmann of Heinrich-Heine-University in Dusseldorf, Germany, such examples of individual efforts to evade environmental pollutants raise serious ethical questions. “Only some people can afford to take preventive measures,” she says. “Others don’t have the means to do it.”

As a result, Hoffmann says, burdening individuals with the responsibility to reduce their own environmental exposures can create or worsen environmental injustice, in which harmful exposures are inequitably distributed across a population.

Moreover, Hoffmann fears that the tacit acceptance by a population of individualized efforts may limit their government’s incentive to protect all citizens through centralized actions. “This is not the way we want to go—to put the responsibility for breathing clean air on the individual instead of on the state,” she says. Instead, she argues, the state or central government can be much more effective at providing clean air and water to rich and poor residents alike through the passage and enforcement of adequate laws.

The ethical line between centralized and distributed (i.e., individualized) solutions is a fine one, suggests Nino Künzli, deputy director of the Swiss Tropical and Public Health Institute and dean of the Swiss School of Public Health. “We have to acknowledge that there are individual rights and governmental duties,” he says. “The success of the latter may take years to fully materialize. In the meantime, individuals have the right to know how to protect themselves.”

This right, however, does not guarantee awareness of the risk in the first place, the financial means to respond to it, or accurate knowledge of what protective measures to take.4 Even for those who can afford them, individual efforts like personal breathing masks and in-home water filters are no guarantee of safety, as they can still be ineffective if chosen or used improperly. Partially as a result, studies suggest, exposures to harmful chemicals through air and water around the world are distributed unevenly—and often inequitably.5

Air Pollution Exposures

The notoriously bad air in many Chinese cities is slowly improving overall, but long-term exposures are still many times higher than World Health Organization recommendations.6,7 And seasonal spikes can be far worse. At least once every winter, it seems, a Chinese “airpocalypse” grabs international headlines as coal burning escalates and weather patterns trap polluted air, subjecting citizens to visibly heavy, hazardous smog.1

Media reports often include photos of individuals who have turned to breathing masks—flimsy, mostly useless surgical masks as well as tighter-fitting particulate respirators—to protect themselves from the dangerous air. For many Western viewers, these images may be a symbol of the country’s failure to keep air pollution in check during its recent industrialization. For others, the masks may be an accepted part of modern life in rapidly developing nations like China and India: a small price to pay for lifting millions out of poverty. After all, countries in the developed world also experienced extreme pollution as they industrialized during the eighteenth and nineteenth centuries.8

The better masks can have benefits, even for healthy individuals.9 In a recent study evaluating the short-term cardiovascular health effects of wearing particulate respirators, researchers from Fudan University in Shanghai and Texas A&M University’s School of Public Health measured heart rate variability and blood pressure in 24 healthy young adults with and without masks. Half the participants wore their respirators whenever possible for 48 hours, while the other half went mask-free. Three weeks later, the randomly assigned groups swapped places. All participants wore devices that measured their blood pressure throughout each 48-hour period.

Across both groups the researchers found respirator use to be associated with healthy blood pressure and increased autonomic nervous function, both indicators of good cardiovascular health. This was true at a time when concentrations of fine particulate matter averaged 74.2 µg/m3, which is three times the WHO daily standard (yet only a fraction of what Beijing saw at the start of 201710).

Künzli welcomes the study’s findings even as he laments the need for respirators. “They are not the solution in China, but they give some reduction in exposure,” he says. “I don’t think we should fundamentally question that in the context of these highly polluted megacities where people have no other choice. … In that sense I have to say I appreciate these scientists critically evaluating wearing masks and what kind of mask we need to wear to get any benefit.”

Such information is sorely needed in China, believes Zhuohui Zhao, a coauthor of the paper and an associate professor in Fudan University’s School of Public Health. The mask tested in the study was 3M’s 8210V model, which is an N95 respirator, meaning it filters out at least 95% of solid and water-based particulates up to 0.3 µm in diameter.11 But the 8210V is just one of countless options available on the market. While breathing masks can vary widely in performance and fit, with cheap and ineffective surgical masks perhaps the most widely used, the differences among them are not always apparent to the consumer, Zhao says.

“For common people, it’s really hard to identify what kind of mask is useful or what kind of air purifier is efficient,” she says. “That’s true even for people working at universities. I have staff and friends in other colleges who say they are very confused about the market. There are many new brands of air purifiers, and there are lots of masks for them to choose from.”

Zhao says the 8210V retails in China for US$1.50–2.00 each. The nation’s average yearly salary is US$9,000,12 putting regular use of effective N95 respirators within reach of the middle and upper classes but potentially not the poor.

Many other people in China choose not to wear any kind of mask at all—a group that may be larger than some expect, Zhao believes. She says this is due in large part to a low perception of risk arising from limited knowledge of the potential adverse health effects of particulate matter. “People that know more about the pollution tend to take protective actions,” she explains. “But this group of people is still a small part of the population.”

Reliable statistics on mask usage are hard to come by, Zhao says, but observations from Beijing reported in January 2017 suggest regular mask wearers are indeed a minority.13 Over the course of five weeks in the preceding November and December—a time of year when local air quality is typically at its worst—daily usage rates along two streets in the capital ranged from about 10% to 50%. Nevertheless, other reports suggest risk perception is on the rise nationwide.14

While some companies in China’s polluted cities distribute respirators as an employee benefit, Zhao notes, no such subsidy exists for the unemployed or working poor. Nor would Hoffman, Künzli, or their colleagues in epidemiology and public health accept such a policy, they say. “If that was part of the official strategies of governments, of course we would need to step up very loudly and make clear that this was a wrong way of going forward,” says Künzli. “I want to see governments invest in cleaning up the air.”

Water Pollution Exposures

Drinking water comes with its own set of expectations around centralized versus distributed approaches to risk mitigation. In the United States, the Safe Drinking Water Act requires that water delivered to ratepayers by public water systems meet high health-based standards for both chemical and microbial contaminants.15 Yet this doesn’t always guarantee the elimination of all harmful agents.

One growing concern among citizens and many local governments in recent months has been the safe delivery of drinking water through lead-bearing plumbing materials. Such materials were banned under the Safe Drinking Water Act Amendments of 1986 but remain in use in many older homes and buildings.16

Lead is just one of many potential threats facing the 150,000 public water systems in the United States. The federal Safe Drinking Water Act regulates 80 different chemicals including disinfection by-products, organic and inorganic chemicals, radionuclides, and six microorganism groups, including Cryptosporidium and coliforms.17 States may impose their own additional standards.

Most large water systems in the nation have few problems meeting these standards. That’s less true for small systems (often defined as those having fewer than a few thousand connections), which serve about 38 million people, or 12% of the U.S. population.18 “Noncompliance occurs more frequently at smaller public water systems because they often have fewer resources to operate and maintain compliance,” states a 2015 U.S. Environmental Protection Agency (EPA) report on the nation’s public water systems.19

Arsenic, for example, is a carcinogen that can be prohibitively expensive to remove centrally in small systems lacking economies of scale, says John Pujol, CEO of water-testing firm SimpleWater, Inc., which holds a license for a proprietary arsenic remediation technology it hopes to commercialize in the United States. As a result, arsenic often goes untreated in such systems, even at levels known to be unsafe and that significantly exceed the federal limit of 10 ppb.19

That’s what has happened in nearly 100 small systems serving 55,000 people in California alone, according to a recent report by the Environmental Integrity Project, an environmental watchdog group.20 Public water systems that fall out of compliance with EPA standards may receive warning letters or notices of violation, or in more severe cases be subject to citations, administrative orders, criminal charges, or other sanctions.19

Many Americans simply don’t trust centralized treatment systems of any size to deliver safe water to their homes, suggests Joseph Cotruvo, a public health consultant and former director of the EPA Drinking Water Standards Division. A 2015 national telephone survey by the Water Quality Association, which represents the water treatment industry, found that 59% of respondents were highly concerned about contaminants in their drinking water, 43% used a water filter, and 70% maintained that the municipality, not the individual, still bore ultimate responsibility for drinking water safety.21

“If you look at public perception and concerns, a very high percentage of people have negative perceptions of their drinking water,” says Cotruvo. “So they’re voting with their pocket books, buying bottled water or filters.” Yet Cotruvo, whose tenure at the EPA began in the agency’s earliest days, also believes that in some cases this perception is misdirected: “Municipal water in the U.S. is actually safer than ever, especially since the implementation of the Safe Drinking Water Act,” he says.

That said, the recent crisis in Flint, Michigan, did not help public perception of centralized water treatment. It also revealed the challenges of distributed treatment on a large scale. In late 2015, a full year and a half after switching its water source from the Detroit Water and Sewerage Department to the corrosive Flint River and setting off perhaps the highest-profile drinking water crisis this country has ever seen, the city began offering free filters to residents concerned about the safety of their water. This improved people’s access to safe water but only among those who were already aware of and able to act on the problem.22,23,24

Two months later, Flint mayor Karen Weaver expanded the program’s reach by declaring an official emergency and advising residents to drink only bottled water or filtered tap water.25 The state soon followed suit and began ramping up the distribution of bottled water, filters, replacement cartridges, and at-home test kits through official centers and limited home delivery.26 However, it was not yet absolved of the responsibility to deliver safe drinking water to all residents.

In November 2016 a federal court asserted as much by ordering the home delivery of bottled water to any Flint resident lacking a verified water filter. City and state officials twice fought the order—saying it would cost Michigan at least $10.45 million a month—but lost.27,28,29 On 24 January 2017, state officials reported that lead levels in Flint’s water were finally back below federal limits,30 but to date the delivery order stands—as does the city’s own recommendation to use a filter.31

Aside from lead, other agents and factors raise new questions about what constitutes drinking water safety. These include emerging chemicals of concern that cannot always be removed, such as pharmaceuticals32 and nanoparticles33; aging infrastructure under streets and inside homes; and other persistent threats, such as nitrates34 and Legionella bacteria.35

At the heart of these issues is a philosophical and practical question about the role of centralized versus distributed solutions within public water systems. The centralized water treatment plants and distribution systems built in American cities and suburbs over the last 100 years were designed under the premise that in-home treatment shouldn’t be necessary, says David Sedlak, a professor of environmental engineering at the University of California, Berkeley. That view persists, he says, even as weaknesses of the centralized model have emerged in recent years.

Mitigation Challenges

If the most ethical solutions to air pollution are always centralized, with drinking water it’s not so black and white. Programs like Flint’s that use existing public funds to subsidize filters and bottled water during an emergency appear to be on solid ground ethically—provided, of course, all residents have equal access. A public water system that is out of compliance with federal water standards and delivering potentially unsafe water, by contrast, places the burden squarely on individuals to be aware of and then attempt to mitigate the problem themselves.

Still, most experts don’t recommend that large or even medium-size water systems seek to systematize point-of-use treatment—that is, gain EPA approval—because it can become a logistical (and financial) nightmare. In their appeal of the federal order to deliver bottled water to any of Flint’s 100,000 residents without an approved filter, state officials said such a plan would require a “Herculean effort” and increase the scope of Michigan’s emergency response “to an unnecessary and insurmountable degree.”28

Instead, experts including Cotruvo believe reverse-osmosis units and other under-sink or faucet-mounted filters represent an attractive option for small, often cash-strapped systems across the country hoping to provide safe water and stay in compliance with federal and state regulations.36 “There is an economy of scale level where central treatment is more cost-effective, although still expensive in a small community compared to a large community,” he says. Point-of-use strategies are “a more efficient way of providing safe water because you’re really only treating the water that people consume.”

Yet a bevy of regulatory and economic hurdles can stand in the way of this strategy. According to federal policy, the water system operator or utility bears all responsibility for purchasing, installing, and maintaining the devices, including inspecting every unit annually.37 This entails routine visits inside private homes at considerable expense and effort that climbs rapidly with the number of connections.

“The problem is that it requires a pretty sophisticated maintenance and regulatory apparatus that doesn’t currently exist,” says Pujol. There is also a problem with perception, notes Sedlak. He explains, “There is this belief among people who are involved in centralized water systems that moving toward point-of-use is regressing and is basically admitting that we can’t deliver safe water in a centralized system.”

If point-of-use treatment were to be more readily accepted at the federal and state levels, Cotruvo believes, many people nationwide could quickly gain access to safer water. Instead, only a tiny number of water systems nationwide—fewer than 100 out of 150,000, Cotruvo estimates—have sought and gained approval to use distributed solutions to maintain safe water system-wide. Many others simply fall out of compliance with federal law.

The EPA reports that in 2013, the most recent year for which it has tabulated data, 27% of all public water systems—serving roughly a quarter of the U.S. population—had at least one “significant” violation of the Safe Drinking Water Act, a category that includes both technical and health-based violations.18 More than two-thirds of these were related to monitoring and reporting, which the EPA considers a serious violation because it makes it impossible to know whether drinking water standards are being met.

Since out-of-compliance systems are more likely to be small, and small systems are more likely to serve rural and low-income communities, the current system puts citizens who are already underserved at a potential disadvantage. Additionally, contaminants such as naturally occurring arsenic and agricultural chemicals are more likely to be present in lower-income regions like the rural Midwest and California’s Central Valley.38 “Many members of the low-income and environmental justice [advocacy] community take point-of-use seriously, because they see it as pretty much the only way forward,” Pujol says.

Peter Gleick, president emeritus and chief scientist of the nonprofit Pacific Institute, says he believes recent developments have led U.S. public water systems to a fork in the road. Down one path is the potential to rebuild water systems to a higher standard, including through less-centralized approaches.

Down the other is a future he describes as a “downward spiral in the quality and cost of our water systems,” where the rich install point-of-use systems and the poor are left relying on bottled water or drinking whatever comes out of the tap. Already, Gleick says, poor infrastructure, rural water contamination, and poverty threaten basic water services for millions of Americans. “Inequitable access to safe and affordable drinking water is a large and growing problem,” he says, “and should be an embarrassment for a country like the United States.”


1. Flanagan E. Millions in China learn to live with smog ‘airpocalypse.’ NBC News, World section, online edition (7 January 2017). Available:​ns-china-learn-live-smog-airpocalypse-n7​03911 [accessed 20 March 2017].

2. Talamo L. East Mooringsport customer: “I don’t trust this water.” The Shreveport Times, News section, online edition, 13 December 2016. Available:​s/2016/12/13/east-mooringsport-customer-​dont-trust-water/85260338/ [accessed 20 March 2017].

3. Unger L, Nichols M. 4 million Americans could be drinking toxic water and would never know. USA Today, online edition (13 December 2016). Available:​12/13/broken-system-means-millions-of-ru​ral-americans-exposed-to-poisoned-or-unt​ested-water/94071732/ [accessed 20 March 2017].

4. Tyrrell J, et al. Associations between socioeconomic status and environmental toxicant concentrations in adults in the USA: NHANES 2001–2010. Environ Int 59:328–335 (2013), doi: 10.1016/j.envint.2013.06.017.

5. Hajat A, et al. Socioeconomic disparities and air pollution exposure: a global review. Curr Environ Health Rep 2(4):440–450 (2015), doi: 10.1007/s40572-015-0069-5.

6. Liu C, et al. Associations between ambient fine particulate air pollution and hypertension: a nationwide cross-sectional study in China. Sci Total Environ 584–585:869–874 (2017), doi: 10.1016/j.scitotenv.2017.01.133.

7. Ma Z, et al. Satellite-based spatiotemporal trends in PM2.5 concentrations: China, 2004–2013. Environ Health Perspect 124(2):184–192 (2016), doi: 10.1289/ehp.1409481.

8. Szreter S. The population health approach in historical perspective. Am J Public Health 93(3):421–431 (2003), PMID: [Pubmed].

9. Shi J, et al. Cardiovascular benefits of wearing particulate-filtering respirators: a randomized crossover trial. Environ Health Perspect 125(2):175–180 (2017), doi: 10.1289/EHP73.

10. Gallucci M. Beijing’s New Year smog wave causes flight cancellations, road closures. Mashable, World section (1 January 2017). Available:​g-new-year/ [accessed 20 March 2017].

11. CDC. NIOSH Guide to the Selection and Use of Particulate Respirators [website]. Atlanta, GA:National Institute for Occupational Safety and Health, U.S. Centers for Disease Control and Prevention (updated 6 June 2014). Available: [accessed 20 March 2017].

12. Trading Economics. China Average Yearly Wages [website]. Lisbon, NY:Trading Economics (2017). Available:​ges [accessed 20 March 2017].

13. Mingrui M, et al. How many people wear masks in Beijing? A street mask observation. Surging News Network, City Hall section (3 January 2017). Available:​d_1592513 [accessed 20 March 2017].

14. VanderKlippe N. Tensions rising as Chinese no longer willing to hold their breath on pollution problems. The Globe and Mail, World section, online edition (19 March 2017). Available:​d/how-china-is-losing-its-war-onpollutio​n/article34342466/ [accessed 20 March 2017].

15. EPA. Safe Drinking Water Act [website]. Washington, DC:U.S. Environmental Protection Agency (updated 12 January 2017). Available: [accessed 20 March 2017].

16. EPA. The Lead Ban: Preventing the Use of Lead in Public Water Systems and Plumbing Used for Drinking Water. EPA 570/9-89-BBB. Washington, DC:Office of Water, U.S. Environmental Protection Agency (August 1989).

17. EPA. Table of Regulated Drinking Water Contaminants [website]. Washington, DC:U.S. Environmental Protection Agency (updated 4 October 2016). Available:​nking-water/table-regulated-drinking-wat​er-contaminants [accessed 20 March 2017].

18. EPA. Providing Safe Drinking Water in America: 2013 National Public Water Systems Report. EPA 305R15001. Washington, DC:U.S. Environmental Protection Agency (2015). Available:​es/2015-06/documents/sdwacom2013.pdf [accessed 20 March 2017].

19. EPA. Drinking Water Requirements for States and Public Water Systems Chemical Contaminant Rules [website]. Washington, DC:U.S. Environmental Protection Agency (updated 29 April 2016). Available:​ontaminant-rules [accessed 20 March 2017].

20. Pelton T, et al. Arsenic in California Drinking Water: Three Years After EPA Notice of Noncompliance to State, Arsenic Levels Still Unsafe in Drinking Water for 55,000 Californians. Washington, DC:Environmental Integrity Project (September 2016). Available:​tent/uploads/CA-Arsenic-Report.pdf [accessed 20 March 2017].

21. WQA. Consumer Opinion Studies [website]. Lisle, IL:Water Quality Association (2017). Available:​sources/consumer-opinion-studies [accessed 20 March 2017].

22. City of Flint, Michigan. Water Filters Available for All Flint Residents. Flint, MI:City of Flint (5 October 2015). Available:​ater-filters-available-for-mdhhs-clients​-all-flint-residents-beginning-tuesday/ [accessed 20 March 2017].

23. Freiss S. For All They Know. (28 November 2016). Available:​-water-crisis-uncertainty/ [accessed 20 March 2017).

24. Kennedy M. Lead-Laced Water in Flint: A Step-By-Step Look at the Makings of a Crisis., America section (20 April 2016). Available:​016/04/20/465545378/lead-laced-water-in-​flint-a-step-by-step-look-at-the-makings​-of-a-crisis [accessed 20 March 2017].

25. City of Flint, Michigan. State of Emergency Declared in the City of Flint. Flint, MI:City of Flint (undated). Available:​rgency/ [accessed 20 March 2017].

26. Southall A. State of emergency declared over man-made water disaster in Michigan City. The New York Times, U.S. section, online edition (17 January 2015). Available:​ama-flint-michigan-water-fema-emergency-​disaster.html [accessed 20 March 2017].

27. Oosting J. State: Flint water delivery order ‘unnecessary’ burden. The Detroit News, News section, online edition (17 November 2016). Available:​chigan/flint-water-crisis/2016/11/17/fli​nt-water-order/94035088/ [accessed 20 March 2017].

28. Oosting J, Chambers J. Schuette backs getting bottled water delivered in Flint. The Detroit News, News section, online edition (17 January 2017). Available:​chigan/flint-water-crisis/2017/01/17/fli​nt-bottled-water/96687386/ [accessed 20 March 2017].

29. Egan P. Judge: state dragging its feet on Flint water deliveries. Detroit Free Press, News section, online edition (24 January 2017). Available:​chigan/flint-water-crisis/2017/01/24/fli​nt-water-bottle-lead-crisi/96993194/ [accessed 20 March 2017].

30. Eggert D. Water Lead-Level Falls Below Federal Limit in Flint. Associated Press (24 January 2017). Available:​a444c308b286d8c14b171ef/apnewsbreak-flin​t-water-has-fallen-below-federal-lead-li​mit [accessed 20 March 2017].

31. Burns G. Unfiltered Flint water is safe, just don’t drink it, says state attorney., News section (24 January 2017). Available:​ssf/2017/01/unfiltered_flint_water_is_sa​fe.html [accessed 20 March 2017].

32. Donn J, et al. PharmaWater I: Pharmaceuticals Found in Drinking Water, Affecting Wildlife and Maybe Humans. Associated Press (undated). Available:​es/pharmawater_site/day1_01.html [accessed 20 March 2017].

33. Abbott Chalew TE, et al. Evaluating nanoparticle breakthrough during drinking water treatment. Environ Health Perspect 121(10):1161–1166 (2013), doi: 10.1289/ehp.1306574.

34. CDC. Diseases and Contaminants. Nitrate and Drinking Water from Private Wells [website]. Atlanta, GA:U.S. Centers for Disease Control and Prevention (updated 1 July 2015). Available:​g/private/wells/disease/nitrate.html [accessed 20 March 2017].

35. EPA. Ground Water and Drinking Water. Legionella [website]. Washington, DC:U.S. Environmental Protection Agency (updated 21 September 2016). Available:​nking-water/legionella [accessed 20 March 2017].

36. Cotruvo J, Cotruvo J Jr. Nontraditional approaches for providing potable water in small systems: part 1. J Am Water Works Assoc 95(3):69–76 (2003).

37. EPA. Drinking Water Requirements for States and Public Water Systems Point-of-Use and Point-of-Entry Treatment Devices [website]. Washington, DC:U.S. Environmental Protection Agency (updated 2 November 2016). Available:​and-point-entry-treatment-devices [accessed 20 March 2017].

38. Ryker SK. Mapping arsenic in ground water: a real need, but a hard problem. Geotimes Newsmagazine of the Earth Sciences 46(11):34–36 (2001). Available: [accessed 20 March 2017].

A Blend of Old and New: Biomonitoring Methods to Study the Exposome

Rachel Cernansky is a freelance journalist in Denver, Colorado, covering science, health, and the environment. She has written for publications including Yale Environment 360, Nature, Civil Eats, and The New York Times.

About This Article open

Citation: Cernansky R. 2017. A blend of old and new: biomonitoring methods to study the exposome. Environ Health Perspect 125:A74;

Published: 31 March 2017

PDF icon PDF Version (502 KB)

Related EHP Article

Biomonitoring in the Era of the Exposome

Kristine K. Dennis, Elizabeth Marder, David M. Balshaw, Yuxia Cui, Michael A. Lynes, Gary J. Patti, Stephen M. Rappaport, Daniel T. Shaughnessy, Martine Vrijheid, and Dana Boyd Barr

The exposome, a concept introduced in 2005, reflects the totality of chemical and nonchemical exposures that an individual accumulates over a lifetime, beginning during prenatal development.1 Whereas traditional biomonitoring targets specific analytes to measure in a sample, exposomic approaches include quantifying hundreds or thousands of analytes simultaneously in what is known as untargeted analysis, and measuring an even greater number of metabolites in so-called high-resolution metabolomics. A new commentary in EHP discusses why both traditional and exposomic approaches are critical to advancing the science of exposure assessment.2

Photo of a rack of blood samples

The authors of a new commentary on exposomic research recommend conducting untargeted analyses of samples collected previously for traditional chemical studies. Researchers must also develop methods to detect and identify low-abundance chemicals in samples and to differentiate between endogenous and exogenous molecules.

© Krisana Sennok/Shutterstock

The commentary is one paper in a six-part series resulting from a workshop held in January 2015. Coauthor David Balshaw, chief of the Exposure, Response, and Technology Branch at the National Institute of Environmental Health Sciences (NIEHS), says that environmental health scientists are increasingly aware of the idea of the exposome, but that some view it with some skepticism—both because of concerns about its untargeted hypothesis-generating approach and because existing technologies are still catching up to the concept of measuring the exposome.

“What this series is intended to do is address those concerns—and to say that [exposomics] is still an emerging concept that needs additional capability and additional validation,” Balshaw says. “As I view the exposome and similar untargeted approaches, they are a tool for hypothesis generation. They do not replace the scientific method; they supplement it.”

The authors identified gaps in existing biomonitoring technology, which formed the basis of eight recommendations discussed in the commentary for advancing exposomic research. Among these recommendations are conducting untargeted analyses of samples collected previously for traditional targeted chemical studies, creating tools to search across multiple complementary databases, and developing chemistry methods to detect low-abundance chemicals and differentiate between endogenous and exogenous molecules among the thousands measured in an untargeted analysis.

A fourth recommendation is developing bioinformatics techniques to enhance detection of unknown chemicals. Balshaw points to coauthor Gary Patti’s work on untargeted metabolomics as an example of how the recommendations build on existing work. In a 2012 study3 Patti and colleagues described a way to more efficiently identify metabolites detected through untargeted studies. The authors came up with a database and workflow to automate the processing of the voluminous data produced by such studies.

Looking ahead, Balshaw says the NIEHS-led Children’s Health Exposure Analysis Resource (CHEAR)4 will focus on developing infrastructure to help realize the recommendations in the commentary. Among other services, CHEAR conducts both targeted and untargeted analyses of biosamples collected by children’s health researchers funded by the National Institutes of Health (NIH).

Robert Wright, director of the Lautenberg Laboratory for Environmental Health at Mount Sinai’s Icahn School of Medicine, thinks the discussion of current data-processing capacity is the commentary’s key contribution. “This is probably the best paper I’ve ever read in terms of detailing how to measure the exposomics assays, and laying out the bioinformatics challenges as well,” he says. “I think people get caught up in the technology of measuring the assay, but you have to do something with [the data produced].” Wright was not involved in the commentary.

For Tracey Woodruff, director of the Program on Reproductive Health and the Environment at the University of California, San Francisco, another important consideration is the potential to identify chemical signatures—unique patterns of changes in molecules associated with a particular exposure or health end point—which she sees as a crucial link between research and application. “If our goal is to improve health, then we have to figure out [which signatures] are bad,” she says. Woodruff was not involved with the commentary.

Woodruff is encouraged by the emphasis on untargeted studies. To her, the commentary is an important indication that NIH is looking to include more hypothesis-generating research in its funding portfolio—as opposed to projects using the traditional approach of starting with a particular hypothesis, which, like targeted studies, is more limiting.

“Post-its weren’t created because someone was trying to create a Post-it. They were invented because someone was trying to make glue for something else. We’re trying to find a new Post-it,” Woodruff says. “That NIH is saying we want to see more of this type of broad research in the field is very, very important. It represents a commitment to a shift in the type of research that’s being funded and ultimately toward supporting efforts to identify environmental contributors to disease.”


1. Wild CP. Complementing the genome with an “exposome”: the outstanding challenge of environmental exposure measurement in molecular epidemiology. Cancer Epidemiol Biomarkers Prev 14(8):1847–1850 (2005), doi: 10.1158/1055-9965.EPI-05-0456.

2. Dennis KK, et al. Biomonitoring in the era of the exposome. Environ Health Perspect 125(4):502–510 (2017), doi: 10.1289/EHP474.

3. Tautenhahn R, et al. An accelerated workflow for untargeted metabolomics using the METLIN database. Nat Biotechnol 30(9):826–828 (2012), doi: 10.1038/nbt.2348.

4. Children’s Health Exposure Analysis Resource (CHEAR) [website]. Research Triangle Park, NC:National Institute of Environmental Health Sciences, National Institutes of Health, U.S. Department of Health and Human Services (updated 29 November 2016). Available:​rted/exposure/chear/ [accessed 9 March 2017].

Way to Go: Identifying Routes for Walkers and Cyclists to Avoid Air Pollutants

Carol Potera, based in Montana, also writes for Microbe, Genetic Engineering News, and the American Journal of Nursing.

About This Article open

Citation: Potera C. 2017. Way to go: identifying routes for walkers and cyclists to avoid air pollutants. Environ Health Perspect 125:A71;

Published: 31 March 2017

PDF icon PDF Version (2.1 MB)

Exposures to air pollutants may offset a portion of the health benefits of walking and bicycling in cities.1 However, taking a detour just a block or two away from the busiest streets and roads “can make a big difference in your exposure,” says Steve Hankey, an assistant professor at Virginia Polytechnic Institute and State University and coauthor of a new study in EHP.2

Photo of the author’s bicycle and mobile air sampler

Over the course of the study, coauthor Steve Hankey covered about 1,000 miles on his bike, pulling a mobile air sampler through the streets of Minneapolis.

© Steve Hankey

For every street block in Minneapolis, Minnesota—13,604 in all—Hankey and his colleagues modeled the exposure of pedestrians and cyclists to particulate air pollution during the afternoon rush hour (4:00–6:00 p.m.). Their models of pedestrian and bicycle traffic and of air quality showed that 3–7% of city blocks were what they called “sour spots,” with high levels of air pollution and high numbers of walkers and cyclists. These sour spots occurred downtown around retail stores, in the business district, and along main traffic arteries.

In contrast, 2–3% of blocks were “sweet spots,” with high rates of walking and cycling but low air pollution. Sweet spots were mostly close to the city center (and thus still walkable) but just outside of the actual downtown area (and thus had lower levels of vehicular pollution). The researchers estimate that shifting cyclists and pedestrians during rush hour from high-traffic roads onto low-traffic roads just one block away could decrease these individuals’ exposure concentrations of ultrafine particulate matter by 11%, black carbon by 19%, and fine particulate matter by 3%.2

“There was a high spatial mismatch between where people walked and biked and where pollution was high,” says Hankey. For example, 49% of walking and 29% of cycling in Minneapolis occurred on high-traffic, polluted streets.2 Hankey suggests that people living or working in cities likely could find cleaner air by walking or biking in less-trafficked areas.

The modeling results suggest strategies that city planners might use to redesign cycling and vehicular traffic flows as a way to improve public health. Practical solutions could include the establishment of bike-friendly corridors by adding speed bumps or creating one-way roads. Alternatively, bicycle routes could be relocated onto low-traffic roads, and bus traffic along popular walking routes could be shifted to corridors directly adjacent to those routes.

Hankey was a graduate student at the University of Minnesota at the time of the study. In 2012, on approximately 40 runs between late August and the end of October, he rode a bicycle pulling a trailer with instruments that measured particulate air pollutant levels. He rode around Minneapolis on three 20-mile routes that covered different road types and land uses. The data he collected were used to build a statistical model to estimate air quality block by block across the city.

“We were testing how mobile sampling by bike compares to traditional fixed-site monitoring,” says Hankey. As their name suggests, fixed-site monitors detect air quality only in the area immediately surrounding them. In contrast, the mobile sampling method used in the current study detects small changes in concentrations of pollutants all along routes where people actually walk and bike.

Hankey plans to collect more data on air pollution and numbers of pedestrians and cyclists throughout the entire day, rather than just during afternoon rush hour. “I want to scale up … to compare cities nationwide and provide more useful information for all Americans,” he says.

“[Hankey’s] novel study provides important new insights in population-level spatial patterns of exposure to air pollution during active travel that may be important for planning low-exposure cities that are overall health protective,” says Mark Nieuwenhuijsen, director of Air Pollution and Urban Environment at ISGlobal in Barcelona, Spain. In addition to shifting active travel away from major roads, Nieuwenhuijsen, who was not involved in the work, says a more sensible approach would be to reduce car use and try to create car-free zones or even car-free cities.3 He says, “This would reduce air pollution, noise, heat island effects, and sedentary behavior, and increase green space.”


1. Hankey S, et al. Health impacts of the built environment: within-urban variability in physical inactivity, air pollution, and ischemic heart disease mortality. Environ Health Perspect 120(2):247–253 (2012), doi: 10.1289/ehp.1103806.

2. Hankey S, et al. Population-level exposure to particulate air pollution during active travel: planning for low-exposure, health-promoting cities. Environ Health Perspect 125(4):527–534 (2017), doi: 10.1289/EHP442.

3. Nieuwenhuijsen MJ, Khreis H. Car free cities: pathway to healthy urban living. Environ Int 94(16):251–262 (2016), doi: 10.1016/j.envint.2016.05.032.

A Satellite–Ground Hybrid Approach: Relative Risks for Exposures to PM2.5 Estimated from a Combination of Data Sources

Nate Seltenrich covers science and the environment from Petaluma, CA. His work has appeared in High Country News, Sierra, Yale Environment 360, Earth Island Journal, and other regional and national publications.

About This Article open

Citation: Seltenrich N. 2017. A satellite–ground hybrid approach: relative risks for exposures to PM2.5 estimated from a combination of data sources. Environ Health Perspect 125:A73;

Published: 31 March 2017

PDF icon PDF Version (360 KB)

Related EHP Article

Comparing the Health Effects of Ambient Particulate Matter Estimated Using Ground-Based versus Remote Sensing Exposure Estimates

Michael Jerrett, Michelle C. Turner, Bernardo S. Beckerman, C. Arden Pope III, Aaron van Donkelaar, Randall V. Martin, Marc Serre, Dan Crouse, Susan M. Gapstur, Daniel Krewski, W. Ryan Diver, Patricia F. Coogan, George D. Thurston, and Richard T. Burnett

Satellite instruments offer researchers powerful new perspectives and data sources for studying the environment. A new study used associations between fine particulate matter (PM2.5) and mortality from circulatory diseases as a test scenario to explore how exposure estimates derived from remote sensing alone compare with those produced by a combination of satellite- and ground-based data.1 The findings showed associations between PM2.5 and mortality regardless of the method used, but specific relative risk estimates varied widely, with hybrid models generally predicting the strongest associations.

Photo of a man receiving an electrocardiograph in the emergency department

Fine particulate matter has been strongly implicated in cardiovascular problems including heart attacks and strokes. Neither remote sensing nor ground-based monitors alone can fully capture a population’s exposures to air pollutants, but the combined use of these technologies may paint a more complete picture.

© BSIP SA/Alamy Stock Photo

NASA’s Terra satellite carries a variety of instruments for observing our planet’s atmosphere and surface from a height of 443 miles.2 Remote sensing is best at detecting differences on a broad scale, such as changes in air quality over time or across geographic regions, as well as pollution sources and sinks, says professor Yang Liu of Emory University, who was not involved in the study.3,4 Ground-based monitors, on the other hand, directly measure air pollutants in their vicinity but may not be accurate for estimating exposures to individuals who are not at the same location. In addition, ground-based monitors are not available in many locations.

As researchers worldwide have enjoyed increasing access to high-resolution air quality and atmospheric data obtained through remote sensing, they have developed ways of using the data to more accurately estimate human exposures to air pollution. Recent trends and advances call for a closer investigation into similarities and differences among satellite- and ground-based models, says lead author Michael Jerrett, chair of the Department of Environmental Health Sciences at the University of California, Los Angeles, Fielding School of Public Health.

“We felt that with the proliferation of models and different approaches, we wanted to be able to understand what this would mean for exposure estimates and therefore for policy and global burden-of-disease calculations,” Jerrett says.

Jerrett and colleagues selected a representative sample of seven such models for comparison, to see how differences among them would influence estimates of health effects on U.S. residents enrolled in the American Cancer Society Cancer Prevention Study II.5 Some of the models relied exclusively on satellite data, some on ground-based monitors, and others on a combination of data sources.

The researchers geocoded the residences of 668,629 participants for the years 2002–2004 and estimated individual exposures to PM2.5 using the methods dictated by the various models. The researchers then estimated relative risk of mortality from circulatory diseases in association with the PM2.5 exposures predicted by each approach.6

They discovered that associations between PM2.5 and mortality were significant for all seven models—a meaningful though tangential result, Jerrett says. However, the models differed considerably in their estimate of the magnitude of that association. The smallest estimate of relative risk came from the two remote-sensing models without ground data, and the largest from the model with the most extensive ground data: a blend of air quality measurements from monitors, land-use modeling, and traffic density data.

“The key finding of the paper is that when you don’t have any ground data represented in remote-sensing models, they do tend to produce risk estimates that appear to be biased toward the null,” Jerrett explains—in other words, the actual relative risk may be underestimated. This has important implications for researchers investigating health effects of air pollution in parts of the world without sufficient ground-based monitors to fill in the complementary remote-sensing data. The findings support the conclusion that whenever possible, hybrid models and combinations of multiple models should be used for maximum coverage and accuracy.

Coauthors Randall Martin and Aaron von Donkelaar of Dalhousie University, who developed the study’s remote-sensing exposure models, agree that satellites should not stand alone. “There are a variety of information sources from which to obtain and learn about PM2.5, and we have often promoted the use of as many of those as possible,” Martin says. “The results provide motivation to pay close attention to how ground data are fused with remote-sensing estimates. They are independent measurements that have arisen for largely different reasons, but there’s synergy in combining both information sources to learn what we can about fine particulate matter.”

Beyond clarifying this relationship, the study should have even more tangible, direct benefits for the field, says professor Julian Marshall of the University of Washington, who was not affiliated with the study. Marshall praised the study for its use of multiple rigorous estimates of exposure. “The authors went to significant lengths to get many robust exposure estimates,” he says. “Many of the methods used here are publicly available and could be useful for many, many cohorts. That opens up many new doors.”


1. Jerrett M, et al. Comparing the health effects of ambient particulate matter estimated using ground-based versus remote sensing exposure estimates. Environ Health Perspect 125(4):552–559 (2017), doi: 10.1289/EHP575.

2. NASA. Terra Spacecraft [website]. Washington, DC:National Aeronautics and Space Administration (updated 30 July 2015). Available:​spacecraft/index.html [accessed 20 September 2016].

3. van Donkelaar A, et al. Global estimates of ambient fine particulate matter concentrations from satellite-based aerosol optical depth: development and application. Environ Health Perspect 118(6):847–855 (2010), doi: 10.1289/ehp.0901623.

4. Geddes JA, et al. Long-term trends worldwide in ambient NO2 concentrations inferred from satellite observations. Environ Health Perspect 124(3):281–289 (2016), doi: 10.1289/ehp.1409567.

5. ACS. Cancer Prevention Study II (CPS II) [website]. Atlanta, GA:American Cancer Society (undated). Available:​opreventcancer/currentcancerpreventionst​udies/cancer-prevention-study [accessed 20 September 2016].

6. MedCalc. Cox proportional-hazards regression [website]. Ostend, Belgium:MedCalc Software (undated). Available:​rtional_hazards.php [accessed 20 September 2016].

Programming the Future: Epigenetics in the Context of DOHaD

Julia R. Barrett, MS, ELS, a Madison, WI–based science writer and editor, is a member of the National Association of Science Writers and the Board of Editors in the Life Sciences.

About This Article open

Citation: Barrett JR. 2017. Programming the future: epigenetics in the context of DOHaD. Environ Health Perspect 125:A72;

Published: 31 March 2017

PDF icon PDF Version (191 KB)

Related EHP Article

Small-Magnitude Effect Sizes in Epigenetic End Points are Important in Children’s Environmental Health Studies: The Children’s Environmental Health and Disease Prevention Research Center’s Epigenetics Working Group

Carrie V. Breton, Carmen J. Marsit, Elaine Faustman, Kari Nadeau, Jaclyn M. Goodrich, Dana C. Dolinoy, Julie Herbstman, Nina Holland, Janine M. LaSalle, Rebecca Schmidt, Paul Yousefi, Frederica Perera, Bonnie R. Joubert, Joseph Wiemels, Michele Taylor, Ivana V. Yang, Rui Chen, Kinjal M. Hew, Deborah M. Hussey Freeland, Rachel Miller, and Susan K. Murphy

Recent studies have shown that variable responses to environmental exposures within a population arise in part from individuals’ genetic differences.1,2,3 Research on these differences is increasingly focusing on the epigenome, in which small chemical tags on DNA and associated proteins fine-tune genetic expression.1,2,4 A new review in EHP takes stock of the methods, analyses, and complexity of environmental epigenetics research in the context of the developmental origins of health and disease (DOHaD).1

According to the DOHaD hypothesis, environmental exposures during pre- and postnatal development can affect health years or even decades later.1,4 A potential bridge between these exposures and outcomes involves changes in, or reprogramming of, the epigenome.2,3,4,5,6

Illustration depicting a strand of DNA with a methyl group attached

Methylation is the attachment of a methyl group to DNA at regions where cytosine and guanine bases are paired. Methyl groups can activate or silence genes. This process affects whether factors that would normally cause the gene to be expressed will still do so.Jane Whitney for EHP

Epigenetic modifications are generally a critical part of normal development, helping to activate or silence specific genes during cell differentiation and thereby directing the formation of various tissues.7 But not all changes are benign, as suggested by associations between specific epigenetic alterations and disorders including cancer, neurodegeneration, and diseases of the cardiovascular and immune systems.2

Epigenetic programming has therefore drawn intense scrutiny as a potential biological mechanism through which environmental factors may influence health and susceptibility to disease.2,3 In addition, assessment of epigenetic alterations may help researchers to detect effects of exposures long after they have occurred and to better characterize disease risks.1,2,6

The review summarizes research pertaining to the detection and interpretation of epigenetic changes as the basis for the DOHaD hypothesis, which is a major objective of research under way in the federally funded Children’s Environmental Health and Disease Prevention Research Centers.1 “There are a lot of individuals within different children’s health centers who are very interested first and foremost in environmental health effects, but also in what role epigenetics can play in driving and explaining some of these health effects, specifically thinking about early-life development,” says first author Carrie Breton, an assistant professor of preventive medicine at the University of Southern California Keck School of Medicine.

For the review, Breton and her co-authors focused on DNA methylation, a specific type of epigenetic modification in which the presence or absence of a methyl tag can control whether a gene is active (transcribed) or silent (not transcribed). Researchers have identified subtle differences in blood measures of methylation between exposed and unexposed populations, with differences in methylation ranging from less than 2% up to 10%.1 “Often, we only see something like two to five methylated loci out of one hundred. That’s what, in this article, we are referring to when we talk about small magnitudes of change,” says Breton.

Whether such small changes are biologically meaningful is very much an open question in environmental health research. Previous genetic research has suggested that the location of epigenetic alterations can be a crucial determinant of the overall effect. For example, one study estimated that a 1% increase in methylation in a specific area of the IGF2 gene halved transcription, while a 1% decrease doubled transcription.8 The scale of the estimated change in transcription is on par with that found for other genes in cancerous tissues.1 This finding illustrates that even small changes may have a large impact. In some cases, the effect may be indirect, with epigenetic changes poising a gene to react to a later trigger.1

However, it is important not to lose sight of the fact that epigenetic changes occur against the backdrop of the entire genome. “We can’t forget that it’s still a layer of information on top of our genetic code,” says Breton. She says that genome-wide association studies and many years of looking at genetic variation have shown that some diseases have an important genetic component.

“You really need to start thinking about how the epigenome responds to the environment and affects disease risk—but on top of the backbone of the genetic code,” Breton says. “You also need to think about measuring and adjusting for genetic variation or looking at interactions between the genetic code and the epigenetic code.”

Ultimately, researchers hope to be able to determine if specific epigenetic alterations bridge the gap between an environmental exposure and a particular outcome. “I think this is a very useful paper to a lot of people in terms of having a kind of common lexicon to work with [in terms of how to identify and describe small changes], and I like that they encourage discussion about small effect sizes,” says Daniele Fallin, a professor at the Johns Hopkins Bloomberg School of Public Health who was not involved in the review. “We are at that stage where there is still so much to learn; bringing people together and then communicating a common landscape are important.”


1. Breton CV, et al. Small magnitude effect sizes in epigenetic endpoints are important in children’s environmental health studies. Environ Health Perspect 125(4):511–526 (2017), doi: 10.1289/EHP595.

2. Ladd-Acosta C, Fallin MD. The role of epigenetics in genetic and environmental epidemiology. Epigenomics 8(2):271–283 (2016), doi: 10.2217/epi.15.102.

3. Bakulski KM, Fallin MD. Epigenetic epidemiology: promises for public health research. Environ Mol Mutagen 55(3):171–183 (2014), doi: 10.1002/em.21850.

4. Burris HH, Baccarelli AA. Environmental epigenetics: from novelty to scientific discipline. J Appl Toxicol 34(2):113–116 (2014), doi: 10.1002/jat.2904.

5. Mitchell C, et al. DNA methylation, early life environment, and health outcomes. Pediatr Res 79(1–2):212–219 (2016), doi: 10.1038/pr.2015.193.

6. Rozek LS, et al. Epigenetics: relevance and implications for public health. Annu Rev Public Health 35:105–122 (2014), doi: 10.1146/annurev-publhealth-032013-182513.

7. Tang WW, et al. A unique gene regulatory network resets the human germline epigenome for development. Cell 161(6):1453–1467 (2015), doi: 10.1016/j.cell.2015.04.053.

8. Murphy SK, et al. Gender-specific methylation differences in relation to prenatal exposure to cigarette smoke. Gene 494(1):36–43 (2012), doi: 10.1016/j.gene.2011.11.062.

The Emergence of Environmental Health Literacy—From Its Roots to Its Future Potential

Author Affiliations open
Division of Extramural Research and Training, National Institute of Environmental Health Sciences, National Institutes of Health, Department of Health and Human Resources, Research Triangle Park, North Carolina, USA

PDF icon PDF Version (764 KB)

  • Background: Environmental health literacy (EHL) is coalescing into a new subdiscipline that combines key principles and procedural elements from the fields of risk communication, health literacy, environmental health sciences (EHS), communications research, and safety culture. These disciplines have contributed unique expertise and perspectives to the development of EHL. Since 1992, the National Institute of Environmental Health Sciences (NIEHS) has contributed to the evolution of EHL and now seeks to stimulate its scientific advancement and rigor.

    Objectives: The principal objective of this article is to stimulate a conversation on, and advance research in, EHL.

    Discussion: In this article, we propose a definition of and conceptual framework for EHL, describe EHL in its social and historical context, identify the complementary fields and domains where EHL is being defined and implemented, and outline a research agenda. Extensive reviews of web and literature searches indicate that the concept of EHL is evolving rapidly, as are the definitions of its scope and inquiry. Although several authors have outlined different frameworks, we believe that a more nuanced model based on Bloom’s taxonomy is better suited to EHL and to future research in this area.

    Conclusions: We posit that EHL can potentially benefit the conduct and outcomes of community-engaged and health disparities EHS research and can ensure that the translation of research findings will lead to greater understanding of specific risks, reduction of exposures, and improvement of health outcomes for individuals and communities. We provide four recommendations to advance work in EHL.

  • Citation: Finn S, O’Fallon L. 2017. The emergence of environmental health literacy—from its roots to its future potential. Environ Health Perspect 125:495–501;

    Address correspondence to L. O’Fallon, Program Analyst, Division of Extramural Research and Training, National Institute of Environmental Health Sciences, National Institutes of Health, U.S. Department of Health and Human Services, P.O. Box 12233 (MD K3-13), Research Triangle Park, NC 27709. Telephone: (919) 541-7733. E-mail:

    The authors declare they have no actual or potential competing financial interests.

    Received: 10 October 2014
    Accepted: 25 June 2015
    Advance Publication: 30 June 2015
    Final Publication: 31 March 2017

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact Our staff will work with you to assess and meet your accessibility needs within 3 working days.

Defining the Scope and Purpose of Environmental Health Literacy (EHL)

Fundamentally, environmental health literacy (EHL) begins with an understanding of the link between environmental exposures and health. EHL has recently coalesced as a new subdiscipline combining key principles and procedural elements from the fields of health literacy, risk communication, environmental health sciences (EHS), communications research, and safety culture (Biocca 2004; Chinn 2011; Edwards et al. 2013; Fitzpatrick-Lewis et al. 2010; Nicholson 2000). Each of these disciplines has contributed unique frameworks and perspectives to the development of EHL as a distinct subfield and is likely to continue to inform the evolution of EHL.

The purpose of this article is to propose a definition of and a conceptual framework for EHL, to understand EHL in its social and historical contexts, and to identify the complementary fields and domains where EHL is being defined and implemented. This commentary acknowledges the value of current academic efforts to delineate the progressive nature of EHL that begins with an individual’s understanding and proceeds to the ability to create new information because similar to health literacy, EHL is not a static achievement, but an evolutionary process.

Another purpose of this article is to highlight the role that the National Institute of Environmental Health Sciences (NIEHS) has played in advancing the concept of EHL and to outline a research agenda that will move forward and stimulate the development of research on this topic. Similar to the validated benefits health literacy can provide in biomedical settings (Benjamin 2010; Lin et al. 2004), we propose that EHL can potentially benefit the conduct and outcomes of community-engaged and health disparities environmental health sciences (EHS) research as well as efforts to promote environmental justice. We also propose that EHL can ensure that the translation of research findings leads to a greater understanding of specific risks, reduction of exposures, and improvement of health outcomes for individuals and communities.

Our extensive literature searches of PubMed ( and Web of Science ( confirm that the field is evolving rapidly, as are definitions of the scope of inquiry and purpose of EHL. Academic endeavors to date have focused primarily on elucidating the attributes of EHL and on the stages of becoming literate about environmental health concepts and issues (Kaphingst et al. 2012; Sørensen et al. 2012). These academic efforts have built upon conceptual frameworks from the fields of health literacy and risk communication to define the progression of understanding from basic knowledge to comprehension and application (Colucci-Gray et al. 2006; Guidotti 2013; Nutbeam 2008). Addressing gaps in education and promoting EHL among health care professionals via curricula and educational module development is another major theme that emerged from the literature review (Barnes et al. 2010; Gehle et al. 2011).

A review of the existing literature related to EHL makes it clear, however, that raising EHL is more than simply the stages of an educational process. It also represents a philosophical perspective, a public health policy to improve literacy and health literacy in the general public, and a set of strategies to empower individuals and communities to exert control over the environmental exposures that may lead to adverse health outcomes (Estacio 2013; Minkler et al. 2008; Mogford et al. 2011; Zoller 2012).

Environmental health literacy integrates concepts from both environmental literacy and health literacy to develop the wide range of skills and competencies that people need in order to seek out, comprehend, evaluate, and use environmental health information to make informed choices, reduce health risks, improve quality of life and protect the environment. (Society for Public Health Education;​/key_ehl.asp)

Existing definitions of EHL, such as the one that the Society for Public Health Education (SOPHE) first outlined in 2008, often include language connoting the evolutionary nature and stages of EHL (Hatfield 1994; Nutbeam 2009); however, we propose a baseline definition that emphasizes the underlying issue: an understanding of the connection between environmental exposures and human health. As we discuss later, this understanding is only the first stage of a hierarchy of increasing literacy. We believe that this baseline definition enables EHL to be described through related disciplinary perspectives such as health literacy, risk communication, EHS, communications, public health, and the social sciences. As EHL evolves, it will be measured and applied in many ways depending on the disciplinary lens, the aim, and the audience.

The Historical Roots of Environmental Health Literacy

There are a number of different sources of the emergence of environmental health literacy (Figure 1). Risk communication, one of EHL’s roots, has deep historical origins and can be traced to the display of symbols in ancient cultures to connote tribal and state affiliations on the battlefield. More recent historical examples of risk communication also utilized symbols to connote danger: the well-known skull and crossbones symbol used initially by pirates and then later as the symbol for poison, and the color red that is widely used to indicate “stop” or “danger” (Hancock et al. 2004). World War II expanded the symbolic vocabulary for dangerous and toxic situations, and the postwar era adopted much of this military iconography in high-risk and dangerous settings related to toxic chemicals, imminent danger, poison, and, increasingly in the 1950s, as symbols for nuclear energy’s threat (Matthews et al. 2014; Young 1998). Symbolic representations are recognized as an effective and appropriate method of communicating hazards; however, cultural differences in risk perception and in the interpretation of specific colors or icons has led to the consideration of universal symbols and to research evaluating the optimal formats for communicating environmental risks (Chan and Ng 2012; Lesch et al. 2009).

Figure 1. Conceptual diagram.

Figure 1. The cultural context: streams leading to the coalescence of environmental health literacy.

View larger image (TIF File)

More recent impetus for the development of EHL began in the late 20th century with the recognition that risks to human health came from a number of different environmental sources and had varying levels of immediate toxicity that could not be sufficiently communicated via an icon or symbol (Hancock et al. 2004). This understanding of complex risk was encapsulated in the 1960s with the publication of Silent Spring (Carson et al. 1962) and was further elucidated by Rachel Carson’s testimony to Congress on pesticides in 1963. Although much of the data that she presented was known to the scientific community, Carson was the first to explain to policy makers and the general public the far-reaching consequences of the introduction of chemicals into the environment in such compelling and convincing terms. Through her vision of a compromised environment, “Carson, the citizen–scientist, spawned a revolution” (Griswold 2012) that led to the rise of organized environmental activism.

Concurrent with this new societal awareness of environmental risks, the NIEHS and the U.S. Environmental Protection Agency (EPA) were established (1966 and 1970, respectively), and early efforts to explore environmental sciences expanded into consideration of the effects of pollutants and other environmental exposures on human health. Since 1970, EHL has been coalescing as a distinct field in direct proportion to the federal commitment to provide information to the public, including EHS research findings, and to the increased public awareness of environmental risks.

Articles describing the historical basis for the emergence of EHL often point to its roots in the health literacy movement in the United States. However, EHL is more than an extension of health literacy, it is the logical and inevitable outcome of the validation of health literacy to improve health outcomes and treatment adherence (Benjamin 2010; Paasche-Orlow et al. 2005) and the extrapolation of that value to the prevention of environmentally induced disease. The coalescence of EHL as a distinct subfield may also be attributed to the recognition of the public health implications of environmental health research with affected communities (Brown et al. 2012; Perez et al. 2012) and the need for research to identify and address environmental risks. Recent reports show that health literacy efforts have evolved, and these reports indicate recognition of the need to move beyond the health care setting and system [Institute of Medicine of the National Academies (IOM) 2004, 2011]. EHL acknowledges this need and addresses the health context of the individual and the community. The goals of EHL are consistently focused on preventing illness by raising awareness of risks from environmental factors and by providing approaches that individuals and communities can take to avoid, mitigate, or reduce such exposures.

The cultural shift in the value of scientific literacy among the general public also stimulated the evolution of the concept of EHL. Analogous to the rise of bioethics in the context of genetics research, EHL arose in response to growing public interest in the environment as well as to scientific and technological advancements that were increasingly available to the public. Furthermore, the emergence of the Environmental Justice movement drew political attention to inequitable and disproportionate environmental exposures faced by low-income, minority, and indigenous populations (Stokes et al. 2010). These and other concerns about environmental pollutants in air, food, and water also led to the emergence of citizen science and the necessity for health risk communications related to environmental exposures (Bonney et al. 2014; Conrad and Hilchey 2011; Minkler et al. 2010).

Scientific and technological developments also contributed to the evolution of communication modalities related to environmental risk that are not dependent on reading ability. In this context, the emergence of EHL can be considered the next stage in risk assessment and a reflection of advances in the fields of exposure assessment and exposure biology. In the 1980s and 1990s, technologies were developed to measure environmental toxicants, standards and regulations were established for chemical exposure and “levels of concern,” and there was an increase in the availability of computer-based visual representations of risk. With the widespread adoption of computers in the 1990s and the development of geographic information system (GIS) mapping software, computer-based visual representations of risk and the ability to link relative risk to geographic locations emerged as an accessible and cost-effective communication modality for the public (Lahr and Kooistra 2010; Severtson 2013). The field of risk communication was an early adopter of visual representations of risk. Such communications represented the most rapid means of translating evidence into risk messages and offered a modality that was both understandable and meaningful for individuals with varying levels of basic and scientific literacy (Hermer and Hunt 1996; LePrevost et al. 2013).

The roots of EHL can also be traced to widespread public awareness of human-made technological disasters that caused large-scale environmental pollution (Brennan 2009). Since the 1980s, media attention to such accidents has been so extensive that one need only mention the Bhopal chemical spill, Love Canal, the Three Mile Island, Chernobyl, or Fukushima nuclear accidents, or the Exxon Valdez or Deepwater Horizon oil spills to elicit images of severe and pervasive contamination. The impact of these disasters was communicated by newspaper photos of oil-soaked marine birds or workers in HazMat suits, televised images of billowing clouds of oil gushing from the wellhead, or YouTube videos of tar balls on the beach. Public attention to such extreme polluting events is heightened by the ever-increasing amounts of information on the Internet about the negative health impacts of the multiple exposures we all experience throughout our lives (Murphy et al. 2010).

The Social Context Underlying the Development of EHL

Although several authors recognize the various roots that have come together and flowered into the emergence of EHL (Baur 2010; Huber et al. 2012), there is little in the literature that explores the larger cultural context that underlies how the public understands environmental health risks. As efforts are made to promote the value of EHL, it will be important to comprehend and address public understanding and misunderstanding of environmental risks and how this knowledge has been informed and defined by cultural media (i.e., books, films, television) (Frayling 2005; Kennedy et al. 2011; Moore 2015; Murphy et al. 2010).

Films have historically explored and exploited public awareness of the negative aspects of increasing environmental exposures. Film studies of cinematic trends have consistently recognized the thematic prevalence of “nuclear anxiety” in films from the 1950s and the plethora of films that depicted the horrendous “atomic mutations and mass devastation” resulting from nuclear exposure (Newman 2000). Films produced since the 1970s, in contrast, have focused on pollution more generally and the threats posed by toxic waste, contamination of the food chain and water supplies, and the increasing reality of diminishing resources (Frayling 2005). Unfortunately, cultural expressions about the outcomes of environmental pollution, as depicted in movies and books, have too often portrayed such scenarios in overly dramatic or unrealistic terms (Murray and Heumann 2014). Despite a few examples of positive outcomes (e.g., A Civil Action, Silkwood), the majority of cultural depictions of diminishing resources do not reflect optimism that science can “fix” pollution. Rather, the postapocalyptic film trend reflects a pervasive attitude that our current actions will lead to barbaric societies where diminishing resources have been completely depleted and climatological changes have spun out of control (e.g., Mad Max, The Hunger Games, The Day After Tomorrow).

The scientific community recognizes that media, and most recently social media, play a key role in public understanding of environmental risk (Fitzpatrick-Lewis et al. 2010; McCallum et al. 1991). Publications and news reports that are evidence based and reflect an understanding of science represent positive examples of media representation of environmental risks. However, the media can misrepresent environmental risks (and indeed have done so), tending to focus on the most dramatic aspects of exposure events and disasters, and presenting news about the outcomes of environmental health science research as a means of driving specific political agendas (Jaspal and Nerlich 2014a, 2014b). These information challenges must be considered as efforts are made to build EHL, especially when attempting to raise public understanding of actual versus perceived risks from environmental exposures.

Ultimately, evidence-based environmental health risk communications can help to provide more accurate evidence to counterbalance media and cultural representations of environmental degradation and its impact on human health. Furthermore, raising EHL can help individuals to navigate the abundance of information, of varying quality and veracity, that is available on the Internet (e.g., on-line blogs, chat rooms, other forms of social media) and can empower them to decide what choices are best for their health and that of their families (Wilcox 2012). More important, improving knowledge about environmental health risks can be used to promote a more optimistic view of the potential that exists to reduce, mitigate, or eliminate the worst environmental exposures and improve the health of both humans and the environment.

EHL Methodology and Approaches

EHL builds on, synthesizes, and encompasses validated tools and methodologies from existing fields of research such as health literacy, risk communication, and education. Although the development of these approaches is most closely based on health literacy concepts and practices, several authors working in this emerging field conceptualize EHL as a process that individuals and communities embrace as a means of critical reflection within their local socioeconomic context rather than as a type of health literacy that incorporates specialized knowledge of environmental factors (Chinn 2011; Sykes et al. 2013). This concept of critical reflection was initially proposed by Nutbeam as one of three phases of learning and processing that reflect the evolutionary nature of health literacy (Nutbeam 2008). Although a number of articles cite this three-stage conceptual framework for EHL, we propose adapting Bloom’s taxonomy of educational objectives as a more nuanced model for the evolutionary nature of becoming more literate about environmental health issues (Bloom 1956) (see Figure 2).

Figure 2. Conceptual diagram.

Figure 2. Conceptual model of environmental health literacy adapted from Bloom (1956), representing the potential for different levels of EHL across various environmental health science topics.

View larger image (TIF File)

Since its publication, more than 5,000 authors have found Bloom’s taxonomy to be a useful construct (Flinders 1996). Bloom’s stepwise progression of six distinct educational stages is a fitting approach for the development of targeted interventions for the various stages of EHL. The value of this model for describing the evolution of learning and understanding is that it acknowledges an individual’s potential for environmental health literacy at each stage. For example, those at the earliest stage, “Recognition,” know that a specific substance is toxic and may affect their health without any other understanding of how this occurs, what levels are concerning, or how to mitigate the exposure. This is, nonetheless, an initial stage of environmental health literacy. As the model suggests, the goal of EHL is to continue to promote greater understanding, to improve an individual’s extrapolation of knowledge to other potential environmental risks, and to stimulate actions based on the understanding of risk. However, the model is not meant to suggest a single path upward to total literacy or an equal level of literacy about different exposures; like the disease-specific nature of health literacy, an individual’s environmental health literacy may vary from topic to topic. For example, someone may have achieved a high degree of EHL related to asthma because of ongoing family experiences with this condition as well as the widespread public information linking asthman to air pollution, and yet possess a very low EHL regarding breast cancer and its lesser-known connections to environmental exposures.

The stages in the taxonomy also indicate the type of action individuals and communities might take based on their level of EHL. These actions can be wide-ranging, from an individual decision to avoid certain personal care products to a union movement to improve workplace conditions, each of which might represent a single stage of environmental health literacy. An example that represents all stages of this model could be a statewide movement to address potential health effects from hydraulic fracturing that builds from the recognition of the exposure to an extrapolation of a health risk to the creation of policy to address the risk. Individuals who are proficient in EHL are able to recognize their exposures and exert some manner of control over them rather than feeling as if “there’s nothing I can do.”

Environmental exposures most commonly affect communities as a whole; however, individual health outcomes arising from these exposures are dependent on an individual’s socioeconomic, biological, and psychological susceptibility to these exposures (Lee et al. 2005; Quandt et al. 2004). Therefore, efforts to promote EHL should include ways to measure literacy at individual and community levels as well as a range of information that recognizes the psychosocial and demographic heterogeneity within communities and the potential for distinct medical, psychological, or cultural responses to a common source of exposure(s). To be truly effective, efforts to promote EHL should be based on the types of awareness and knowledge needed, and they should use validated and culturally sensitive strategies to best promote the uptake of information by individuals, communities, public health officials, health care providers, or in regulatory or policy settings (Arcury et al. 2010; Ramos et al. 2001). An understanding of environmental health risks could serve as a needed mediator to improve media representations of environmental health science and in popular cultural representations of the relationship between the environment and health (Fitzpatrick-Lewis et al. 2010). More critically, raising EHL could be an important goal of science, technology, engineering, and math (STEM) educational efforts in vulnerable communities and could provide future generations with the knowledge, skills, and evidence to address environmental injustices that lead to health disparities.

NIEHS Contributions to EHL

NIEHS has played an influential role in the emergence of EHL since the early 1990s. Since then, NIEHS programs have focused on building the capacity of researchers and community members to work together to address the environmental health concerns of community residents and related concerns about environmental justice and environmental health disparities. Although not specifically stated, these programs have shared a common goal: to build and strengthen EHL. To further this goal, NIEHS included community outreach, dissemination, translation, and education cores as required components of key programs (Hursh et al. 2011). Moreover, the institute transitioned from communication to the public to communicating with the public. One-way communication strategies changed to bidirectional and multidirectional approaches, including social media and other Internet-based modalities, to ensure that all partners could contribute to a dialogue about environmental health risks (Sullivan et al. 2003).

Community-engaged research (CEnR) programs at NIEHS have demonstrated how raising EHL can also serve as a tool for empowering individuals to actively participate in efforts to address environmental exposures of local or regional concern (Adams et al. 2011; Haynes et al. 2011). An additional positive consequence for promoting EHL is raising general scientific literacy and numeracy among the public.

Over time, these community-engaged programs fostered novel partnerships (Shepard et al. 2002), taught researchers how to work collaboratively with community residents (DeLemos et al. 2007), empowered community groups to be actively involved in the conduct and dissemination of research (Minkler et al. 2010), and trained teachers how to bring environmental health concepts into the classroom (Moreno and Tharp 1999). The NIEHS experience shows that cultivating equity in community–academic partnerships enables projects to develop effective and culturally appropriate materials for local communities. Additionally, sustained support for CEnR, which includes capacity building of all partners, allows projects to address environmental health disparities in vulnerable populations, such as Latino, Native American, African American, and low-socioeconomic-status communities. These programs have all addressed essential components of an EHL model that emphasizes the importance of health literacy for public health and prevention (Freedman et al. 2009; Sørensen et al. 2012). These successes, and the continued need to raise EHL and public health awareness of risks, have kept multidirectional communication and engagement as a central goal in the NIEHS 2012–2017 Strategic Plan (NIEHS 2012).

EHL as a Research Topic

The trans-National Institutes of Health (NIH) Health Literacy program exemplifies NIH recognition of the need to explore fundamental issues in HL. For NIEHS, the focus is on validating effective ways of communicating about environmental health risks. Although the term EHL is increasingly used by investigators to denote a type of communications research, environmental health risk messaging is understudied, and relatively little is known about

  • whether there are specific stages of EHL that are amenable to intervention
  • whether raising EHL correlates with improved health outcomes
  • the relationship between EHL and resilience, for example, whether EHL increases the ability of an individual or a community to cope in challenging circumstances
  • the effectiveness of EHL resources and educational materials to inform intended audiences (within the context of their existing beliefs and attitudes)
  • different approaches for measuring success
  • the level of cultural acceptance of environmental risk messages in different ethnic and socioeconomic settings
  • the utilization and sustainability of evidence-based tools and approaches to raise EHL
  • whether risk messaging about environmental factors leads to behavior change
  • whether risk messaging leads to prevention, reduction or mitigation of environmental risk factors.

A key focus of EHL research will involve formal and rigorous assessment and validation to move from projects that produce new educational materials to projects that explore the effectiveness of educational resources. Additionally, research that explores EHL and advances the science of environmental risk messaging will require transdisciplinary or team science approaches. Environmental health scientists, individuals with expertise in community-engaged research, risk communication specialists, health educators, anthropologists, experts in dissemination and implementation science, community partners in research, and “citizen scientists” from affected communities will be critical to the success of this research.

Conclusions and Recommendations

Examine the influence of sociocultural context on EHL. When research focuses on ways to improve the EHL of individuals and communities, it will be important to understand the larger cultural context for how the public understands risks and to address misperceptions driven by media and cultural expressions. It is likely that media and films form the basis of beliefs and perception because they are widely accessed forms of communication and are often easier for the public to understand, rather than the more technical and scientific communication that investigators have historically disseminated. Effective efforts to raise EHL must therefore make risk messaging more understandable and more relevant to individuals, and they must provide not only the results of research but also address existing misinformation and misperceptions.

Develop conceptual models. As EHL evolves, measuring its stages will be beneficial. We have modified Bloom’s taxonomy to enable targeted interventions for each stage of attainment in EHL. Our model should be tested and others developed or adapted, perhaps by utilizing or extending existing instruments from related fields to accurately measure and quantify the stages of EHL. Ideally, models should account for sociocultural context and how it influences EHL, and they should acknowledge skills and empowerment at each measurable stage of EHL.

Use EHL as a tool for all partners. NIEHS embraces the evolution of EHL as an empowering component of community-engaged and environmental public health research. EHL research should include community partners in the research and provide capacity building and education at various levels of literacy for individuals and communities at risk from environmental exposures. Such education should extend beyond simply providing descriptions of specific risks to including some elucidation of the pace of science, the uncertainty principle, and the relevance of various risk measurements (e.g., ppb and levels of concern). Additionally, education and training of investigators in effective and appropriate communication modalities and creation of active partnerships with affected individuals will improve the development of culturally relevant messages. Health care professionals are another stakeholder group that could benefit from targeted education and training to enable them to recognize symptoms caused by environmental exposures and to diagnose environmentally induced diseases.

Conduct EHL research. NIEHS is committed to advancing EHL, expanding on existing efforts, and addressing gaps in knowledge and practice. This commitment could include investigations to

  • characterize the process for increasing environmental health literacy
  • develop and validate measures of EHL at both individual and community levels
  • assess the effectiveness of existing environmental risk messages
  • measure the extent of behavior change based on health risk messaging
  • create or adapt environmental risk messaging to increase the EHL of specific audiences
  • identify statistical methods or develop models that correlate the role of EHL to improving the understanding of complex risk and health outcomes.

To be most effective, this research will require a transdisciplinary or team science approach, community–academic partnerships, and sufficiently broad expertise to allow development and dissemination of targeted messaging for local communities in modalities and languages that are culturally and linguistically appropriate. Special attention could be given to improving the EHL of low-literacy and non–English-speaking individuals or that of individuals living and working in health-disparate and low-income communities. Additionally, these projects should broaden the identification of relevant stakeholders and raise the EHL of not only affected community members but also that of health care professionals, public health and lay health workers, decision makers, teachers, and students.

Coordinate federal resources. We recognize that NIEHS is only one player in the advancement of EHL and must work together with our federal partners such as the National Library of Medicine, the Centers for Disease Control and Prevention, the U.S. EPA, the National Science Foundation, and the Agency for Healthcare Research and Quality. As a coordinated group, representatives of these agencies could catalog and make available existing educational resources for the general public and for researchers working with chronically affected communities. Such a compilation of resources could provide a reliable, evidence-based source of information to the general public that may help to counteract the unsubstantiated (mis)information available on the Internet or disseminated through the media and films about environmental risks. This coordination will maximize the federal investments to date and help to ensure that research builds on previous efforts and utilizes effective tools and validated approaches developed in related fields.

Finally, the concept of EHL has emerged and is being embraced by investigators as a relevant research topic within environmental health sciences. We believe that the definitions and scope of EHL will continue to evolve and that research will help define the optimal approaches for measuring and raising EHL. Ultimately, efforts to improve EHL are intended to prevent environmentally induced disease and to empower individuals to gain a sense of control through understanding the environmental risks that affect their families and their communities.


Adams C, Brown P, Morello-Frosch R, Brody JG, Rudel R, Zota A, et al. 2011. Disentangling the exposure experience: the roles of community context and report-back of environmental exposure data. J Health Soc Behav 52:180–196.

Arcury TA, Estrada JM, Quandt SA. 2010. Overcoming language and literacy barriers in safety and health training of agricultural workers. J Agromedicine 15:236–248.

Barnes G, Fisher B, Postma J, Harnish K, Butterfield P, Hill W. 2010. Incorporating environmental health into nursing practice: a case study on indoor air quality. Pediatr Nurs 36:33–39.

Baur C. 2010. New directions in research on public health and health literacy. J Health Commun 15(suppl 2):42–50.

Benjamin R. 2010. Improving health by improving health literacy. Public Health Rep 125:784–785.

Biocca M. 2004. Risk communication and the Precautionary Principle. Int J Occup Med Environ Health 17:197–201.

Bloom BS. 1956. Taxonomy of Educational Objectives: the Classification of Educational Goals. 1st ed. New York:Longmans, Green.

Bonney R, Shirk J, Phillips T, Wiggins A, Ballard H, Miller-Rushing A, et al. 2014. Citizen science. Next steps for citizen science. Science 343:1436–1437.

Brennan VM. 2009. Natural Disasters and Public Health: Hurricanes Katrina, Rita, and Wilma. Baltimore, MD:Johns Hopkins University Press.

Brown P, Brody JG, Morello-Frosch R, Tovar J, Zota AR, Rudel RA. 2012. Measuring the success of community science: the northern California Household Exposure Study. Environ Health Perspect 120:326–331, doi: 10.1289/ehp.1103734.

Carson R, Darling L, Darling L. 1962. Silent Spring. Boston, MA:Houghton Mifflin, Riverside Press.

Chan AH, Ng AW. 2012. The guessing of mine safety signs meaning: effects of user factors and cognitive sign features. Int J Occup Saf Ergon 18:195–208.

Chinn D. 2011. Critical health literacy: a review and critical analysis. Soc Sci Med 73:60–67.

Colucci-Gray L, Camino E, Barbiero G, Gray D. 2006. From scientific literacy to sustainability literacy: an ecological framework for education. Sci Educ 90:227–252.

Conrad CC, Hilchey KG. 2011. A review of citizen science and community-based environmental monitoring: issues and opportunities. Environ Monit Assess 176:273–291.

DeLemos J, Rock T, Brugge D, Slagowski N, Manning T, Lewis J. 2007. Lessons from the Navajo: assistance with environmental data collection ensures cultural humility and data relevance. Prog Community Health Partnersh 1:321–326.

Edwards JRD, Davey J, Armstrong K. 2013. Returning to the roots of culture: a review and re-conceptualisation of safety culture. Saf Sci 55:70–80.

Estacio EV. 2013. Health literacy and community empowerment: it is more than just reading, writing and counting. J Health Psychol 18:1056–1068.

Fitzpatrick-Lewis D, Yost J, Ciliska D, Krishnaratne S. 2010. Communication about environmental health risks: a systematic review. Environ Health 9:67, doi: 10.1186/1476-069X-9-67.

Flinders DJ. 1996. Bloom’s taxonomy: a forty-year retrospective [Book review]. Hist Educ Q 36:76–78.

Frayling C. 2005. Mad, Bad and Dangerous? The Scientist and the Cinema. London:Reaktion Books.

Freedman DA, Bess KD, Tucker HA, Boyd DL, Tuchman AM, Wallston KA. 2009. Public health literacy defined. Am J Prev Med 36:446–451.

Gehle KS, Crawford JL, Hatcher MT. 2011. Integrating environmental health into medical education. Am J Prev Med 41(4 suppl 3):S296–S301.

Griswold E. 2012. How ‘Silent Spring’ ignited the environmental movement. New York Times Magazine (New York) 21 September.​ne/how-silent-spring-ignited-the-environ​mental-movement.html?_r=0) [accessed 16 December 2015].

Guidotti TL. 2013. Communication models in environmental health. J Health Commun 18:1166–1179.

Hancock HE, Rogers WA, Schroeder D, Fisk AD. 2004. Safety symbol comprehension: effects of symbol type, familiarity, and age. Hum Factors 46:183–195.

Hatfield TH. 1994. A risk communication taxonomy for environmental-health. J Environ Health 56:23–28.

Haynes EN, Beidler C, Wittberg R, Meloncon L, Parin M, Kopras EJ, et al. 2011. Developing a bidirectional academic-community partnership with an Appalachian-American community for environmental health research and risk communication. Environ Health Perspect 119:1364–1372.

Hermer J, Hunt A. 1996. Official graffiti of the everyday. L & Soc Rev 30:455–480.

Huber JT, Shapiro RM, Gillaspy ML. 2012. Top down versus bottom up: the social construction of the health literacy movement. Libr Q 82:429–451.

Hursh DW, Martina CA, Trush MA, Davis HB. 2011. Teaching Environmental Health to Children: An Interdisciplinary Approach. Dordrect, Netherlands:Springer Science and Business.

IOM (Institute of Medicine of the National Academies). 2004. Health Literacy: a Prescription to End Confusion (Nielsen-Bohlman L, Panzer AM, Kindig DA, eds). Washington, DC:National Academies Press.

IOM. 2011. Innovations in Health Literacy Research: Workshop Summary. Washington, DC:National Academies Press.

Jaspal R, Nerlich B. 2014a. Fracking in the UK press: threat dynamics in an unfolding debate. Public Underst Sci 23:348–363.

Jaspal R, Nerlich B. 2014b. When climate science became climate politics: British media representations of climate change in 1988. Public Underst Sci 23:122–141.

Kaphingst KA, Kreuter MW, Casey C, Leme L, Thompson T, Cheng MR, et al. 2012. Health Literacy INDEX: development, reliability, and validity of a new tool for evaluating the health literacy demands of health information materials. J Health Commun 17(suppl 3):203–221.

Kennedy MG, Turf EE, Wilson-Genderson M, Wells K, Huang GC, Beck V. 2011. Effects of a television drama about environmental exposure to toxic substances. Public Health Rep 126(suppl 1):150–159.

Lahr J, Kooistra L. 2010. Environmental risk mapping of pollutants: state of the art and communication aspects. Sci Total Environ 408:3899–3907.

Lee JEC, Lemyre L, Mercier P, Bouchard L, Krewski D. 2005. Beyond the hazard: the role of beliefs in health risk perception. Hum Ecol Risk Assess 11:1111–1126.

LePrevost CE, Storm JF, Blanchard MR, Asuaje CR, Cope WG. 2013. Engaging Latino farmworkers in the development of symbols to improve pesticide safety and health education and risk communication. J Immigr Minor Health 15:975–981.

Lesch MF, Rau PL, Zhao Z, Liu C. 2009. A cross-cultural comparison of perceived hazard in response to warning components and configurations: US vs. China. Appl Ergon 40:953–961.

Lin S, Gomez MI, Hwang SA, Franko EM, Bobier JK. 2004. An evaluation of the asthma intervention of the New York State Healthy Neighborhoods Program. J Asthma 41:583–595.

Matthews B, Andronaco R, Adams A. 2014. Warning signs at beaches: do they work? Saf Sci 62:312–318.

McCallum DB, Hammond SL, Covello VT. 1991. Communicating about environmental risks: how the public uses and perceives information sources. Health Educ Q 18:349–361.

Minkler M, Garcia AP, Williams J, LoPresti T, Lilly J. 2010. Sí se puede: using participatory research to promote environmental justice in a Latino community in San Diego, California. J Urban Health 87:796–812.

Minkler M, Vásquez VB, Tajik M, Petersen D. 2008. Promoting environmental justice through community-based participatory research: the role of community and partnership capacity. Health Educ Behav 35:119–137.

Mogford E, Gould L, Devoght A. 2011. Teaching critical health literacy in the US as a means to action on the social determinants of health. Health Promot Int 26:4–13.

Moore EE. 2015. Green screen or smokescreen? Hollywood’s messages about nature and the environment. Environ Commun 1–17, doi: 10.1080/17524032.2015.1014391.

Moreno NP, Tharp BZ. 1999. An interdisciplinary national program developed at Baylor to make science exciting for all K-5 students. Acad Med 74:345–347.

Murphy MW, Iqbal S, Sanchez CA, Quinlisk MP. 2010. Postdisaster health communication and information sources: the Iowa flood scenario. Disaster Med Public Health Prep 4:129–134.

Murray RL, Heumann JK. 2014. Film and Everyday Eco-disasters. Lincoln, NE:University of Nebraska Press.

Newman K. 2000. Apocalypse movies: end of the world cinema. 1st St. Martin’s Griffin ed. New York:St. Martin’s Griffin.

Nicholson PJ. 2000. Communicating occupational and environmental issues. Occup Med (Lond) 50:226–230.

NIEHS (National Institute of Environmental Health Sciences). 2012. Advancing Science, Improving Health: A Plan for Environmental Health Research. 2012-2017, Strategic Plan. NIH Publication No. 12-7935. Bethesda, MD:National Institutes of Health, U.S. Department of Health and Human Services.​plan/strategicplan2012_508.pdf [accessed 16 December 2015].

Nutbeam D. 2008. The evolving concept of health literacy. Soc Sci Med 67:2072–2078.

Nutbeam D. 2009. Defining and measuring health literacy: what can we learn from literacy studies? Int J Public Health 54:303–305.

Paasche-Orlow MK, Riekert KA, Bilderback A, Chanmugam A, Hill P, Rand CS, et al. 2005. Tailored education may reduce health literacy disparities in asthma self-management. Am J Respir Crit Care Med 172:980–986.

Perez H, Sullivan EC, Michael K, Harris R. 2012. Fish consumption and advisory awareness among the Philadelphia Asian community: a pilot study. J Environ Health 74:24–28.

Quandt SA, Doran AM, Rao P, Hoppin JA, Snively BM, Arcury TA. 2004. Reporting pesticide assessment results to farmworker families: development, implementation, and evaluation of a risk communication strategy. Environ Health Perspect 112:636–642, doi: 10.1289/ehp.6754.

Ramos IN, May M, Ramos KS. 2001. Environmental health training of promotoras in colonias along the Texas-Mexico border. Am J Public Health 91:568–570.

Severtson DJ. 2013. The influence of environmental hazard maps on risk beliefs, emotion, and health-related behavioral intentions. Res Nurs Health 36:330–348.

Shepard PM, Northridge ME, Prakash S, Stover G. 2002. Advancing environmental justice through community-based participatory research. Environ Health Perspect 110(suppl 2):139–140.

Sørensen K, Van den Broucke S, Fullam J, Doyle G, Pelikan J, Slonska Z, et al. 2012. Health literacy and public health: a systematic review and integration of definitions and models. BMC Public Health 12:80, doi: 10.1186/1471-2458-12-80.

Stokes SC, Hood DB, Zokovitch J, Close FT. 2010. Blueprint for communicating risk and preventing environmental injustice. J Health Care Poor Underserved 21:35–52.

Sullivan J, Diamond PD, Kaplan CL, Mader TJ, Santa R, Lloyd RS. 2003. Community outreach as an iterative dialogue among scientists and communities in the Texas gulf coast region. Mutat Res 544:331–338.

Sykes S, Wills J, Rowlands G, Popple K. 2013. Understanding critical health literacy: a concept analysis. BMC Public Health 13:150, doi: 10.1186/1471-2458-13-150.

Wilcox C. 2012. Guest editorial. It’s time to e-volve: taking responsibility for science communication in a digital age. Biol Bull 222:85–87.

Young SL. 1998. Connotation of hazard for signal words and their associated panels. Appl Ergon 29:101–110.

Zoller HM. 2012. Communicating health: political risk narratives in an environmental health campaign. J Appl Commun Res 40:20–43.

Biomonitoring in the Era of the Exposome

Author Affiliations open
1Department of Environmental Health, Rollins School of Public Health, Emory University, Atlanta, Georgia, USA; 2Exposure, Response, and Technology Branch, Division of Extramural Research and Training, National Institute of Environmental Health Sciences, National Institutes of Health, Department of Health and Human Services, Research Triangle Park, North Carolina, USA; 3Department of Molecular and Cell Biology, College of Liberal Arts and Sciences, University of Connecticut, Storrs, Connecticut, USA; 4Department of Chemistry, and 5Department of Medicine, Washington University, St. Louis, Missouri, USA; 6Department of Environmental Health Sciences, School of Public Health, University of California, Berkeley, Berkeley, California, USA; 7Centre for Research in Environmental Epidemiology (CREAL), Barcelona, Spain

PDF icon PDF Version (211 KB)

  • Background: The term “exposome” was coined in 2005 to underscore the importance of the environment to human health and to bring research efforts in line with those on the human genome. The ability to characterize environmental exposures through biomonitoring is key to exposome research efforts.

    Objectives: Our objectives were to describe why traditional and nontraditional (exposomic) biomonitoring are both critical in studies aiming to capture the exposome and to make recommendations on how to transition exposure research toward exposomic approaches. We describe the biomonitoring needs of exposome research and approaches and recommendations that will help fill the gaps in the current science.

    Discussion: Traditional and exposomic biomonitoring approaches have key advantages and disadvantages for assessing exposure. Exposomic approaches differ from traditional biomonitoring methods in that they can include all exposures of potential health significance, whether from endogenous or exogenous sources. Issues of sample availability and quality, identification of unknown analytes, capture of nonpersistent chemicals, integration of methods, and statistical assessment of increasingly complex data sets remain challenges that must continue to be addressed.

    Conclusions: To understand the complexity of exposures faced throughout the lifespan, both traditional and nontraditional biomonitoring methods should be used. Through hybrid approaches and the integration of emerging techniques, biomonitoring strategies can be maximized in research to define the exposome.

  • Citation: Dennis KK, Marder E, Balshaw DM, Cui Y, Lynes MA, Patti GJ, Rappaport SM, Shaughnessy DT, Vrijheid M, Barr DB. 2017. Biomonitoring in the era of the exposome. Environ Health Perspect 125:502–510;

    Address correspondence to D.B. Barr, Department of Environmental Health, Rollins School of Public Health, 1518 Clifton Rd. NE, Mailstop: 1518-002-2BB. Emory University, Atlanta, GA 30322 USA. Telephone: (404) 727-9605. E-mail:

    This manuscript is based upon the work of the Biomonitoring Working Group of the NIEHS Exposome Workshop held 14–15 January 2015 in Research Triangle Park, NC, USA.

    This work was supported by the National Institutes of Health (NIH), the National Institute of Environmental Health Sciences (NIEHS) and NIH grant P30 ES019776.

    K.K.D. and D.B.B are supported by Emory University’s Health and Exposome Research Center: Understanding Lifetime Exposures (HERCULES; grant P30 ES019776). M.A.L. has a patent that he shares with Ciencia, Inc. for a biomonitoring instrument that uses both grating-coupled surface plasmon resonance (GCSPR) imaging and grating-coupled surface plasmon–coupled emission (GCSPCE) imaging in a microarray format for the analysis of functional cell phenotyping. M.A.L. has consulted for Ciencia, Inc. in the past, but is not currently compensated as a consultant. M.A.L. also has had (and currently has) NIH/NIEHS support to develop this technology. G.J.P. is a scientific advisory board member for Cambridge Isotope Laboratories. All other authors declare they have no actual or potential competing financial interests.

    Received: 28 January 2016
    Revised: 10 May 2016
    Accepted: 21 June 2016
    Published: 6 July 2016

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact Our staff will work with you to assess and meet your accessibility needs within 3 working days.


More than ten years ago, shortly after the human genome was sequenced, Christopher Wild proposed an environmental complement to the genome in determining risk of disease, termed the exposome. He defined the exposome as the totality of exposures throughout the lifespan (Wild 2005).

Since the exposome was originally defined, research efforts have begun, leading to a revised working definition that may be summarized by the following elements. The exposome includes the cumulative measure of exposures to both chemical and nonchemical agents such as diet, stress, and sociobehavioral factors. It includes a series of quantitative and repeated metrics of exposures—both endogenous and exogenous—that describe, holistically, environmental influences or exposure over a lifetime (from conception to death). The exposome can include traditional measures of exposure (e.g., traditional biomonitoring, environmental monitoring) but also includes untargeted discovery of unknown chemicals of biological importance (Miller and Jones 2014; Rappaport and Smith 2010; Wild 2005, 2012). Exposomic approaches go a step beyond traditional biomonitoring, aiming to capture all exposures that potentially affect health and disease.

As a cancer epidemiologist, Wild understood the importance of the environment to health and that current disease trends cannot be explained by genetics alone (Wild 2005). We are only beginning to understand the complexities of environmental exposures and their impacts on human health, whereas genetic influences on health have been extensively studied. At present, we have limited estimates of the impact of environmental exposures on health, and uncertainty exists even in those (Jones 2016; Rappaport 2016; Rappaport and Smith 2010). Biomonitoring serves as a key tool to define exposure–disease risks given the biological significance of internal exposure measurements. With the continued advancement of methods, biomonitoring strategies will be critical in achieving a comprehensive understanding of exposures that have personal and public health relevance. With full understanding of the complex interactions between genetics and environmental exposures, the mysteries of the etiology, trends, and prevention of many diseases can be solved.

In an effort to advance the framework for developing exposome approaches and characterization, a diverse group of scientists gathered at the National Institute of Environmental Health Sciences (NIEHS) Exposome Workshop in January 2015 to discuss the current state of the science and to provide recommendations to the environmental health sciences community on how to best advance exposome research. The state of the science along with the perspectives and recommendations of our working group, Biomonitoring for the Exposome, are described here.


Traditional Biomonitoring

Exposure is commonly assessed by a spectrum of questionnaire data and ecological, environmental, or biological measurements. Biological measures of exposure that determine an internalized dose are often preferred because they are usually more relevant to the health outcome studied. Traditional biological measurements, also called targeted analyses, measure a target chemical, metabolite, or reaction product in a biological medium such as urine or blood (see Appendix 1). These traditional biomonitoring measurements have become a key component of exposure assessment in many epidemiologic studies that attempt to link exposures to health outcomes.

Molecular epidemiology studies and regulatory agencies rely primarily on targeted analyses because of their current availability and historical use. Broad surveys such as the National Health and Nutrition Examination Study (NHANES) utilize these methods, allowing for quantification and longitudinal surveillance of known exposures across the U.S. population. NHANES data facilitate comparative identification of abnormal exposure levels in select population subsets. Major epidemiology studies such as those evaluating blood lead levels and mean IQ in children and prenatal pesticide exposures and neurological deficits in children and neurodegenerative disease in adults have linked significant health outcomes to specific exposures, informing opportunities for further mechanistic studies (Chin-Chan et al. 2015; Kaufman et al. 2014; Rosas and Eskenazi 2008). Other federal efforts in the United States include the National Biomonitoring Program (NBP) of the Division of Laboratory Sciences at the Centers for Disease Control and Prevention (CDC). The NBP produces a National Report on Human Exposure to Environmental Chemicals and regularly updates the NHANES biomonitoring data in that report (CDC 2009, 2015). Chemicals of potential concern such as arsenic, perchlorate, and environmental phenols, among others, continue to be added to NHANES, with the most recent report including data on > 250 chemicals. The CDC also provides grant funding to a variety of state laboratories to increase public health laboratory capacity for surveillance. Targeted analytical capabilities and worldwide use continue to expand through both public health and academic entities.

Historical use of biomonitoring. Traditional biomonitoring methods are well established for exposure assessment in epidemiology studies and in federal and state surveillance activities. Because of their historical use, they provide a number of strong advantages for exposure research (see Appendix 2). Biologically persistent chemicals are well-characterized with traditional methods, whereas short-lived chemicals are effectively measured only if the individual is undergoing continuous or continual exposures or if the timing of exposures is known. Chemicals such as phthalates, bisphenols, and parabens are well-characterized by targeted methods given their widespread use and presence in the environment. Often, chemicals of particular toxicological interest may be difficult to measure owing to barriers such as stability or presence in readily accessible biological matrices such as blood or urine. For example, short-lived chemicals such as various current-use pesticide and phthalate metabolites can only be detected in urine samples if exposure occurs within a few days of testing; therefore, continuous or longitudinal sample collection is necessary to capture exposure. For a selected group of 250–300 known persistent (~30–40%) and nonpersistent (~60–70%) chemicals, sample analysis provides exposure information for the chemical of concern within a specific window of exposure; reference data are available for most of these chemicals (CDC 2015).

Appendix 2. Key advantages and disadvantages of traditional biomonitoring for determination of exposure.

Appendix 2. Key advantages and disadvantages of traditional biomonitoring for determination of exposure.

View Table (HTML Version)
View larger image (TIF File)

The ~250 chemicals that are commonly measured in the United States are primarily driven by the CDC biomonitoring list of target analytes (CDC 2015). Most other programs also follow the CDC list because selection of these agents was informed by a public nomination process followed by expert ranking of the nominated chemicals (CDC 2012). An important caveat of this process is the target list is partially based on ease of performance and compatibility with existing methods. Another concern is that some of the chemicals have little toxicological relevance or diminishing exposure across the population resulting from successful regulation of their release into the environment, or a combination of the two.

Biomonitoring methods. Although method development for traditional biomonitoring can be quite rigorous, this also translates into a slow and expensive process when developing analysis protocols for new chemicals of interest. These analyses often require relatively high volumes of sample, typically 0.5–1 mL for a single method (~ 10 mL urine and > 20 mL serum to measure the 250–300 currently biomonitored chemicals), which can be limiting for certain biospecimen types and age groups under study. For exposome research, these requirements restrict the number and types of chemicals that can be measured at any one time. Unknown or suspected chemicals of concern may not be measured or identifiable through targeted methods (see Appendix 2) (Rappaport et al. 2014); yet targeted analyses are valuable given the accuracy and depth at which a chemical of interest can be assessed. By coupling traditional biomonitoring methods with broader exposomic approaches, the benefits of both strategies can be fully realized.

Exposomic Approaches

An exposomic approach differs from traditional biomonitoring in that it can theoretically include all exposures of potential health significance, whether they are derived from exogenous sources (e.g., pollutants, diet, drugs) or endogenous sources (e.g., hormones, human and microbial metabolites) (Rappaport and Smith 2010; Rappaport et al. 2014). Because levels of chemicals in blood or other biospecimens reflect a wide range of exposures or the metabolic consequences of exposures, including psychosocial stress, other nonchemical stressors such as noise, and nutritional factors, exposomic biomonitoring offers an efficient means for characterizing individual exposure profiles. Incorporating the exposome paradigm into traditional biomonitoring approaches offers a means to improve exposure assessment in many ways (Wild 2012).

Untargeted analyses. With only a few hundred chemicals routinely assessable through targeted methods and with limitations for short-lived compounds, exposomic approaches are critical to understanding the thousands of chemicals people are exposed to daily through direct chemical exposures or consequences of exposure (e.g., cortisol levels due to stress or noise exposures) (CDC 2015). Through untargeted biomonitoring approaches such as high-resolution metabolomics (HRM), > 1,500 metabolites can be monitored with a relatively small amount of biological specimen (≤ 100 μL) and for the cost of a single traditional biomonitoring analysis of 8–10 target chemicals (Johnson et al. 2010; Jones 2016).

Untargeted analyses of small molecules or macromolecular adducts in blood, urine, or other matrices are well suited for exposome-wide association studies (EWAS), which compare profiles of hundreds or thousands of chemical features—analogous to ions with a given mass-to-charge ratio and a specified retention time in traditional biomonitoring —between diseased and healthy subjects (Rappaport 2012, 2016). Indeed, untargeted analyses performed using the current generation of liquid chromatography–high resolution mass spectrometers (LC-HRMSs) can detect > 30,000 small-molecule features (Ivanisevic et al. 2013) and > 100 human serum albumin (HSA) adducts of reactive electrophilic chemicals (including reactive oxygen species) at the nucleophilic locus Cys34 (Grigoryan et al. 2012; Rappaport et al. 2012). Processing the rich sets of data from untargeted analyses of archived biospecimens offers a path for discovering health-impairing exposures that have thus far escaped scrutiny, a largely unrecognized benefit of exposomics. It is important to note that full annotation of molecular features is not required for case–control comparisons provided that LC-HRMS signatures are available (e.g., accurate mass, retention time, and MS/MS fragmentation). Archived biospecimens from well-designed cohort studies already exist. With continued advancement in untargeted analyses, there is potential to make significant advances in human health through uncovering unknown exposures (da Silva et al. 2015; Zhou et al. 2012).

High-resolution metabolomics. Although untargeted analyses encompass a wide range of the -omics techniques, HRM is a technique that is poised to advance exposomics research because of the breadth of coverage it offers of both endogenous and exogenous chemicals. At the present time, it is routine to detect tens of thousands of features with HRM, and this number will increase as the sensitivity of mass analyzers continues to improve. These features do not necessarily represent different chemical constituents but provide extensive data for evaluation of alterations in biological pathways (Mahieu et al. 2014). Extensive comparisons of the features of these various instruments are available elsewhere (Marshall and Hendrickson 2008). With the additional advancements that have been made in bioinformatics methods to aid in feature extraction and data analysis, HRM has become an increasingly viable tool for broad exposome-level characterization (Jones 2016). Although features linked to human health will require chemical identification, the technology is in place for the feature extraction methods and annotation efforts that will increase the total number of chemicals that can be monitored (Soltow et al. 2013). Researchers are already demonstrating this expanded potential along with the ability to quantify chemicals under a high-resolution metabolomics platform (Go et al. 2015; Li et al. 2015). By definition, untargeted approaches are agnostic, allowing unknown or emerging exposures of concern (see Appendix 3) to be detected. These approaches are often hypothesis-generating and may require testing of newly discovered analytes/exposures in experimental models to confirm effects on biological responses.

Appendix 3. Key advantages and disadvantages of exposomic approaches for determination of exposure.

Appendix 3. Key advantages and disadvantages of exposomic approaches for determination of exposure.

View Table (HTML Version)
View larger image (TIF File)

Detection of low-level xenobiotic exposures. Persistent challenges exist with detecting chemicals present at low levels, defining reference values of “normal” exposure, and ultimately linking these measures to an exogenous source so intervention can occur. Because blood concentrations of xenobiotics (femtomolar to micromolar) tend to be much lower than those of chemicals derived from food, drugs, and endogenous sources (nanomolar to micromolar), untargeted analyses of xenobiotics are not as efficient and reliable as those of ingested and endogenous sources at detecting many exposures of interest (Rappaport et al. 2014). To determine the health impacts of these exposures, it will be necessary to develop semi-targeted or multiplexed methods that increase the signals of exogenous molecules relative to those of endogenous origin (Rappaport et al. 2014; Southam et al. 2014; Wei et al. 2010). Analyses of suspected chemicals of concern, also referred to as suspect screening, can be prioritized through measuring panels of chemicals with known biological effects but without identifying a specific hypothesis regarding the toxicological pathway. Untargeted and suspected chemical analyses both fall under exposomic biomonitoring and offer extraordinary potential for increased understanding of complex chemical exposures.

Hybrid approaches. Various terms are used to describe hybrid approaches, including suspect screening or semi-targeted analyses. Because both targeted and untargeted approaches have beneficial attributes as well as drawbacks, using a hybrid exposomics approach may enable us to exploit advantages while minimizing the limitations of each technique. One of the obvious limitations of a targeted approach is its inability to provide exposure information on a wide array of chemicals. However, targeted analysis can typically provide validated and quality-assured detection and quantification at very low concentrations that may not be available using an untargeted approach until HRM and the necessary bioinformatic data extraction techniques mature. As mentioned above, the development of these quantitative techniques for HRM is underway with advancements in instrumentation (Go et al. 2015; Marshall and Hendrickson 2008). Furthermore, the generic extraction methods used in untargeted analysis may not be able to capture all of the chemicals of interest (e.g., limited extraction of nonpolar chemicals using a typically polar solvent extraction), whereas more specialized extractions can specifically target chemical classes.

Semi-targeted analysis. Semi-targeted analysis can utilize various approaches including a two-step strategy: discovery using metabolomics followed by a more fully quantitative targeted measure. Another potential approach would involve a known or measured chemical exposure in individuals for which metabolomic measurements could also be made. For instance, untargeted metabolomic analysis of each group would then allow for a search for new exposure biomarkers and unique metabolic pathway pertubations to help elucidate the effect mechanism.

Traditionally, targeted analysis data have been used for risk assessment purposes, so shifting solely to a newer platform may take some time. The hybrid approach can be useful for both exposomic analysis and informing targeted analysis approaches. For example, a targeted chemical concentration can be used as an “outcome” for metabolome-wide association studies (e.g., evaluating biochemical alterations relative to targeted chemical concentrations), or a metabolomic analysis can help identify important chemicals that need to be rigorously quantified for health or risk assessments. Of course, each of the two approaches stands on its own, and they have done so for many decades. By combining the two, however, we have a much more powerful approach to understanding chemical exposures, biological alterations, and disease.

Overarching Issues

Matrix selection. Whether using a traditional biomonitoring or an exposomic approach, careful attention must be given to which matrices can be practically collected and which matrices are relevant for assessing chemical exposures. The matrices available for collection during different life stages and a nonexhaustive list of the chemicals that are appreciably present in these matrices have been reviewed elsewhere (Barr et al. 2005). Typically, the least-invasive matrix in which the chemicals appreciably collect, such as blood and urine, is the preferred matrix.

Although most analysis of exposure is performed with urine or blood samples as a consequence of the ease with which they can be collected, there are other sample types that have begun to be explored for their value in exposome interrogation. For example, saliva, which can be collected from school-age children and adults, is a problematic matrix to collect from infants and toddlers because of choking dangers associated with the collection devices and the inability of young children to actively secrete it. Even if the matrix, in this case saliva, can be noninvasively collected, the target chemical or suite of potential chemicals may not enter the matrix for a variety of reasons, including protein binding of chemicals that prevents their secretion into saliva (Lu et al. 1998). In addition, saliva is nonsterile; therefore, contributions of the oral microbiome can influence the composition of the analytes to be measured. Buccal and nasal swabs have also been used to assess the biological consequences of external exposures. In those sample types, DNA, mRNA, and their adducts have been the principal focus to date (Beane et al. 2011; Spira et al. 2004; Zhang et al. 2010), but these samples (as well as fecal samples) are also compromised by the presence of a strong microbial community that can influence the composition of the exposome constituents.

Other biological samples (e.g., selected blood cells, sweat, teeth, nails) can include information about recent historical exposures in their composition. The use of alternative samples as historical measures of exposure may become important in future studies. Teeth are a matrix that has demonstrated particular promise for characterizing prenatal exposures to metals and to some organic chemicals because of their defined growth patterns (Andra et al. 2015). We can use the “record” of prior exposures recorded in hair, deciduous teeth, or molecular “fingerprints” in other samples to provide historical measures of certain exposures (Arora et al. 2012; Hu et al. 2007); however, validation of the time represented in the exposure history may be laborious.

There are limitations to these sample sets because external deposits of specific chemicals can make the interpretation of measured levels in these samples different from those measured in blood, for example. In addition, standardized protocols and reference standards are lacking for many alternative matrices, making standardization of results across studies difficult.

An important consideration when choosing samples for exposome-type research is the anticipated presence of the particular chemical(s) in the harvested samples. Because chemicals may display unanticipated pharmacodynamics and biotransformation, it may ultimately be essential that multiple sample types are collected from each individual in the effort to define the exposome. Blood circulates throughout the body, so there is an advantage to its assessment because it has been exposed to the myriad of routes by which an environmental chemical may enter the body. However, some analytes are known to specifically accumulate in particular tissues; thus, a broad-spectrum assessment of multiple patient samples will provide the best insights into exposures.

Analytical considerations for matrix effects. In addition to the relevant matrices that can be collected, we must consider the alterations in response that may be obtained in analytic systems related to other components of the matrix. Such matrix effects can enhance analytic signals or work to suppress signals (Panuwet et al. 2016). In fact, each individual sample will exert its own matrix effects that can make quantification difficult, particularly in mass spectrometry–based methods. Mass spectrometers are inherently sensitive to matrix effects such that the analytical signal of a given concentration can vary over orders of magnitude if appropriate internal standards for normalizing the mass spectral signal have not been used (Baker et al. 2005). In particular, these variations could present challenges when attempting to quantify features in untargeted analysis approaches.

Sample collection and storage. Collection and storage procedures are particularly important considerations for internal exposure measurements. Failures in the proper collection and storage of specimens can result in lost sample integrity, samples that are not suitable for analysis, and contamination/degradation of important chemicals. Because of the sensitivity of some methods such as HRM, biospecimens must be carefully collected and well-maintained. Specific attention to freeze-thaw cycles, potential contamination risks, and collection protocols is needed to ensure that the data extracted from each sample are accurate. It is nearly impossible to control for every preanalytic challenge in sample collection and storage for an untargeted analysis, which is one reason that both targeted and untargeted analyses are quite complementary. In addition, both targeted and untargeted approaches can only measure a limited amount of the exogenous and endogenous chemicals that are present in our bodies. The types and number of chemicals within us that are measureable largely depend upon the matrix selected and the method used [Children’s Health Exposure Analysis Resource (CHEAR) 2016a, 2016b].

Variability of exposures. Temporal variability. Temporal, spatial, and genetic variability and variability in biological distribution of chemicals are important elements to characterize in internal exposure studies. It is important to understand if a single sample obtained at a given life stage represents average exposure over time [e.g., a blood sample for dichlorodiphenyldichloroethylene (DDE) measurements during adulthood obtained during a time of much physiologic change, such as pregnancy], or if peak exposures during a critical window are more important to consider. For short-lived chemicals, new technologies and approaches that facilitate collection of real-time data, high-dimensional analyses, and uncovering biological response markers of transient exposures offer strategies for capturing historically difficult measurements (Dennis et al. 2016).

Spatial variability. In addition, it is important to understand how temporal variability may vary over geographic areas and in different exposure scenarios. In this respect, exposure assessment can become very complex. Multiple samples within a population are generally preferred over a single sample so that both temporal and spatial variability can be assessed; however, the collection of multiple samples is often cost-prohibitive and can be an undue burden on participants. To appropriately interpret internal exposure data in the context of risk or health outcome, it is imperative to ascertain the degree of variability in space and time.

Pharmacodynamic variability. Ideally, we would have information on variability in pharmacodynamics to potentially evaluate resulting exposure data (e.g., whether a given chemical distributes to tissues differently among individuals). Most of the pharmacodynamic information we have for specific chemicals is derived from animal studies, and these processes may not be the same in humans. In addition to exposure and pharmacokinetic variability, laboratory and sampling variability should also be assessed and, if possible, teased apart from true intra-person variability.

Fit-for-purpose use. The “fit-for-purpose” concept has gained popularity in traditional biomonitoring (Lee et al. 2006). This concept addresses the balance between overall cost of analysis and the degree of analytical rigor required to use the internal exposure measure results for a given purpose. In instances where legal implications exist or regulatory decisions are to be made, maximum analytical rigor is required. But for exploratory studies and for many epidemiologic studies, statistical power derived from a larger number of samples, but with sufficient precision to detect differences, is often preferred. In these cases, relaxation of analytical rigor may translate into lower costs that, in turn, could enable the number of samples analyzed to increase. Furthermore, in untargeted approaches, authentic standards are not always necessary to evaluate a chemical’s relationship to disease or alterations in biomolecular concentrations. In addition, many “add-on” studies use samples collected for different analyses for which the sample collection/storage may represent more imprecision, thus not warranting the increased cost of strict analytical rigor.

For each given study or study question, it is important to consider the analysis and the criteria that are necessary to meet the study objectives. For example, if a study seeks to control for smoking but needs validation of the questionnaire, a low-resolution method such as an immunoassay for molecular indications of smoking may be most suitable for the study; this would maximize the money available for other needs in the study. Many times, substantial resources are dedicated to perfecting an analytic method rather than using a portion of those funds to determine which measurements are actually critical to answering a research question. The issue of balance in analytic rigor and cost needs to be addressed in each study.

Extant data also represent a “fit-for-purpose” approach. Extant data were often collected to answer a certain set of research questions; thus, they are not always applicable to a different study question. However, extant data do represent a source for generating hypotheses that can be further tested using prospective, longitudinal studies. For example, NHANES data offer a resource to evaluate the extent of U.S. population exposures to particular chemicals and can serve as a tool for the exposure component of risk assessment. Although the data are cross-sectional, they serve as a useful hypothesis-generating resource.

Unknown analytes. Characterizing unknown analytes remains a major challenge for understanding the exposome. Research efforts should prioritize the development of methods to determine relevant exposures and to identify sources of specific chemical signatures. By linking shifts in the microbiome, the metabolome, the proteome, and so forth to unknown analytes, we can start to determine the profiles of unknown toxicant exposures and their consequences. Additionally, biomonitoring techniques that can assess changes in cellular composition or in the developmental capacity of cells may indicate risks for later health conditions such as cancer and neurodegenerative diseases. Even if the identity of an analyte is unknown, linking unknown exposures to potential disease consequences creates further support for the investment of resources necessary to understanding cumulative lifetime exposures.

Annotation of spectra for unknown chemicals can be quite time-consuming and therefore only completed on a select number of features. Limitations regarding chemical annotation will best be overcome through a concerted effort across many research groups to identify, catalogue, and disseminate information related to newly identified small molecules. Additionally, continued focus on bioinformatics techniques to extract information about chemical features of importance will allow semi-targeted approaches to be used for unknown and low-abundance chemicals.

The omics technologies all have potential for discovering unknown analytes. Through ongoing advancements in mass spectrometry, low-abundance chemicals can be targeted and characterized. With comprehensive coverage of the metabolome, reference metabolic profiles combined with health outcome data would provide a baseline for identification of unknown analytes with health relevance. Through a concerted effort across laboratories, identification and cataloguing unknown analytes will become a tangible task for advancing the exposome.

Overcoming Gaps and Barriers to Exposome Research

Several data gaps or barriers exist in both targeted and untargeted analyses. For untargeted analyses, the ability to identify and quantify low-abundance analytes—most environmental chemicals—is still immature. Untargeted approaches may need new, more sensitive mass spectrometric approaches or chemoselective probes to improve the detection of low-abundance chemicals. We reemphasize that analytic standards are not required for discovery of new and relevant biomarkers; they become necessary only when a new biomarker is identified and needs to be validated.

Although many biomonitoring resources are available through public health and academic laboratories, few laboratories exist with the capacity to measure a wide array of “known” toxicants, particularly in nonstandard matrices (i.e., matrices other than blood and urine) (see Appendix 4). Having access to such capacity is particularly important for new investigators, who may not have established relationships with such laboratories. Additionally, accurate and reproducible measures across laboratories remain a challenge. The CHEAR initiative, led by the National Institute of Environmental Health Sciences, represents a unique opportunity to provide a standardized laboratory network with access to targeted and untargeted analyses of biospecimens and so may serve to fill these gaps (NIEHS 2015).

Databases. The application of untargeted metabolomics to identify environmental exposures correlated with human health has its own unique challenges. The largest reference databases for metabolomics are the Metabolite and Tandem MS Database (METLIN) and the Human Metabolome Database (HMDB) (Tautenhahn et al. 2012; Wishart et al. 2009). To date, METLIN and HMDB have largely focused on naturally occurring metabolites. To our knowledge, the number of compounds in METLIN and HMDB that may be potentially relevant to exposure studies has not yet been carefully assessed. The number of databases available for metabolomics continues to expand and has unique utility depending on the research question. A more expansive discussion of metabolomics database resources is available (Go 2010). To facilitate large-scale exposomic studies, the field may benefit from having a database or from having database search functionalities specifically dedicated to environmental exposure chemicals. As discussed above, discovery experiments are typically most successful when a small subset of features can be targeted for structural identification. Thus, databases and repositories curating information on the human exposome would provide powerful mechanisms for prioritizing features of interest to environmental health scientists.

Bioinformatic Approaches

Although bioinformatics were covered under the scope of the Biostatistics and Informatics Workgroup at the NIEHS Exposome Workshop, it is worthwhile to mention a few bioinformatic needs that are specific to the development of exposomic biomonitoring approaches. As highlighted throughout this article, characterizing the complexities of the exposome requires use of broad coverage techniques to link internal biochemical perturbations to external exposures. Bioinformatic requirements for these types of data analyses are substantial, yet they offer a high return on investment. Through pathway analysis and data extraction algorithms, biological pathway perturbations can provide great insight into broad disease processes. Additionally, detection of low-level xenobiotic and unknown chemicals can be greatly enhanced through bioinformatic techniques. Further development of bioinformatic tools and data storage and handling will be key to advancing our understanding of the health impact of complex exposures.

Implementing the Exposome

External exposures and the actual body burden of said exposures can be quite variable. There is much to be learned about combining external and internal measures to maximize understanding of exposure and how to mitigate exposures that have negative health consequences. Coupling technologies and utilizing real-time monitoring tools can increase our overall understanding of exposures spatially and temporally. Exposome studies in Europe such as The Human Early-life Exposome (HELIX); Health and Environment-wide Association Studies based on Large population Surveys (HEALS); and EXPOsOMICS have started to demonstrate specific approaches for capturing this type of information [Community Research and Development Information Service (CORDIS) 2014, 2015; Vrijheid et al. 2014].

Similarly, Emory University’s NIEHS-funded Human Exposome Research Center: Understanding Lifetime Exposures (HERCULES) has developed infrastructure that has supported several environmental health studies using hybrid biomonitoring approaches (Go et al. 2014, 2015; Jones 2016; Zhang et al. 2014). HELIX also uses a hybrid approach for data collection. HELIX specifically focuses on cohorts of mother-child pairs to better understand which developmental periods may be particularly vulnerable to environmental exposures (Vrijheid et al. 2014). Along with personal external exposure monitoring strategies, traditional biomonitoring techniques have been combined with untargeted omics analyses (e.g., metabolomics, proteomics, transcriptomics, epigenomics) with a particular focus on repeat sampling to capture nonpersistent biomarkers. By performing omics–exposure and omics–health association studies, researchers aim to uncover biologically meaningful omics signatures. The HELIX design is one example of a current approach that integrates traditional and nontraditional techniques to better understand the exposome. Although HELIX offers one initial study structure for understanding the exposome, continued emphasis for exposomic approaches should be placed on developing techniques to measure nonpersistent chemicals that do not place undue burdens on study participants or significant financial constraints on the research study.


The following recommendations are suggested for approaching internal exposure assessment for exposome research:

Recommendation 1: Encourage secondary analyses of samples collected for traditional targeted chemical studies. High-quality samples (i.e., samples that have been collected and stored properly) from longitudinal epidemiology studies should be used for untargeted analysis and alternative measurement techniques. For this aim to be successful, it is critical that methods for sample collection and storage be standardized. Investment should be made in maintaining established cohorts and in developing protocols that optimize stabilization of samples for storage (e.g., does one analyte stabilizer actually destabilize other analytes of interest? Would adding a known xenobiotic act as a standard for normalization? Should multiple small aliquots be stored at the time of collection to facilitate different analytical needs?).

Recommendation 2: Evaluate and use standardized measurement platforms with measurement harmonization. A general prototype platform or reference samples should be established under which different technologies can be tested. By establishing this platform, researchers can have a standardized way of demonstrating capacity with new approaches, which would allow efficient integration of effective methods into research protocols. One approach would be to use samples from NHANES or from a similarly well-characterized data set as a “challenge” or “quality control” set for new and emerging technologies. Moreover, development of or participation in multi-lab proficiency testing programs will ensure harmonization of data across studies.

Recommendation 3: Use existing resources and databases to obtain information on current exposures that may be important. Significant efforts have been made to expand databases such as the HMDB, Kyoto Encyclopedia of Genes and Genomes (KEGG) human metabolic pathways, and METLIN databases (Kanehisa and Goto 2000; Kanehisa 2002; Smith et al. 2005; Wishart et al. 2009, 2013). Mining these well-developed resources in conjunction with new data analyses will enable a more comprehensive exposure characterization.

Recommendation 4: Provide guidance for the use of existing databases and develop tools to allow searches across multiple databases. To facilitate researchers’ integrating exposomic approaches into their studies, resources regarding existing databases should be streamlined. Integration of existing databases such as the HMDB, LIPID MAPS Structure Database and METLIN or search options that can readily work across these resources would enhance their utility for exposome research (LIPID MAPS 2015; Smith et al. 2005; Wishart et al. 2009, 2013).

Recommendation 5: Foster and facilitate discussion with people from different disciplines to discuss the reality of targeted and untargeted analytic capabilities. Discussions should focus around the development of semi-targeted or multiplexing strategies (Wei et al. 2010). Specific discussions should emphasize approaches for capturing short-lived chemicals while minimizing undue financial and participant burdens. Through generating discussion regarding established methods, researchers can have a structured dialogue concerning the utility of targeted, untargeted, and hybrid methods.

Recommendation 6: Develop chemistry methods to enable the detection of low-abundance chemicals and to enable differentiation of endogenous molecules from exogenous molecules. Through methods such as multiplexing, interfering chemicals can be removed to allow detection of low-level environmental chemicals that are often difficult to detect because of higher-abundance endogenous chemicals from food, drugs, and normal metabolic processes (Rappaport et al. 2014). Investments in the development of semi-targeting or multiplexing strategies should be a high priority.

Recommendation 7: Develop bioinformatics techniques to enhance detection of unknown chemicals using untargeted methods. With continued efforts such as ExpoCast, untargeted analysis can be combined with advanced bioinformatic techniques to help prioritize risk assessment, to determine which exposures often co-occur, and to establish markers of disease risk (Dennis et al. 2016; Johnson et al. 2015; Rager et al. 2016; Yu et al. 2013; Wambaugh et al. 2013).

Recommendation 8: Encourage development of pharmacokinetic models. Through building simulated human response models, researchers would be able to incorporate kinetic and dynamic variability to inform interpretation of biomonitoring data.


Measurable long-term improvements to human health are attainable through working towards a holistic understanding of environmental influences. In order to assess the exposome, traditional biomonitoring should be coupled with untargeted discovery of unknown chemicals of biological importance. It is critical to note that the advances described here, including those still in early stages of development, require commitment of scientific resources and energy to bring such approaches to fruition. Continued discussion and integration of approaches will be necessary to address the inherent complexity of the exposome. Broad characterization and understanding of internal exposures and their consequences are achievable under the exposome paradigm through combining emerging technologies and untargeted approaches with traditional biomonitoring techniques.


Andra SS, Austin C, Arora M. 2015. Tooth matrix analysis for biomonitoring of organic chemical exposure: current status, challenges, and opportunities. Environ Res 142:387–406, doi: 10.1016/j.envres.2015.06.032.

Arora M, Bradman A, Austin C, Vedar M, Holland N, Eskenazi B, et al. 2012. Determining fetal manganese exposure from mantle dentine of deciduous teeth. Environ Sci Technol 46:5118–5125, doi: 10.1021/es203569f.

Baker SE, Olsson AO, Needham LL, Barr DB. 2005. High-performance liquid chromatography–tandem mass spectrometry method for quantifying sulfonylurea herbicides in human urine: reconsidering the validation process. Anal Bioanal Chem 383:963–976.

Barr DB, Wang RY, Needham LL. 2005. Biologic monitoring of exposure to environmental chemicals throughout the life stages: requirements and issues for consideration for the National Children’s Study. Environ Health Perspect 113:1083–1091, doi: 10.1289/ehp.7617.

Beane J, Vick J, Schembri F, Anderlind C, Gower A, Campbell J, et al. 2011. Characterizing the impact of smoking and lung cancer on the airway transcriptome using RNA-Seq. Cancer Prev Res (Phila) 4:803–817, doi: 10.1158/1940–6207.

CDC (Centers for Disease Control and Prevention). 2009. Fourth National Report on Human Exposure to Environmental Chemicals. Department of Health and Human Services, CDC.​urthReport.pdf [accessed 24 August 2015].

CDC. 2012. Chemical Selection, National Report on Human Exposure to Environmental Chemicals, Chemical Selection.​al_selection.html [accessed 7 October 2015].

CDC. 2015. Fourth National Report on Human Exposure to Environmental Chemicals Updated Tables. Department of Health and Human Services, CDC.​rthReport_UpdatedTables_Feb2015.pdf [accessed 1 September 2015].

CHEAR (Children’s Health Exposure Analysis Resource). 2016a. Lab Analysis Services. Targeted Analysis.​ysis-services/targeted [accessed 30 April 2016].

CHEAR. 2016b. Lab Analysis Services. Untargeted Analysis. Metabolomics.​ysis-services/untargeted/metabolomics [accessed 30 April 2016].

Chin-Chan M, Navarro-Yepes J, Quintanilla-Vega B. 2015. Environmental pollutants as risk factors for neurodegenerative disorders: Alzheimer and Parkinson diseases. Front Cell Neurosci 9:124, doi: 10.3389/fncel.2015.00124.

CORDIS (Community Research and Development Information Service. 2014. EXPOSOMICS (Enhanced Exposure Assessment and Omic Profiling for High Priority Environmental Exposures in Europe.).​3_en.html [accessed 14 April 2016].

CORDIS. 2015. Periodic Report Summary 1 – HEALS (Health and Environment-wide Associations based on Large population Surveys).​8_en.html [accessed 14 April 2016].

da Silva RR, Dorrestein PC, Quinn RA. 2015. Illuminating the dark matter in metabolomics. Proc Natl Acad Sci U S A 112(41):12549–12550, doi: 10.1073/pnas.1516878112.

Dennis KK, Auerbach SS, Balshaw DM, Cui Y, Fallin MD, Smith MT, et al. 2016. The importance of the biological impact of exposure to the concept of the exposome. Environ Health Perspect 124:1504–1510, doi: 10.1289/EHP140.

Go EP. 2010. Database resources in metabolomics: an overview. J Neuroimmune Pharmacol 5(1):18–30, doi: 10.1007/s11481-009-9157-3.

Go YM, Uppal K, Walker DI, Tran V, Dury L, Strobel FH, et al. 2014. Mitochondrial metabolomics using high-resolution Fourier-transform mass spectrometry. Methods Mol Biol 1198:43–73, doi: 10.1007/978-1-4939-1258-2_4.

Go YM, Walker DI, Liang Y, Uppal K, Soltow QA, Tran V, et al. 2015. Reference standardization for mass spectrometry and high-resolution metabolomics applications to exposome research. Toxicol Sci 148(2):531–543, doi: 10.1093/toxsci/kfv198.

Grigoryan H, Li H, Iavarone AT, Williams ER, Rappaport SM. 2012. Cys34 adducts of reactive oxygen species in human serum albumin. Chem Res Toxicol 25:1633–1642, doi: 10.1021/tx300096a.

Hu H, Shih R, Rothenberg S, Schwartz BS. 2007. The epidemiology of lead toxicity in adults: measuring dose and consideration of other methodologic issues. Environ Health Perspect 115:455–462, doi: 10.1289/ehp.9783.

Ivanisevic J, Zhu ZJ, Plate L, Tautenhahn R, Chen S, O’Brien PJ, et al. 2013. Toward ‘omic scale metabolite profiling: a dual separation–mass spectrometry approach for coverage of lipid and central carbon metabolism. Anal Chem 85:6876–6884, doi: 10.1021/ac401140h.

Johnson CH, Ivanisevic J, Benton HP, Siuzdak G. 2015. Bioinformatics: the next frontier of metabolomics. Anal Chem 87(1):147–156, doi: 10.1021/ac5040693.

Johnson JM, Yu T, Strobel FH, Jones DP. 2010. A practical approach to detect unique metabolic patterns for personalized medicine. Analyst 135(11):2864–2870, doi: 10.1039/c0an00333f.

Jones DP. 2016. Sequencing the exposome: a call to action. Toxicol Rep 3:29–45.

Kanehisa M. 2002. The KEGG database. Novartis Found Symp 247:91–101.

Kanehisa M, Goto S. 2000. KEGG: Kyoto Encyclopedia of Genes and Genomes. Nucleic Acids Res 28:27–30.

Kaufman AS, Zhou X, Reynolds MR, Kaufman NL, Green GP, Weiss LG. 2014. The possible societal impact of the decrease in U.S. blood lead levels on adult IQ. Environ Res 132:413–420, doi: 10.1016/j.envres.2014.04.015.

Lee JW, Devanarayan V, Barrett YC, Weiner R, Allinson J, Fountain S, et al. 2006. Fit-for-purpose method development and validation for successful biomarker measurement. Pharm Res 23(2):312–328.

Li S, Todor A, Luo R. 2015. Blood transcriptomics and metabolomics for personalized medicine. Comput Struct Biotechnol J 14:1–7, doi: 10.1016/j.csbj.2015.10.005.

LIPID MAPS (LIPID Metabolites and Pathways Strategy). 2015. Lipodomics Gateway. Resources Overview.​rces.html [accessed 24 August 2015].

Lu C, Anderson LC, Morgan MS, Fenske RA. 1998. Salivary concentrations of atrazine reflect atrazine plasma levels in rats. J Toxicol Environ Health A 53:283–292.

Mahieu NG, Huang X, Chen YJ, Patti GJ. 2014. Credentialing features: a platform to benchmark and optimize untargeted metabolomic methods. Anal Chem 86:9583–9589, doi: 10.1021/ac503092d.

Marshall AG, Hendrickson CL. 2008. High resolution mass spectrometers. Annu Rev Anal Chem (Palo Alto Calif) 1:579–599, doi: 10.1146/annurev.anchem.1.031207.112945.

Miller GW, Jones DP. 2014. The nature of nurture: refining the definition of the exposome. Toxicol Sci 137(1):1–2, doi: 10.1093/toxsci/kft251.

NIEHS (National Institute of Environmental Health Sciences). 2015. Children’s Health Exposure Analysis Resource (CHEAR).​rted/dert/programs/chear/index.cfm [accessed 1 September 2015].

Panuwet P, Hunter RE Jr, D’Souza PE, Chen X, Radford SA, Cohen JR, et al. 2016. Biological matrix effects in quantitative tandem mass spectrometry-based analytical methods: advancing biomonitoring. Crit Rev Anal Chem 46(2):93–105, doi: 10.1080/10408347.2014.980775.

Rager JE, Strynar MJ, Liang S, McMahen RL, Richard AM, Grulke CM, et al. 2016. Linking high-resolution mass spectrometry data with exposure and toxicity forecasts to advance high-throughput environmental monitoring. Environ Int 88:269–280, doi: 10.1016/j.envint.2015.12.008.

Rappaport SM. 2012. Biomarkers intersect with the exposome. Biomarkers 17:483–489, doi: 10.3109/1354750X.2012.691553.

Rappaport SM. 2016. Genetic factors are not the major causes of chronic diseases. PloS One 11(4):e0154387, doi: 10.1371/journal.pone.0154387.

Rappaport SM, Barupal DK, Wishart D, Vineis DP, Scalbert A. 2014. The blood exposome and its role in discovering causes of disease. Environ Health Perspect 122:769–774, doi: 10.1289/ehp.1308015.

Rappaport SM, Li H, Grigoryan H, Funk WE, Williams ER. 2012. Adductomics: characterizing exposures to reactive electrophiles. Toxicol Lett 213:83–90, doi: 10.1016/j.toxlet.2011.04.002.

Rappaport SM, Smith MT. 2010. Environment and disease risks. Science 330:460–461, doi: 10.1126/science.1192603.

Rosas LG, Eskenazi B. 2008. Pesticides and child neurodevelopment. Curr Opin Pediatr 20:191–197, doi: 10.1097/MOP.0b013e3282f60a7d.

Smith CA, O’Maille G, Want EJ, Qin C, Trauger SA, Brandon TR, et al. 2005. METLIN: a metabolite mass spectral database. Ther Drug Monit 27:747–751.

Soltow QA, Strobel FH, Mansfield KG, Wachtman L, Park Y, Jones DP. 2013. High-performance metabolic profiling with dual chromatography-Fourier-transform mass spectrometry (DC-FTMS) for study of the exposome. Metabolomics 9(1 suppl):S132–S143.

Southam AD, Lange A, Al-Salhi R, Hill EM, Tyler CR, Viant MT. 2014. Distinguishing between the metabolome and xenobiotic exposome in environmental field samples analyses by direct-infusion mass spectrometry based metabolomics and lipidomics. Metabolomics 10(6):1050–1058.

Spira A, Beane J, Schembri F, Liu G, Ding C, Gilman S, et al. 2004. Noninvasive method for obtaining RNA from buccal mucosa epithelial cells for gene expression profiling. Biotechniques 36:484–487.

Tautenhahn R, Cho K, Uritboonthai W, Zhu Z, Patti GJ, Siuzdak G. 2012. An accelerated workflow for untargeted metabolomics using the METLIN database. Nat Biotechnol 30:826–828, doi: 10.1038/nbt.2348.

Vrijheid M, Slama R, Robinson O, Chatzi L, Coen M, van den Hazel P, et al. 2014. The Human Early-Life Exposome (HELIX): project rationale and design. Environ Health Perspect 122:535–544, doi: 10.1289/ehp.1307204.

Wambaugh JF, Setzer RW, Reif DM, Gangwal S, Mitchell-Blackwood J, Arnot JA, et al. 2013. High-throughput models for exposure-based chemical prioritization in the ExpoCast project. Environ Sci Technol 47(15):8479–8488, doi: 10.1021/es400482g.

Wei R, Li G, Seymour AB. 2010. High-throughput and multiplexed LC/MS/MRM method for targeted metabolomics. Anal Chem 82(13):5527–5533, doi: 10.1021/ac100331b.

Wild CP. 2005. Complementing the genome with an “exposome”: the outstanding challenge of environmental exposure measurement in molecular epidemiology. Cancer Epidemiol Biomarkers Prev 14:1847–1850.

Wild CP. 2012. The exposome: from concept to utility. Int J Epidemiol 41:24–32, doi: 10.1093/ije/dyr236.

Wishart DS, Jewison T, Guo AC, Wilson M, Knox C, Liu Y, et al. 2013. HMDB 3.0—the Human Metabolome Database in 2013. Nucleic Acids Res 41(database issue):D801–D807.

Wishart DS, Knox C, Guo AC, Eisner R, Young N, Gautam B, et al. 2009. HMDB: a knowledgebase for the human metabolome. Nucleic Acids Res 37(database issue):D603–D610.

Yu T, Park Y, Li S, Jones DP. 2013. Hybrid feature detection and information accumulation using high-resolution LC-MS metabolomics data. J Proteome Res 12(3):1419–1427, doi: 10.1021/pr301053d.

Zhang J, Shen H, Xu W, Xia Y, Barr DB, Mu X, et al. 2014. Urinary metabolomics revealed arsenic internal dose-related metabolic alterations: a proof-of-concept study in a Chinese male cohort. Environ Sci Technol 48(20):12265–12274, doi: 10.1021/es503659w.

Zhang X, Sebastiani P, Liu G, Schembri F, Zhang X, Dumas YM, et al. 2010. Similarities and differences between smoking-related gene expression in nasal and bronchial epithelium. Physiol Genomics 41:1–8, doi: 10.1152/physiolgenomics.00167.2009.

Zhou B, Xiao JF, Tuli L, Ressom HW. 2012. LC-MS-based metabolomics. Mol Biosyst 8(2):470–481, doi: 10.1039/c1mb05350g.

Climate and Health Co-Benefits in Low-Income Countries: A Case Study of Carbon Financed Water Filters in Kenya and a Call for Independent Monitoring

Author Affiliations open
1Civil and Environmental Engineering, Stanford University, Stanford, California, USA; 2Center for Innovation in Global Health, Stanford University, Stanford, California, USA; 3Division of Epidemiology, School of Public Health, University of California, Berkeley, Berkeley, California, USA; 4Department of Nutrition, University of California, Davis, Davis, California, USA; 5Innovations for Poverty Action, Nairobi, Kenya; 6Mathematica Policy Research, Washington, DC, USA

PDF icon PDF Version (363 KB)

  • Background: The recent global climate agreement in Paris aims to mitigate greenhouse gas emissions while fostering sustainable development and establishes an international trading mechanism to meet this goal. Currently, carbon offset program implementers are allowed to collect their own monitoring data to determine the number of carbon credits to be awarded.

    Objectives: We summarize reasons for mandating independent monitoring of greenhouse gas emission reduction projects. In support of our policy recommendations, we describe a case study of a program designed to earn carbon credits by distributing almost one million drinking water filters in rural Kenya to avert the use of fuel for boiling water. We compare results from an assessment conducted by our research team in the program area among households with pregnant women or caregivers in rural villages with low piped water access with the reported program monitoring data and discuss the implications.

    Discussion: Our assessment in Kenya found lower levels of household water filter usage than the internal program monitoring reported estimates used to determine carbon credits; we found 19% (n = 4,041) of households reported filter usage 2–3 years after filter distribution compared to the program stated usage rate of 81% (n = 14,988) 2.7 years after filter distribution. Although carbon financing could be a financially sustainable approach to scale up water treatment and improve health in low-income settings, these results suggest program effectiveness will remain uncertain in the absence of requiring monitoring data be collected by third-party organizations.

    Conclusion: Independent monitoring should be a key requirement for carbon credit verification in future international carbon trading mechanisms to ensure programs achieve benefits in line with sustainable development goals.

  • Citation: Pickering AJ, Arnold BF, Dentz HN, Colford JM Jr., Null C. 2017. Climate and health co-benefits in low-income countries: a case study of carbon financed water filters in Kenya and a call for independent monitoring. Environ Health Perspect 125:278–283;

    Address correspondence to A.J. Pickering, Civil and Environmental Engineering, 473 Via Ortega, Y2E2 Building, Room 247, Stanford University, Stanford, CA 94305 USA. Telephone: (650) 736-8668. E-mail:

    We thank G. Nyambane, C. Stewart, T. Bourdier, M. Wolfe, T. Wolfe, M. Harris, and M. Kremer for their help in making this work possible. We thank A. Kleysteuber for helpful comments on the manuscript.

    The data collection for this paper was supported by a grant from the Bill and Melinda Gates Foundation to the University of California, Berkeley to conduct the WASH Benefits study in Kenya (#OPPGD759).

    The funder had no role in data collection or analysis. The nonprofit organization Innovations for Poverty Action (IPA) in Kenya collected the data under the oversight of the authors. At the time of data collection, IPA was also involved in a project implementing community chlorine dispensers in Kenya and was exploring accessing carbon financing to operate the dispensers. None of the authors were involved in IPA’s chlorine dispenser implementation project. At the time of data collection, H.N.D. was employed by Innovations for Poverty Action, Kenya. C.N. is currently employed by Innovations for Poverty Action, Kenya, and Mathematica Policy Research, Washington, DC.

    The authors declare they have no actual or potential competing financial interests.

    Received: 11 April 2016
    Revised: 22 August 2016
    Accepted: 29 August 2016
    Published: 16 September 2016

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact Our staff will work with you to assess and meet your accessibility needs within 3 working days.


The recent Paris Agreement, which was signed by nearly 200 nations, aims to prevent global warming from exceeding 2°C above pre-industrial levels (UNFCCC 2015). The agreement acknowledges global disparities in present and historic energy consumption by including goals of sustainable development and poverty eradication alongside its primary goal of mitigating global greenhouse gas emissions. To enable countries to cost effectively meet their nationally determined emission reduction goals, the Paris Agreement also allows parties (e.g., countries) to transfer mitigation outcomes that help them achieve these goals and establishes an international mechanism to promote mitigation and support sustainable development (Article 6) (UNFCCC 2015). Predating the Paris Agreement, the clean development mechanism (CDM) created by the 1997 Kyoto Protocol (United Nations 1998) facilitated carbon offset projects in lower-income countries. The CDM was designed to reduce carbon emissions more cost effectively by allowing developed countries with emission reduction targets to purchase carbon credits from developing countries where projects cost less to implement. Similar to the Paris Agreement, a second explicitly stated goal of the CDM was to assist developing countries achieve sustainable development (United Nations 1998). A voluntary carbon market also exists outside of the United Nations Framework Convention on Climate Change and its legal instruments, in which corporations, organizations, and individuals typically purchase carbon credits to offset their emissions. Designated Operational Entities (DOE) serve as third-party auditors to validate CDM project proposals and certify carbon emission reductions (UNFCCC 2014). However, in both compliance and voluntary markets, carbon offset program implementers are allowed to collect their own monitoring data to submit to third-party certification agencies to determine the number of carbon credits awarded (UNFCCC 2014; Implementers have financial incentives to report high offsets in order to claim more credits, which could lead to biased results. In this commentary, we argue for the need to independently monitor emissions reduction programs, and we present a case study of a carbon offset project designed to provide safe drinking water in rural Kenya.

Carbon Credits, Water Treatment, and Suppressed Demand

Carbon finance could be a financially sustainable approach to scale up water treatment and improve health in low-income settings, and in doing so would reduce greenhouse gas emissions while contributing to sustainable development (Hodge and Clasen 2014). Almost half (42%) of the global population does not have access to piped water into the home (WHO-UNICEF 2015), and it is estimated that 1.8 billion people drink from a water source with fecal contamination (Bain et al. 2014). Diarrhea is a leading cause of child mortality, causing 700,000 child deaths annually (Walker et al. 2013). Disinfection of drinking water supplies can prevent the transmission of common diarrheal pathogens (Prüss et al. 2002) and historically led to dramatic reductions in mortality associated with waterborne illness in the United States and Europe (Cutler and Miller 2005; Sedgwick and MacNutt 1910).

A number of programs to reduce greenhouse gas emissions are implementing zero-emission household water treatment technologies with carbon credit financing. Carbon credits (awarded for the avoidance, sequestration, or reduction of 1 ton of carbon dioxide equivalent) are generated by calculating the quantity of greenhouse gas emissions avoided by treating water with a zero-emission technology (e.g., water filter) against an estimated baseline of emissions if households had boiled the water using fossil fuel or nonrenewable biomass (Hodge and Clasen 2014). Greenhouse gas emissions resulting from the manufacturing, distribution, and behavior promotion of such water treatment technologies are also accounted for in carbon credit calculations. Baseline emissions can adjust upwards to account for “suppressed demand,” based on the concept that current emissions are constrained by limited resources in developing economies. For example, the baseline would include the carbon emissions theoretically emitted from households that would boil their drinking water if they had access to sufficient fuel and the resources to obtain it (independent of whether households actually boil their water). In addition to baseline emission estimates, a key parameter for awarding carbon credits under this system is the percentage of households that regularly use the water treatment units, determined by internal monitoring by the program implementer. Proponents emphasize the primary benefit of such programs is to improve water quality for high need populations; however, critics note greenhouse gas emissions are not actually reduced because most households would not boil their water in the absence of the programs (Yeo 2013).

Carbon for Water in Western Kenya

During April and May of 2011, Vestergaard Frandsen reported that they distributed 877,505 LifeStraw® Family water filters free of charge to > 4.5 million people in Kenya’s Western Province (Vestergaard Frandsen 2012). The program is registered to earn carbon credits certified by a third-party organization, the Gold Standard Foundation (, for use in the voluntary carbon market with a crediting period of 10 years. In this “Carbon for Water” program, all households received a LifeStraw® Family water filter [model 1.0; Vestergaard,​ily-1-0), a point-of-use water treatment product that does not require electricity to operate and is classified as “Highly Protective” by World Health Organization testing guidelines (WHO 2011). The Carbon for Water program implemented three household health education campaigns (July–August 2011, April–May 2012, and October 2012), employing almost 2,000 community health workers to personally visit households that had received a filter, and broadcasted messages about the program over the radio. The health messaging included training on proper filter use, as well as promotion of safe water storage to prevent recontamination, and handwashing with soap and filtered water. The program also established 32 complementary maintenance, repair, and education facilities in Western Province (Vestergaard Frandsen 2012). Vestergaard Frandsen hired staff to collect their own monitoring data (sampling between approximately 15,000 and 20,000 households each round), as well as contracted a local Kenyan firm to conduct audits (100–300 households each round). The program reported that it distributed filters to 91% of all households in Kenya’s Western Province and that water filter usage rates over time since distribution were 91% (0–6 months), 75% (7–18 months), and 81% (19–32 months) (Vestergaard Frandsen 2012, 2013, 2014a). The program assumed that 79.6% of households would boil their water if they had access to adequate resources in monitoring period 1 (0–6 months) and monitoring period 2 (7–18 months); this assumption was lowered to 52.8% of households for monitoring period 3 (19–32 months). The usage rates combined with the baseline emissions estimated using suppressed demand earned a total of 4,476,205 carbon credits during the period 1 June 2011–31 January 2014 (32 months) (Vestergaard Frandsen 2014a). Due to fluctuations in the price of carbon credits and unreleased implementation costs, actual program profits (or losses) to date are unknown.


The Kenya Carbon for Water program happened to overlap a large randomized controlled trial run by our research team evaluating the health effects of water, sanitation, hygiene, and nutritional interventions among newborns in rural villages in Kakamega, Bungoma, and Vihiga counties in Western Province (WASH Benefits study). As part of the WASH Benefits study, we collected household survey data to pilot test interventions and characterize baseline water management practices and drinking water quality in the WASH Benefits study population (Arnold et al. 2013; Christensen et al. 2015). The coincidental temporal and geographic overlap of WASH Benefits with the Carbon for Water program allowed us to measure household filter usage and microbial water quality over a 3-year period that included the first 32 months of program monitoring. Because of the potential importance of filter usage rates in interpreting the primary outcomes of the WASH Benefits trial, we added measures of filter ownership to existing household surveys that took place approximately 6, 18, and 24–36 months after the filter distribution by Vestergaard Frandsen during April and May in 2011. We sampled a different study population during each time period (Table 1). All surveys contained a standardized module to assess household water management practices, use of the LifeStraw® Family filter, and reasons for nonuse. Figure 1 is a map of household survey locations within the program area.

Table 1. Select View Table (HTML Version) for a 508-conformant version

Table 1. Self-reported ownership and usage of LifeStraw® Family filters among three study populations (A, B, and C) at 6, 18, and 24–36 months post-filter distribution and assessment of filter use based on observations by study staff.

View Table (HTML Version)
View larger image (TIF File)

Figure 1. Map of the study area showing the location of study households, with a second map showing the location of the study area within Kenya.

Figure 1. The Carbon for Water program distribution area is shown in light yellow (includes Bungoma, Kakamega, Vihiga, and Busia counties in Western Kenya). Each circle represents the location of one household enrolled in our assessment (n = 8,721); dark gray circles indicate study population A (surveyed approximately 6 months after LifeStraw® filter distribution), light gray circles indicate study population B (surveyed approximately 18 months post filter distribution), open circles indicate study population C (surveyed approximately 24–36 months post filter distribution). The location of the study area within Kenya is shown at right.

View larger image (TIF File)

The 6-month survey (1–29 November 2011) draws on data collected during the baseline assessment for two pilot randomized controlled trials conducted as part of WASH Benefits, simultaneously implemented in 72 rural villages in Western Kenya; the trials evaluated the adoption of household water, sanitation, and hand hygiene interventions [see Christensen et al. (2015) for further details]. These trials enrolled households with caregivers of 4- to 16-month-old children in Kakamega (367 households in the Shianda Location), and pregnant women and caregivers with children < 3 months near the town of Bungoma (132 households in the Kibingei Location). The 18-month survey (27 November–19 December 2012) and the 24–36 month survey (19 June 2013–21 May 2014) data were obtained from the baseline assessment and enrollment survey of the full-scale WASH Benefits study (Arnold et al. 2013). The WASH Benefits study enrolled pregnant mothers in their second or third trimester in rural villages with low levels of piped water access (< 20% of households) in Kakamega, Bungoma, and Vihiga counties. All respondents provided written informed consent; those not comfortable signing their name provided a thumb print. Human subjects institutional review boards (IRBs) at the University of California, Berkeley, Stanford University, and the Kenya Medical Research Institute (KEMRI) approved the study protocols.

Field staff asked respondents to fetch a cup of water the way they normally would for a young child, then observed from where the respondent obtained the water and how it was stored and extracted. Field staff inquired if anyone in the household “had done anything to make the water less cloudy or safer to drink,” and if so, what method was used (without prompting on specific water treatment methods). If the respondent did not report a water treatment method, the field staff asked if the respondent ever treats drinking water and to list all methods used. Field staff questioned if the household had received a LifeStraw® Family filter. If the household reported receiving a filter, we observed if the filter was present, hanging on the wall, looked unused (e.g., visible dust), and contained water or moisture. Field staff asked if the filter was working and if there were any issues that prevented use. Finally, respondents reported if and when a representative from the Carbon for Water program had most recently visited their home to promote the LifeStraw® filter.

We also collected a sample of stored drinking water to assess levels of Escherichia coli contamination during the 18-month and 24–36 month surveys, which allowed us to compare water that respondents indicated was filtered with a LifeStraw® filter to water reported by the respondent to be unfiltered. Respondents were asked to pour the fetched cup of stored water into a sterile 100 mL Whirl-Pak® bag (product no. B01040WA; Nasco, The samples were placed on ice, transported to a field lab, and processed by membrane filtration within 8 hr. A volume of 100 mL was vacuum filtered, plated on MI agar, then incubated at 35°C for 24 hr. E. coli were enumerated following the U.S. Environmental Protection Agency (EPA) method 1604 (U.S. EPA 2002).


We completed 499 household surveys 6 months after filter distribution (study population “A” in Table 1), 531 household surveys 18 months after filter distribution (study population “B”), and 7,691 household surveys 24–26 months after filter distribution (study population “C”) as part of our on-going enrollment into the WASH Benefits pilot study and main study. Our measurements indicate that the Carbon for Water program coverage was impressively high; the percentage of households that remembered ever receiving a filter was 91% 6 months after filter distribution. However, this number fell to 70% at 18 months and 53% at 24–36 months (Table 1). Among households that reported filter ownership, 95% said the Carbon for Water program had visited their home to promote the LifeStraw® at the 18-month survey and 55% at 24–36 months (Table 1, this question was not asked at the 6-month visit). Although we did not measure the same households at each time point, we documented a progressive decline in reported filtering of currently stored drinking water among households that had received a filter: 29% at 6 months (118 out of 408 in study population A), 27% at 18 months (93 out of 344 in study population B) and 7% at 24–36 months (236 out of 3,383 in study population C) (Figure 2, Table 1). Similarly, the percentage of households that reported using the filter as a drinking water treatment method was lowest among study population C, and highest among study population A (Figure 2, Table 1). The percentage of filters with observed moisture or water (one indicator of recent use) also declined with time since distribution (32% at 6-months, 25% at 12-months, and 12% at 24–36 months). Half (51%) of households reported filters were not working after 24–36 months. When asked about issues preventing use of the LifeStraw® filter, 35% of households that received a filter said the filter was too slow or took too much time, 17% said the filter was blocked or not working, 8% said the filter had a bad smell or taste, and 7% thought the filter was bad for their health (data combined from all study populations, n = 4,868).

Figure 2. Graph showing the percent of households (y-axis) according to time period (6, 18, or 24–36 months) (x-axis) with different symbols used to indicate filter use based on program monitoring, observed filter use for drinking water, observed filter use for stored drinking water, respectively.

Figure 2. Self reported indicators of LifeStraw® Family filter usage (solid black and gray circles) collected by our assessment of rural households with pregnant women or caregivers of young children that reported receiving a filter in Western Kenya; a separate study population was measured at each time point (6 months, 18 months, and 24–36 months). Open circles indicate reported usage in official program monitoring reports for carbon credit verification; the three monitoring periods ended at 6 months, 19 months, and 32 months post filter distribution. Standard errors (not shown) are less than 3 percentage points for all data points.

View larger image (TIF File)

Among our full study population (all households in study populations A, B, and C), for drinking water most households access protected springs (52%, n = 7,170) or unprotected (16%) springs, 16% access shallow wells, with the remainder accessing borewells, surface water, or collecting rainwater; 71% (n = 7,170) of households access improved water sources according to the WHO/UNICEF Joint Monitoring Program definition (WHO-UNICEF 2016). The percentage of households accessing improved water sources is similar to the 68% reported in the Carbon for Water program’s third monitoring report (Vestergaard Frandsen 2014a), as well as the 67% reported by the 2014 Demographic and Health Survey (DHS Program 2015). Among our full study population, filtering of stored drinking water was not significantly different between households accessing improved water sources versus those accessing unimproved water sources (6.6% vs. 6.4%; p = 0.786). Among all households in our study with available stored water, field staff observed the majority (78%, n = 7,041) of stored drinking water containers to be covered. When asked to retrieve a glass of water, 61% (n = 7,069) of respondents extracted water by dipping a cup into the storage container.

Microbial water quality was improved in filtered stored drinking water we collected from our study populations B and C (measured 18 months and 24–26 months after filter distribution), but typically still reflected contamination with the fecal indicator bacteria Escherichia coli. The geometric mean in filtered water was 20.6 E. coli per 100 mL (n = 296), compared to 27.5 E. coli (n = 5,492) in unfiltered water (p = 0.012). Over half (52%) of respondents reported storing filtered water for 2 or more days, indicating the potential for recontamination during storage. Water treatment methods other than filter usage cited by respondents included adding chlorine (5%, n = 7,122) and boiling (1%, n = 7,122) (data from all three study populations). Stored water reported by households to be treated with locally available chlorine had a geometric mean < 10 E. coli per 100 mL (n = 311 samples from study populations B and C).


Our measurements show lower usage of LifeStraw® Family filters compared with the program’s own monitoring data (Figure 2). We document 19% usage of the filters as a drinking water treatment method 2–3 years after filter distribution (among households with pregnant women) compared to the 81% usage reported by the program approximately 2.7 years after filter distribution (Vestergaard Frandsen 2014a). In addition, while data reported by the Carbon for Water program indicated consistent high use over time, our data suggest that use decreased over time. These findings suggest that providing water treatment technologies like the Lifestraw® Family filter do not automatically translate into improved water quality at the point of use (and by extension improved health). This result is consistent with other studies showing poor long-term adoption and inconsistent use of household water treatment products provided programmatically (Arnold et al. 2009; Luby et al. 2008; Rosa et al. 2016). A large systematic review found no evidence that household water treatment interventions reduce child diarrhea after 12 or more months of implementation, possibly because product usage is not sustained (Clasen et al. 2015). Considering the equivocal success of previous household water treatment programs at achieving sustained access to safe water, future greenhouse gas reduction programs may want to consider implementing emission-free water treatment technologies that automatically treat drinking water at the community level instead of relying on users to consistently treat their own water (Amin et al. 2016; Pickering et al. 2015).

Providing access to safe water in Kenya is a stated goal of the Vestergaard Frandsen Carbon for Water program (Vestergaard Frandsen 2014b). The immense scale of such programs presents enormous opportunity to improve water quality for millions living in poverty. We believe greenhouse gas emissions reduction programs claiming to provide safe water to low-income people should be required to demonstrate that improvements in drinking water quality are actually achieved and sustained though regular independent monitoring of microbial water quality. We found low prevalence of boiling (1%) to treat drinking water by households with and without a LifeStraw® Family filter, indicating the Carbon for Water program in Kenya has achieved minimal reductions in actual greenhouse gas emissions (in contrast with the projected reductions from hypothetical baseline emission levels estimated using suppressed demand). When suppressed demand is employed to calculate baseline scenarios, the potential absence of actual greenhouse gas emissions reductions places additional onus on programs to improve sustainable development outcomes, such as drinking water quality.


The 2015 Paris Agreement provides a timely opportunity to create a new or updated international trading mechanism that could improve the environmental, health, and economic benefits of future emissions reduction programs by mandating independent monitoring for carbon credit verification. Furthermore, carbon offset programs will only contribute to the Paris Agreement’s goals of greenhouse gas emissions reduction, sustainable development, and poverty alleviation if they are implemented successfully. Without independent evaluation, it is difficult to confirm that monitoring results are accurate. Evaluations of international development programs are widely acknowledged to be stronger if conducted independently from the implementing organizations because of the inherent conflict of interest—financial and otherwise—that implementers have in the success of their own programs (Gertler et al. 2011; Purcell 2003; Savedoff et al. 2006; USAID 2016). Similar to monitoring guidelines for clinical research studies and finances of publicly traded companies, we propose that carbon credit program monitoring standards be revised to ensure that monitoring activities are free from real or perceived conflicts of interest.

One strategy for achieving independent cost-effective monitoring would be to transfer funds that program implementers already spend on their own internal monitoring to the third-party credit certification organizations, which in turn would contract independent evaluations. This would expand the scope of credit certification organizations, but would help prevent conflicts of interest arising in the evaluation process. The monitoring process would also ideally include pre-specified indicators used to evaluate non-emissions related benefits (Miguel et al. 2014), such as microbial water quality for water treatment projects (Hodge and Clasen 2014). In addition, a reporting schedule tied to publicly available results would facilitate prompt feedback to program implementers to improve program effectiveness, improve transparency to the global community, and improve data reliability.


Amin N, Crider YS, Unicomb L, Das KK, Gope PS, Mahmud ZH, et al. 2016. Field trial of an automated batch chlorinator system at shared water points in an urban community of Dhaka, Bangladesh. Journal of Water, Sanitation and Hygiene for Development 6(1):32–41, doi: 10.2166/washdev.2016.027.

Arnold B, Arana B, Mäusezahl D, Hubbard A, Colford JM Jr. 2009. Evaluation of a pre-existing, 3-year household water treatment and handwashing intervention in rural Guatemala. Int J Epidemiol 38:1651–1661.

Arnold BF, Null C, Luby SP, Unicomb L, Stewart CP, Dewey KG, et al. 2013. Cluster-randomised controlled trials of individual and combined water, sanitation, hygiene and nutritional interventions in rural Bangladesh and Kenya: the WASH Benefits study design and rationale. BMJ Open 3:e003476, doi: 10.1136/bmjopen-2013-003476.

Bain R, Cronk R, Hossain R, Bonjour S, Onda K, Wright J, et al. 2014. Global assessment of exposure to faecal contamination through drinking water based on a systematic review. Trop Med Int Health 19:917–927, doi: 10.1111/tmi.12334.

Christensen G, Dentz HN, Pickering AJ, Bourdier T, Arnold BF, Colford JM Jr, et al. 2015. Pilot cluster randomized controlled trials to evaluate adoption of water, sanitation, and hygiene interventions and their combination in rural western Kenya. Am J Trop Med Hyg 92:437–447, doi: 10.4269/ajtmh.14-0138.

Clasen T, Alexander KT, Sinclair D, Boisson S, Peletz RL, Chang HH, et al. 2015. Interventions to improve water quality for preventing diarrhoea. Cochrane Database Syst Rev 10:CD004794, doi: 10.1002/14651858.CD004794.pub3.

Cutler DM, Miller G. 2005. The role of public health improvements in health advances: the twentieth-century United States. Demography 42:1–22.

DHS Program. 2015. Kenya Demographic and Health Survey 2014. Final report. Rockville, MD:ICF International, DHS Program.​308.pdf [accessed 14 July 2015].

Gertler PJ, Martinez S, Premand P, Rawlings LB, Vermeersch CMJ. 2011. Impact Evaluation in Practice. Washington, DC:World Bank.

Hodge JM, Clasen TF. 2014. Carbon financing of household water treatment: background, operation and recommendations to improve potential for health gains. Environ Sci Technol 48:12509–12515, doi: 10.1021/es503155m.

Luby SP, Mendoza C, Keswick BH, Chiller TM, Hoekstra RM. 2008. Difficulties in bringing point-of-use water treatment to scale in rural Guatemala. Am J Trop Med Hyg 78:382–387.

Miguel E, Camerer C, Casey K, Cohen J, Esterling KM, Gerber A, et al. 2014. Promoting transparency in social science research. Science 343:30–31, doi: 10.1126/science.1245317.

Pickering AJ, Crider Y, Amin N, Bauza V, Unicomb L, Davis J, et al. 2015. Differences in field effectiveness and adoption between a novel automated chlorination system and household manual chlorination of drinking water in Dhaka, Bangladesh: a randomized controlled trial. PLoS One 10:e0118397, doi: 10.1371/journal.pone.0118397.

Prüss A, Kay D, Fewtrell L, Bartram J. 2002. Estimating the burden of disease from water, sanitation, and hygiene at a global level. Environ Health Perspect 110:537–542.

Purcell R. 2003. Independent Evaluation: Principles, Guidelines And Good Practice: The World Bank Development Grant Facility (DGF) Technical Note.​F/Resources/Evaluation&LearningNote.pdf [accessed 3 August 2016].

Rosa G, Kelly P, Clasen T. 2016. Consistency of use and effectiveness of household water treatment practices among urban and rural populations claiming to treat their drinking water at home: a case study in Zambia. Am J Trop Med Hyg 94(2):445–455, doi: 10.4269/ajtmh.15-0563.

Savedoff WD, Levine R, Birdsall N. 2006. When Will We Ever Learn? Improving Lives through Impact Evaluation. Report of the Evaluation Gap Working Group. Washington, DC:Center for Global Development.​/7973_file_WillWeEverLearn.pdf [accessed 3 August 2016].

Sedgwick WT, MacNutt JS. 1910. On the Mills-Reincke Phenomenon and Hazen’s theorem concerning the decrease in mortality from diseases other than typhoid fever following the purification of public water supplies. J Infect Dis 7:489–564.

U.S. EPA (U.S. Environmental Protection Agency). 2002. Method 1604: Total Coliforms and Escherichia coli in Water by Membrane Filtration Using a Simultaneous Detection Technique (MI Medium). EPA-821-R-02-024. Washington, DC:U.S. EPA.

UNFCCC (United Nations Framework Convention on Climate Change). 2014. CDM Project Cycle.​html [accessed 3 August 2016].

UNFCCC. 2015. Adoption of the Paris Agreement. Proposal by the President.​ound/convention/application/pdf/english_​paris_agreement.pdf [accessed 3 August 2016].

United Nations. 1998. Kyoto Protocol to the United Nations Framework Convention on Climate Change.​kpeng.pdf [accessed 3 August 2016].

USAID (U.S. Agency for International Development). 2016. Strengthening Evidence-Based Development: Five Years of Better Evaluation Practice at USAID, 2011–2016.​s/documents/1870/Strengthening%20Evidenc​e-Based%20Development%20-%20Five%20Years​%20of%20Better%20Evaluation%20Practice%2​0at%20USAID.pdf [accessed 3 August 2016].

Vestergaard Frandsen. 2012. GS0086 Sustainable Deployment of the LifeStraw Family in Rural Kenya Verification Report – Verification 1 – June 1, 2011 to November 30, 2011.​rocessDocument/downloadDocumentById/1030​00000010551 [accessed 14 July 2015].

Vestergaard Frandsen. 2013. GS886 Sustainable Deployment of the LifeStraw Family in Rural Kenya Verification Report–Verification 2–December 1, 2011 to October 31, 2012.​rocessDocument/downloadDocumentById/1030​00000031855 [accessed 14 July 2015].

Vestergaard Frandsen. 2014a. GS886 Sustainable Deployment of the LifeStraw Family in Rural Kenya Monitoring Report – Verification 3 – November 1, 2012 to January 31, 2014.​rocessDocument/downloadDocumentById/1030​00000037580 [accessed 14 July 2015].

Vestergaard Frandsen. 2014b. Lifestraw® Carbon for Water.​lifestraw-carbon-for-water [accessed 3 August 2016].

Walker CLF, Rudan I, Liu L, Nair H, Theodoratou E, Bhutta ZA, et al. 2013. Global burden of childhood pneumonia and diarrhoea. Lancet 381:1405–1416, doi: 10.1016/S0140-6736(13)60222-6.

WHO (World Health Organization). 2011. Evaluating Household Water Treatment Options: Health-Based Targets and Microbiological Performance Specifications.​th/publications/2011/evaluating_water_tr​eatment.pdf [accessed 5 July 2015].

WHO-UNICEF (World Health Organization-United Nations International). 2015. Progress on Sanitation and Drinking Water: 2015 Update and MDG Assessment.​/177752/1/9789241509145_eng.pdf?ua=1 [accessed 3 August 2016].

WHO-UNICEF. 2016. The JMP Method – Deriving Progress Estimates.​ds/method/ [accessed 3 August 2016].

Yeo D. 2013. Towards more responsible carbon finance for water. Waterlines 32:308–314, doi: 10.3362/1756-3488.2013.032.

Health Risk Assessment of Dietary Cadmium Intake: Do Current Guidelines Indicate How Much is Safe?

Author Affiliations open
1Centre for Kidney Disease Research, Translational Research Institute, University of Queensland School of Medicine, Woolloongabba, Brisbane, Queensland, Australia; 2Department of Molecular Biology and Applied Physiology, Tohoku University School of Medicine, Sendai, Japan

PDF icon PDF Version (155 KB)

  • Background: Cadmium (Cd), a food-chain contaminant, is a significant health hazard. The kidney is one of the primary sites of injury after chronic Cd exposure. Kidney-based risk assessment establishes the urinary Cd threshold at 5.24 μg/g creatinine, and tolerable dietary intake of Cd at 62 μg/day per 70-kg person. However, cohort studies show that dietary Cd intake below a threshold limit and that tolerable levels may increase the risk of death from cancer, cardiovascular disease, and Alzheimer’s disease.

    Objective: We evaluated if the current tolerable dietary Cd intake guideline and urinary Cd threshold limit provide sufficient health protection.

    Discussion: Staple foods constitute 40–60% of total dietary Cd intake by average consumers. Diets high in shellfish, crustaceans, mollusks, spinach, and offal add to dietary Cd sources. Modeling studies predict the current tolerable dietary intake corresponding to urinary Cd of 0.70–1.85 μg/g creatinine in men and 0.95–3.07 μg/g creatinine in women. Urinary Cd levels of < 1 μg/g creatinine were associated with progressive kidney dysfunction and peripheral vascular disease. A urinary Cd of 0.37 μg/g creatinine was associated with breast cancer, whereas dietary Cd of 16–31.5 μg/day was associated with 25–94% increase in risk of estrogen receptor–positive breast cancer.

    Conclusion: Modeling shows that dietary intake levels for Cd exceed the levels associated with kidney damage and many other adverse outcomes. Thus, the threshold level of urinary Cd should be re-evaluated. A more restrictive dietary intake guideline would afford enhanced health protection from this pervasive toxic metal.

  • Citation: Satarug S, Vesey DA, Gobe GC. 2017. Health risk assessment of dietary cadmium intake: do current guidelines indicate how much is safe? Environ Health Perspect 125:284–288;

    Address correspondence to S. Satarug, Centre for Kidney Disease Research, Translational Research Institute, University of Queensland School of Medicine, Kent St., Woolloongabba, Brisbane, Australia 4102. Telephone: 61-7-344-38011. E-mail:

    S.S. (L-14568) received Research Professorship from the Japan Society for Promotion of Science (JSPS).

    The authors declare they have no actual or potential competing financial interests.

    Received: 14 March 2016
    Revised: 23 August 2016
    Accepted: 12 September 2016
    Published: 1 March 2017

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact Our staff will work with you to assess and meet your accessibility needs within 3 working days.


Cadmium (Cd) is a nonessential metal, a food-chain contaminant, and a constituent of cigarette smoke and polluted air (IPCS 1992). Diet is a major source of Cd exposure for nonsmokers, while cigarette smoke is an additional source for smokers (Satarug et al. 2013). Cd accumulates in the kidneys. Currently, human exposure to Cd is assumed, primarily, to damage the kidneys, especially the proximal tubular cells where the metal selectively concentrates. Consequently, kidney-based assessment of Cd exposure is often applied, with urinary Cd levels as indicators of Cd exposure. Other means of measuring Cd exposure relate to dietary intake estimates. Cancer risk assessment by the International Agency for Research on Cancer (IARC 1993) established Cd as a human lung carcinogen. Non-cancer risk assessment by the Food and Agriculture Organization/World Health Organization (FAO/WHO 1989, 1993, 2010) established a tolerable exposure and urinary threshold level that should protect against kidney damage. However, a wide diversity of Cd toxicity levels is increasingly apparent from recent studies, including the U.S. National Health and Nutrition Examination Survey (NHANES) (Hyder et al. 2013; Lin et al. 2013, 2014). Challenging kidney-based assessment is the observation that the carcinogenic effects of Cd appear to occur at exposure levels below the levels associated with kidney effects.

In this article, we highlight the basis of risk assessment for dietary Cd intakes together with dietary Cd intake estimates, derived from dietary and modeling studies. We provide evidence supporting the use of urinary Cd, a measure of cumulative lifetime exposure (body burden), in risk assessment, as opposed to the use of dietary intake estimates. We reviewed cross-sectional and longitudinal studies (published between 2010 and 2016) that link current dietary Cd intake levels to kidney damage, chronic kidney disease (CKD), cancer, and many other adverse health outcomes.

Kidney Threshold Risk Assessment

In 1989, the Joint FAO/WHO Expert Committee on Food Additives (JECFA) established the safe dietary intake guideline, known as the Provisional Tolerable Weekly Intake (PTWI), defined as an estimate of the amount of the chemical with no intended function that can be ingested weekly over a lifetime without appreciable health risk (FAO/WHO 1989). The original PTWI for Cd was set at 400–500 μg per person per week, based on kidney critical concentration of Cd at 200 μg/g kidney wet weight, attainable after dietary Cd exposure of 140–260 μg/day for over 50 years or 2,000 mg of Cd over a lifetime (FAO/WHO 1989). It is considered that kidney damage by Cd causes a reduction in tubular re-absorptive function, evident from excess urinary excretion of nutrients (amino acids, glucose, zinc, calcium) and low molecular weight proteins, notably β2-microglobulin (β2-MG), retinol binding protein (RBP). Wallin et al. (2014) observed a positive correlation between kidney Cd levels and urinary α1-microglobulin (α1-MG) levels, and they suggested that urinary α1-MG could serve as a more sensitive biomarker for kidney toxicity, compared with other injury biomarkers like kidney injury molecule-1 (KIM-1), RBP, and β2-MG, which do not correlate with kidney Cd levels in chronic low-level exposure conditions. The original PTWI for Cd was later revised to 7 μg/kg body weight per week (70 μg/day for a 70-kg person) (FAO/WHO 1993). The current tolerable exposure was set at 25 μg/kg body weight per month (62 μg/day for a 70-kg person), and a urinary threshold of 5.24 μg/g creatinine (FAO/WHO 2010).

Dietary Cd Intake Estimates for Swedish and French Populations

In the Swedish National Food Consumption Survey, Sand and Becker (2012) reported a dietary Cd intake of 10.6 μg/day for an average consumer. In this group, 40–50% of the Cd came from staple foods such as potatoes and wheat. For the high-Cd consumer, that is exposure above the 95th percentile, intake was 23 μg/day with additional Cd reportedly coming from seafood and spinach. Data derived from the second French Total Diet Study (TDS) (Arnich et al. 2012) showed the dietary Cd intake of 11.2 μg/day for an average consumer and 18.9 μg/day for the high consumer. For the average consumer, 35% came from bread products and another 26% from potato-based products. The additional Cd came from consumption of mollusks and crustaceans in the high-consumer group. Dietary Cd exposure in the second TDS (2007–2009) was four times higher than that of the first TDS (2000–2004), while the Cd concentrations were reported for the different food groups: Crustaceans and mollusks had the highest Cd content (0.167 mg/kg), followed by offal (0.053 mg/kg), sweet and savory biscuits (0.030 mg/kg) and cereal bars, and chocolate (0.029 mg/kg). However, data on variations in dietary habits were not reported. Likewise, the TDS did not report the body status of essential metals, which governs Cd absorption, or resultant toxicity. A study of 1,764 post-menopausal Danish women indicated leafy vegetables and soy-based products to be dietary Cd sources, but such dietary intake estimates (average 14 μg/day) marginally correlated with urinary Cd levels (Vacchi-Suzzi et al. 2015). These findings suggest such dietary intake estimates to be of limited use in health risk assessment.

Urine Cd Concentration as a Measure of Cumulative Lifetime Cd Intake

Use of urinary Cd concentration, as a measure of cumulative lifetime exposure and body burden, has its foundation in the Cd-toxicokinetics model, developed by Kjellström and Nordberg (1978) from Swedish autopsy data. In a recent analysis of Cd levels in kidney, blood, and urine samples from 109 living kidney donors (mean age 51 years, mean kidney Cd 12.9 μg/g wet weight), Akerstrom et al. (2013) found a urine-to-kidney Cd ratio of 1:60, and urinary Cd of 0.42 μg/g creatinine corresponded to kidney Cd of 25 μg/g kidney. Assuming a urine-to-kidney Cd ratio of 1:20, urinary Cd of 1.25 μg/g creatinine corresponded to 25 μg/g Cd per kidney wet weight, comparable with Cd levels found in kidney cortex samples from Australians, 41–50 years of age (Satarug et al. 2002). Using the 1:20 ratio, the current WHO urinary Cd threshold of 5.24 μg/g creatinine (FAO/WHO 2010) corresponds to kidney Cd > 100 μg/g kidney weight, the levels seen mostly in workers exposed to high-Cd doses via inhalation. In the general population, blood Cd is considered a good estimate of body burden because population blood Cd levels correlate with urine Cd levels (Tellez-Plaza et al. 2010; Wu et al. 2014). Blood Cd is a better estimate of exposure for the elderly, people with diabetes, hypertension, and heavy smokers because the high prevalence of kidney dysfunction in these people may bias associations between their urine Cd levels and health outcomes.

Model-Based Prediction of Urinary Cd Excretion at Tolerable Intake Guideline

Reverse dosimetry theory dictates that dietary Cd exposure and urinary Cd levels can be derived from a Cd toxicokinetic model, which describes mathematical relationships among the parameters, influencing Cd body burden, such as absorption rate, tissue distribution, half-life, and elimination rate. The original Cd-toxicokinetic model predicts that a Cd level of 50 μg/g kidney cortex wet weight corresponds to urinary Cd excretion of 2–4 μg/day, attainable after 50-year intake of dietary Cd at the tolerable weekly intake rate. A simulation model of Cd-toxicokinetics has been developed as a tool kit for prediction of Cd intake via oral (diet, water) versus inhalation (cigarette smoke, air) routes as a function of age and sex (Ruiz et al. 2010). Such simulation models predict that dietary intake of Cd at current tolerable monthly intake rate for 50 years will result in urinary Cd of 0.70–1.85 μg/g creatinine in men and 0.95–3.07 μg/g creatinine in women (Satarug et al. 2013). These urinary Cd levels, derived from modeled tolerable intake of dietary Cd, have been associated with kidney damage and CKD (Ferraro et al. 2010), concurrent with death from cancer (Lin et al. 2013; Adams et al. 2012), liver-related disease (Hyder et al. 2013), cardiovascular disease (CVD), ischemic heart disease, coronary heart disease (Tellez-Plaza et al. 2012b), and Alzheimer’s disease (Min and Min 2016). Tables 1 and 2 show details of population-based studies, giving evidence of Cd intake and exposure biomarkers associated with various adverse outcomes.

Table 1. Select View Table (HTML Version) for a 508-conformant version

Table 1. Adverse outcomes associated with Cd exposure in cross sectional studies.

View Table (HTML Version)
View larger image (TIF File)

Table 2. Select View Table (HTML Version) for a 508-conformant version

Table 2. Adverse outcomes associated with Cd exposure in longitudinal studies.

View Table (HTML Version)
View larger image (TIF File)

Cross Sectional Studies

Table 1 is a summary of cross-sectional and case-control studies of Cd exposure outcomes. In NHANES 1999–2006 (Ferraro et al. 2010), blood and urinary Cd levels > 1 μg/L were associated with kidney damage [OR 1.41, 95% confidence interval (CI): 1.10, 1.82] and CKD (OR 1.48, 95% CI: 1.01, 2.17). An association of Cd exposure and CKD became obscured when creatinine was used to correct for diluting effects of spot urine samples (Ferraro et al. 2010). This may indicate variability in creatinine secretion by kidney. In NHANES 2011–2012 (Lin et al. 2014), blood Cd levels > 0.53 μg/L were associated with kidney damage (OR 2.04, 95% CI: 1.13, 3.69) and low GFR (OR 2.21, 95% CI: 1.09, 4.50), and risk of Cd-induced kidney damage was particularly high (OR 3.38, 95% CI: 1.39, 8.28) in the participants who had lower zinc status, compared with those with higher zinc status. In NHANES 1999–2004 (Tellez-Plaza et al. 2010), urinary Cd levels ≥ 0.69 μg/g creatinine were associated with peripheral arterial disease (PAD) in men (OR 4.90, 95% CI: 1.55, 15.54), and in women (OR 0.56, 95% CI: 0.18, 1.71). Further, PAD risk in male nonsmokers increased with blood Cd levels, but PAD prevalence and blood Cd levels in female nonsmokers showed a U-shape relation, reflecting effects at blood Cd levels below 0.3 μg/L. In NHANES III (1988–1994), Hyder et al. (2013) found that urinary Cd levels ≥ 0.83 μg/g creatinine were associated with liver inflammation in women (OR 1.26, 95% CI: 1.01, 1.57), while urinary Cd levels ≥ 0.65 μg/g creatinine were associated with liver inflammation (OR 2.21, 95% CI: 1.64, 3.00), non-alcoholic fatty liver (OR 1.30, 95% CI: 1.01, 1.68) and non-alcoholic steatohepatitis (OR 1.95, 95% CI: 1.11, 3.41) in men. Ciesielski et al. (2013) found a 1 μg/L increment in urinary Cd was associated with a 1.93% reduction in a neurocognitive test for attention/perception domain among nonsmokers in NHANES III. In NHANES 2007–2010 (Scinicariello and Buser 2015), blood Cd levels ≥ 0.54 μg/L were associated with depressive symptoms in nonsmokers (OR 2.91, 95% CI: 1.12, 7.58), and smokers (OR 2.69, 95% CI: 1.13, 6.42). In NHANES 2005–2008, Wu et al. (2014) found blood Cd levels ≥ 0.66 μg/L were associated with age-related macular degeneration (AMD) (OR 1.56, 95% CI: 1.02, 2.40). The Cd and AMD association was particularly strong in non-Hispanic whites with urinary Cd levels ≥ 0.35 μg/L (OR 3.31, 95% CI: 1.37, 8.01). In NHANES 2005–2008 (Wallia et al. 2014), urinary Cd levels > 1.4 μg/g creatinine were associated with risk of prediabetes among nonsmokers. In NHANES 1999–2008, Gallagher et al. (2010) found urinary Cd levels ≥ 0.37 μg/g creatinine were associated with breast cancer among women (OR 2.50, 95% CI: 1.11, 5.63). In another study, Itoh et al. (2014) found dietary Cd intake levels ≥ 31.5 μg/day were associated with estrogen receptor positive (ER+) breast cancer in Japanese women (OR 1.94, 95% CI: 1.04, 3.63).

Longitudinal Studies

Table 2 is a summary of longitudinal studies of Cd exposure outcomes. In a Swedish cohort (Julin et al. 2012), dietary Cd intake levels ≥ 16 μg/day were associated with breast cancer (RR 1.27, 95% CI: 1.07, 1.50), and ER+ breast cancer (RR 1.25, 95% CI: 1.03, 1.52). In NHANES 1999–2004 follow-up (Tellez-Plaza et al. 2012b), urinary Cd levels ≥ 0.57 μg/g creatinine were associated with death from CVD (HR 1.74, 95% CI: 1.07, 2.83), ischemic heart disease (HR 2.53 95% CI: 1.54, 4.16), and coronary heart disease (HR 2.09, 95% CI: 1.06, 4.13). Population attributed risks suggest that reduction in urinary Cd from 0.57 to 0.14 μg/g creatinine could prevent 8.8% overall deaths and 9.2% CVD deaths. An equivalent analysis using blood Cd data gives parallel results; a reduction of blood Cd from 0.80 to 0.22 μg/L could prevent 7% overall deaths and 7.5% CVD deaths. In NHANES III follow-up, Adams et al. (2012) found urinary Cd levels ≥ 0.58 μg/g creatinine were associated with death from lung cancer in men (HR 3.22, 95% CI: 1.26, 8.25), while Hyder et al. (2013) found female urinary Cd levels ≥ 0.83 μg/g creatinine, and male urinary Cd levels ≥ 0.65 μg/g creatinine were associated with death from liver-related diseases (HR 3.42, 95% CI: 1.12, 10.47). Also in NHANES III follow-up (Lin et al. 2013), urinary Cd levels > 0.79 μg/g creatinine were associated with cancer death in men (HR 3.13, 95% CI: 1.88, 5.20), while urinary Cd levels > 1.05 μg/g creatinine were associated with cancer death in women (HR 1.65, 95% CI: 1.13, 2.41). In the NHANES 1999–2004 follow-up (Patel et al. 2013), a 1-SD change in logged exposure levels was associated with death from all causes (HR 1.6, 95% CI: 1.3, 2.0 for urinary Cd, and HR 1.4, 95% CI: 1.2, 1.6 for blood Cd). In the NHANES 1999–2004 follow-up (Min and Min 2016), blood Cd levels > 0.6 μg/L were associated with death from Alzheimer’s disease (HR 3.83, 95% CI: 1.39, 10.59).


Chronic intake of low-level dietary Cd has long been viewed as a subtle, long term and non-specific impairment. In contrast, such low-level dietary Cd intake has now been implicated in more serious health outcomes than previously perceived. Of concern, NHANES data indicate a significant proportion of the U.S. population is at risk of adverse effects from low-level dietary Cd intakes. Data from the NHANES 1999–2008 participants, aged 20–85 years, indicate Cd exposure prevalence of 94–98% in nonsmokers, and 96–99% in smokers (Riederer et al. 2013). A decline in Cd exposure in the U.S. over the 20-year (1988–2008) period could only be attributed to a reduction in smoking prevalence with little evidence to suggest a reduction in dietary Cd sources (Tellez-Plaza et al. 2012a). Overall Cd exposure prevalence among NHANES 2007–2012 participants, aged ≥ 20 years remains as high as 91.9% (Buser et al. 2016). These high exposure prevalence rates suggest that even a small increase in disease risk by Cd exposure can result in a large number of people affected by a disease that is preventable. In the NHANES 1999–2006, overall (female) prevalence of urinary Cd > 1, > 0.7 and > 0.5 μg/g creatinine among ≥ 20-year nonsmokers without CKD was 1.7 (2.5)%, 4.8 (7.1)%, and 10.8 (16)%, respectively (Mortensen et al. 2011). These data are a cause for concern because urinary Cd levels ≥ 0.37 to ≥ 0.65 μg/g creatinine have been associated with female breast cancer (Gallagher et al. 2010), death from heart disease (Tellez-Plaza et al. 2012b), death from cancer (Adams et al. 2012; Lin et al. 2013), and liver-related diseases (Hyder et al. 2013). Further, the prevalence of diminished kidney function among the NHANES 2011–2012 participants of 7.4% exceeds the 5% acceptable disease prevalence (Lin et al. 2014). Thus, restrictive dietary intake guidelines are required to safeguard against a further increase in dietary Cd intake.


Current population risk assessment of dietary Cd intake relies on estimates of dietary Cd intake and/or maintenance of threshold levels of urinary Cd that should protect the kidney from Cd-induced damage. Risk assessment using dietary Cd intake estimates has been questioned because they show only a marginal correlation with urinary Cd levels, a well-founded measure of lifetime intakes. Blood Cd levels, however, show a correlation with urinary Cd levels, and they could thus be of value in risk assessment; blood Cd levels ≥ 1 μg/L were associated with CKD, while blood Cd levels above 0.5 μg/L were associated with AMD, depression, and death from Alzheimer’s disease. Using a Cd-toxicokinetic simulation model, we have found that current tolerable dietary intake guidelines do not contain a safety margin, given that the modeled dietary intake levels exceed the levels associated with kidney damage and many other adverse health outcomes seen in cohorts and cross-sectional studies. These data point to the need for a revision of tolerable dietary intake levels for Cd, and public measures to minimize the food-chain contamination by Cd. Risk reduction measures, supported by international food legislation, should not be relaxed. A maximally permissible concentration (MPC) for Cd in foods should be set as low as reasonably achievable. Current MPC for rice is set at 0.4 mg/kg dry grain weight, but global risk assessment suggests 0.1 mg/kg is necessary. Persistence of Cd in the environment, coupled with its high soil-to-plant transfer rates, requires long-term management of Cd in the environment (soil, air, and water), and in agriculture, where consideration should be given to leafy salad vegetables, such as spinach and lettuce, which are known to be hyper accumulators of Cd. In the absence of non-toxic chelating agents to reduce Cd tissue burden, maintenance of the lowest Cd levels in food crops is pivotal.


Adams SV, Passarelli MN, Newcomb PA. 2012. Cadmium exposure and cancer mortality in the Third National Health and Nutrition Examination Survey cohort. Occup Environ Med 69(2):153–156.

Akerstrom M, Barregard L, Lundh T, Sallsten G. 2013. The relationship between cadmium in kidney and cadmium in urine and blood in an environmentally exposed population. Toxicol Appl Pharmacol 268:286–293.

Arnich N, Sirot V, Rivière G, Jean J, Noël L, Guérin T, et al. 2012. Dietary exposure to trace elements and health risk assessment in the 2nd French Total Diet Study. Food Chem Toxicol 50:2432–2449.

Buser MC, Ingber SZ, Raines N, Fowler DA, Scinicariello F. 2016. Urinary and blood cadmium and lead and kidney function: NHANES 2007–2012. Int J Hyg Environ Health 219(3):261–267.

Ciesielski T, Bellinger DC, Schwartz J, Hauser R, Wright RO. 2013. Associations between cadmium exposure and neurocognitive test scores in a cross-sectional study of US adults. Environ Health 12:13, doi: 10.1186/1476-069X-12-13.

FAO/WHO (Food and Agriculture Organization; World Health Organization). 1989. Evaluation of Certain Food Additives and Contaminants: Thirty-third Report of the Joint FAO/WHO Expert Committee on Food Additives. WHO Technical Report Series No. 776. Geneva:WHO.​/39252/1/WHO_TRS_776.pdf [accessed on 28 January 2017].

FAO/WHO. 1993. Evaluation of Certain Food Additives and Contaminants: Forty-first Report of the Joint FAO/WHO Expert Committee on Food Additives. WHO Technical Report Series No. 837. Geneva:WHO.​/36981/1/WHO_TRS_837.pdf [accessed 28 January 2017].

FAO/WHO. 2010. Seventy-third Meeting, Geneva, 8–17 June 2010. Summary and Conclusions. JECFA/73/SC. Geneva, Switzerland:Food and Agriculture Organization of the United Nations; World Health Organization.​ns/chem/summary73.pdf [accessed 28 January 2017].

Ferraro PM, Costanzi S, Naticchia A, Sturniolo A, Gambaro G. 2010. Low level exposure to cadmium increases the risk of chronic kidney disease: analysis of the NHANES 1999–2006. BMC Public Health 10:304, doi: 10.1186/1471-2458-10-304.

Gallagher CM, Chen JJ, Kovach JS. 2010. Environmental cadmium and breast cancer risk. Aging (Albany NY) 2:804–814.

Hyder O, Chung M, Cosgrove D, Herman JM, Li Z, Firoozmand A, et al. 2013. Cadmium exposure and liver disease among US adults. J Gastrointest Surg 17(7):1265–1273.

IARC (International Agency for Research on Cancer). 1993. Cadmium. IARC Monogr Eval Carcinog Risk Hum 58:119–238.

IPCS (International Programme on Chemical Safety). 1992. Environmental Health Criteria 134: Cadmium. Geneva, WHO.​ehc134.htm [accessed 28 January 2017].

Itoh H, Iwasaki M, Sawada N, Takachi R, Kasuga Y, Yokoyama S, et al. 2014. Dietary cadmium intake and breast cancer risk in Japanese women: a case–control study. Int J Hygiene Environ Health 217:70–77.

Julin B, Wolk A, Bergkvist L, Bottai M, Åkesson A. 2012. Dietary cadmium exposure and risk of postmenopausal breast cancer: a population-based prospective cohort study. Cancer Res 72:1459–1466.

Kjellström T, Nordberg GF. 1978. A kinetic model of cadmium metabolism in the human being. Environ Res 16:248–269.

Lin YS, Caffrey JL, Lin JW, Bayliss D, Faramawi MF, Bateson TF, et al. 2013. Increased risk of cancer mortality associated with cadmium exposures in older Americans with low zinc intake. J Toxicol Environ Health A 76(1):1–15.

Lin YS, Ho WC, Caffrey JL, Sonawane B. 2014. Low serum zinc is associated with elevated risk of cadmium nephrotoxicity. Environ Res 134:133–138.

Mortensen ME, Wong LY, Osterloh JD. 2011. Smoking status and urine cadmium above levels associated with subclinical renal effects in U.S. adults without chronic kidney disease. Int J Hyg Environ Health 214(4):305–310.

Min JY, Min KB. 2016. Blood cadmium levels and Alzheimer’s disease mortality risk in older US adults. Environ Health 15(1):69, doi: 10.1186/s12940-016-0155-7.

Patel CJ, Rehkopf DH, Leppert JT, Bortz WM, Cullen MR, Chertow GM, et al. 2013. Systematic evaluation of environmental and behavioural factors associated with all-cause mortality in the United States National Health and Nutrition Examination Survey. Int J Epidemiol 42(6):1795–1810.

Riederer AM, Belova A, George BJ, Anastas PT. 2013. Urinary cadmium in the 1999–2008 U.S. National Health and Nutrition Examination Survey (NHANES). Environ Sci Technol 47(2):1137–1147.

Ruiz P, Mumtaz M, Osterloh J, Fisher J, Fowler BA. 2010. Interpreting NHANES biomonitoring data, cadmium. Toxicol Lett 198(1):44–48.

Sand S, Becker W. 2012. Assessment of dietary cadmium exposure in Sweden and population health concern including scenario analysis. Food Chem Toxicol 50:536–544.

Satarug S, Baker JR, Reilly PE, Moore MR, Williams DJ. 2002. Cadmium levels in the lung, liver, kidney cortex, and urine samples from Australians without occupational exposure to metals. Arch Environ Health 57:69–77.

Satarug S, Swaddiwudhipong W, Ruangyuttikarn W, Nishijo M, Ruiz P. 2013. Modeling cadmium exposures in low- and high-exposure areas in Thailand. Environ Health Perspect 121:531–536, doi: 10.1289/ehp.1104769.

Scinicariello F, Buser MC. 2015. Blood cadmium and depressive symptoms in young adults (aged 20–39 years). Psychol Med 45(4):807–815.

Tellez-Plaza M, Navas-Acien A, Crainiceanu CM, Sharrett AR, Guallar E. 2010. Cadmium and peripheral arterial disease: gender differences in the 1999–2004 US National Health and Nutrition Examination Survey. Am J Epidemiol 172(6):671–681.

Tellez-Plaza M, Navas-Acien A, Caldwell KL, Menke A, Muntner P, Guallar E. 2012a. Reduction in cadmium exposure in the United States population, 1988–2008: the contribution of declining smoking rates. Environ Health Perspect 120:204–209, doi: 10.1289/ehp.1104020.

Tellez-Plaza M, Navas-Acien A, Menke A, Crainiceanu CM, Pastor-Barriuso R, Guallar E. 2012b. Cadmium exposure and all-cause and cardiovascular mortality in the U.S. general population. Environ Health Perspect 120:1017–1022, doi: 10.1289/ehp.1104352.

Vacchi-Suzzi C, Eriksen KT, Levine K, McElroy J, Tjønneland A, Raaschou-Nielsen O, et al. 2015. Dietary intake estimates and urinary cadmium levels in Danish postmenopausal women. PLoS One 10(9):e0138784, doi: 10.1371/journal.pone.0138784.

Wallia A, Allen NB, Badon S, El Muayed M. 2014. Association between urinary cadmium levels and prediabetes in the NHANES 2005–2010 population. Int J Hyg Environ Health 217(8):854–860.

Wallin M, Sallsten G, Lundh T, Barregard L. 2014. Low-level cadmium exposure and effects on kidney function. Occup Environ Med 71(12):848–854.

Wu EW, Schaumberg DA, Park SK. 2014. Environmental cadmium and lead exposures and age-related macular degeneration in U.S. adults the National Health and Nutrition Examination Survey 2005 to 2008. Environ Res 133:178–184.

An Integrated Experimental Design for the Assessment of Multiple Toxicological End Points in Rat Bioassays

Author Affiliations open
1Cesare Maltoni Cancer Research Center, Ramazzini Institute, Bentivoglio, Bologna, Italy; 2Leonardo da Vinci Programme at the Cesare Maltoni Cancer Research Center, Ramazzini Institute, Bentivoglio, Bologna, Italy; 3National Institute of Environmental Health Sciences, National Institutes of Health, Department of Health and Human Services, Research Triangle Park, North Carolina, USA; 4Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland, USA

PDF icon PDF Version (426 KB)

  • Background: For nearly five decades long-term studies in rodents have been the accepted benchmark for assessing chronic long-term toxic effects, particularly carcinogenicity, of chemicals. The European Food Safety Authority (EFSA) and the World Health Organization (WHO) have pointed out that the current set of internationally utilized test methods capture only some of the potential adverse effects associated with exposures to these agents over the lifetime.

    Objectives: In this paper, we propose the adaption of the carcinogenicity bioassay to integrate additional protocols for comprehensive long-term toxicity assessment that includes developmental exposures and long-term outcomes, capable of generating information on a broad spectrum of different end points.

    Discussion: An integrated study design based on a stepwise process is described that includes the priority end points of the Economic Co-operation and Development and the National Toxicology Program guidelines on carcinogenicity and chronic toxicity and developmental and reproductive toxicity. Integrating a comprehensive set of relevant toxicological end points in a single protocol represents an opportunity to optimize animal use in accordance with the 3Rs (replacement, reduction and refinement). This strategy has the potential to provide sufficient data on multiple windows of susceptibility of specific interest for risk assessments and public health decision-making by including prenatal, lactational, neonatal exposures and evaluating outcomes over the lifespan.

    Conclusion: This integrated study design is efficient in that the same generational cohort of rats used for evaluating long-term outcomes can be monitored in satellite parallel experiments to measure biomarkers and other parameters related to system-specific responses including metabolic alterations and endocrine disturbances.

  • Citation: Manservisi F, Babot Marquillas C, Buscaroli A, Huff J, Lauriola M, Mandrioli D, Manservigi M, Panzacchi S, Silbergeld EK, Belpoggi F. 2017. An integrated experimental design for the assessment of multiple toxicological end points in rat bioassays. Environ Health Perspect 125:289–295;

    Address correspondence to F. Belpoggi, Cesare Maltoni Cancer Research Center, Ramazzini Institute, Via Saliceto, 3, 40010 Bentivoglio, Bologna, Italy. Telephone: 39 051 6640460. E-mail:

    *C. Babot Marquillas is a visiting researcher from the University of Barcelona, Barcelona, Spain.

    We thank L. De Angelis and L. Falcioni of the Cesare Maltoni Cancer Research Center, Ramazzini Institute, for their suggestions and assistance in the development of this integrated model.

    There were no funds received to support the writing or production of this paper.

    The authors declare they have no actual or potential competing financial interests.

    Received: 12 October 2015
    Revised: 27 April 2016
    Accepted: 20 June 2016
    Published: 22 July 2016

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (451 KB)

    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact Our staff will work with you to assess and meet your accessibility needs within 3 working days.

    PDF icon Supplemental Table of Contents PDF (105 KB)


Synthetic chemicals have been used individually and as mixtures in consumer products for over a century, gaining intense momentum beginning after World War II. Naturally occurring elements and compounds have been used for millennia. The first bioassays for identifying chemicals posing a greater and immediate danger for carcinogenicity to individuals were first developed about 100 years ago (Yamagiwa and Ichikawa 1918). The chemical carcinogenesis revolution and testing age began when Yamagiwa and Ichikawa in 1918 showed that coal tar applied to rabbit ears caused skin carcinomas (Yamagiwa and Ichikawa 1918). The real impetus for testing chemicals came with passage of legislation, first in the United States in 1976 and then in several European Union member states, requiring evaluation of industrial chemicals, especially those in the workplace and in consumer products. This led to the development of a multi-national effort to harmonize testing methods through the Environment Programme of the Organization for Economic Co-operation and Development (OECD). Over the last 30 years, many test guidelines were developed within the OECD as well as the concepts for assessing risks of chemicals identified as harmful and carcinogenic in the workplace and environment (Hartung 2009; Huff 1992; Maltoni 1976; Soffritti et al. 2002; Tomatis 1979; Silbergeld et al. 2015). Rodent bioassays have been described in the OECD Test Guideline (TG) 453 (OECD 2009) and by the U.S. National Toxicology Program (NTP 2011b), with specifications for design and conduct of studies to evaluate toxic and carcinogenic potential of chemical, biological and physical agents in laboratory animals. Recognizing that carcinogenesis is a multi-step, multivariate process (Brash and Cairns 2009; Hanahan and Weinberg 2011), it may be unrealistic to expect a basic 2-year cancer study to provide all the complex data necessary for cancer risk identification, management, and regulatory decisions. Current OECD guidelines (OECD 2009), as planned, are not aimed to monitor cancer hazards and risks of exposure on susceptible individuals such as children and the elderly. For some test articles, NTP carcinogenicity 2-year protocol might include perinatal exposure, but these are selected only after considering patterns of human exposure (NTP 2011b, 2016). Furthermore, traditional toxicity testing methods could not identify many of the endocrine-related adverse effects of some chemicals, especially subtle effects on specific developmental stages (Bergman et al. 2012, 2015; Birnbaum 2013; Huff 1996; Huff et al. 1996; Manservisi et al. 2015; Melnick et al. 2002; Vandenberg et al. 2012), as happened for bisphenol A (Maffini et al. 2006; Vandenberg et al. 2009; vom Saal et al. 2007). Consistent with these considerations, both OECD and NTP have introduced new guidelines for reproductive and developmental toxicity with more functional end points to assess how agents affect the reproductive and endocrine status of animals (NTP 2011a; OECD 2011).

Study designs and outcomes investigated by current guidelines and our proposed protocol on carcinogenicity and chronic toxicity and reproductive and developmental toxicity are summarized in Table 1. The OECD reference guideline for reproductive and developmental toxicity, OECD TG 443 (Extended One-Generation Reproductive Toxicity Study), provides an evaluation of reproductive and developmental effects that may occur in offspring as a result of pre- and post-natal chemical exposure as well as systemic toxicity in pregnant and lactating females (OECD 2011). In the OECD TG 443 protocol, sexually mature male and female rodents [parental (P) generation] are exposed to graduated doses of test substances starting 2 weeks before mating and continued through mating, gestation, lactation, and weaning of pups (F1 generation). At weaning, pups are assigned to three groups for reproductive and developmental toxicity testing (cohort 1), developmental neurotoxicity testing (cohort 2), and developmental immunotoxicity testing (cohort 3). Other F1 offspring are exposed after weaning through adulthood. Clinical observations and pathology examinations are performed on all animals for signs of toxicity, with special emphasis on integrity and performance of male and female reproductive systems and health, growth, development, and function of offspring. Part of cohort 1 (cohort 1B) may be extended to include an F2 generation: In this case, procedures for F1 animals are similar to those for the P animals. The total number of animals involved in this OECD protocol design is more than one thousand (OECD 2011).

Table 1. Select View Table (HTML Version) for a 508-conformant version

Table 1. Comparison between existing NTP MOG and OECD guidelines and the Ramazzini Institute (RI) proposed study design.

View Table (HTML Version)
View larger image (TIF File)

The NTP reference guideline for reproductive and developmental toxicity, the NTP’s Modified One-Generation (MOG) Reproduction Study (NTP 2011a), employs pregnant animals with exposures beginning at implantation with continued dosing of dams throughout gestation and lactation (Foster 2014). At weaning, offspring are administered the test substance at the same level as their respective dams and are assigned to different cohorts: a prechronic toxicity cohort (analogous to a standard 90-day study) for evaluating clinical pathology and target organ toxicity and pathology; a teratology cohort for evaluating prenatal development; another cohort to evaluate breeding and littering for potential examination of the subsequent generation. This study design involves exposure of pregnant females throughout gestation (the P generation), lifetime exposure of the F1, and generation of two cohorts of F2 animals (developmental and reproductive).

The OECD TG 443 and the NTP MOG were introduced only recently, and there is still no published data comparing studies with the same substance according to the two guidelines. We cannot exclude the possibility that authorities such as the U.S. Environmental Protection Agency (EPA), the U.S. Food and Drug Administration (FDA), the European Chemical Agency (ECHA), and the European Food Safety Agency (EFSA) could require (or have already required) the repetition of the tests with both guidelines considering the need for empirical evidence supporting the use of one of the two. It is our opinion that regularly studying the same substance with both the NTP MOG and OECD TG 443 represents an unnecessary repetition. The NTP’s MOG is able to generate large and robust data sets that include early-life exposure and teratogenicity, but requires a larger number of animals than the OECD TG 443 (Schiffelers et al. 2015; Foster 2014).

Starting from the 1990s, the Cesare Maltoni Cancer Research Centre (CMCRC) of the Ramazzini Institute (RI) performed carcinogenicity studies on low doses of chemical or physical agents that may expose millions or even billions of people to potential carcinogenic risks, such as radiations and food additives (Maltoni et al. 1985, 1999; Soffritti et al. 1999, 2002, 2007, 2008), using an alternative model, more sensitive than the traditional combined carcinogenicity and chronic toxicity 2-year protocol adopted by OECD and NTP (Bucher 2002; Huff 1992; Melnick et al. 2008). The CMCRC protocol includes prolonged periods of exposure and observation of experimental animals and starting exposures from the 12th day of fetal life (gestation) and continuing through lactation and weaning until at least 130 weeks or longer (Soffritti et al. 2002). In fact, human exposures to environmental agents, also at relatively low doses, most often starts prior to and during mother’s gestation, continues through lactation (via breast milk) and lasts until death. In standard bioassays, exposure generally starts in young adulthood and lasts until about 2 years, which is roughly equivalent to only 65 years in humans (Maltoni et al. 1997; Haseman et al. 2001; Huff et al. 2008; Melnick et al. 2008). Group sizes in carcinogenicity studies should also be increased whenever required for sufficient statistical power and to avoid the possibility of false negative response: Bioassays involving 100 animals or more per sex per group might be necessary for identifying carcinogenic effects of low doses and weak carcinogenic activity (Maltoni et al. 1981; McCormick 2013). More than 500 chemical-specific bioassays have been performed at CMCRC, and the results are used worldwide for hazard identification and human cancer risk assessments (NRC 2014a, 2014b).

To satisfy the need to consider multiple effects (e.g., cancer and noncancer) across multiple life stages and to reduce the overall number of animals required for separate studies of these end points, we propose the following experimental design that integrates traditional cancer guidelines with more recent proposals of OECD and NTP for studying reproductive and developmental toxicity. This new integrated experimental design aims to maximize the end points measured for each animal, thus reducing the overall number of animals produced and utilized, in accordance with the 3Rs (replacement, reduction and refinement) (European Union 2010).

The central aim of the methodology proposed in the Integrated Long-Term Toxicity and Carcinogenicity Study is to maximize the breadth of outcomes assessed and to increase the sensitivity of testing beyond that in commonly used protocols to give more reliable and inclusive information on many important end points (Figure 1).

Figure 1. Chart illustrating the numbers of animals (by sex and total) and the treatment schedule for each study group.

Figure 1. Integrated Long-Term Toxicity and Carcinogenicity Study experimental design. Schedule for treatment and duration for each group. Note: ////, continuous treatment; IIII, no treatment (period without dosing); F2, second generation offspring; m, mating; total animals/group, studying at least three exposure groups plus controls, the number for a comprehensive human equivalent hazard identification study is 1,720 animals; WOS, windows of susceptibility.

View larger image (TIF File)

Our Proposal: An Integrated Experimental Design

The integrated experimental design proposed by the CMCRC/RI is outlined in Figure 1 and more details on each specific section of the protocol are available in the Supplemental Material, “Ramazzini Institute’s proposal for Integrated Long-Term Toxicity/Carcinogenicity Study.” The study design is largely based on OECD TG 453 (modified only for duration of the experiment), OECD TG 443, NTP Guidelines. The study comprises the following components:

Carcinogenicity and chronic toxicity study. Animals are treated from fetal life (dams, 12th day of pregnancy) until 104 weeks of age, then observed (with or without continuous exposure, depending on chemical) until 130 weeks of age (30 months). Interim kills are included to provide information on progression of non-neoplastic or neoplastic changes and mechanistic information (e.g., gene expression, serum biomarkers of inflammation, cell proliferation). Animals included for interim evaluation are also exposed from fetal life (dams, 12th day of pregnancy) until 26, 52, 78, and 104 weeks of age following OECD guidelines (OECD 2009).

Reproductive and developmental toxicity. Different windows of susceptibility (WOS) related to reproductive and developmental and other noncancer effects are studied. The possible adverse effects of the substances are studied in prenatal, neonatal, prepubertal, pubertal, and adult parous and nulliparous WOS and compared among them, or with the possible long-term carcinogenicity effect.

Animal Model

The laboratory rat has served as the traditional animal model of choice for research and regulatory developmental and reproductive toxicity testing conducted to support human health hazard identification and risk assessment. The rat has been used extensively for developmental and reproductive physiology and endocrinology research and has been more thoroughly characterized in these research fields than other species, likewise for identifying likely human carcinogens (Gray et al. 2004; Maltoni et al. 1999; Teitelbaum et al. 2015).

Our proposal to use Sprague-Dawley (SD) rats is based on the evidence that they are adequately sensitive, have a long history of being used in research studies, and are also recommended by the OECD (2009, 2011) and the NTP (King-Herbert and Thayer 2006; King-Herbert et al. 2010) and are used by many universities and organizations (Manservisi et al. 2015). SD rats are known and accepted as a human-equivalent model for cancer (Teitelbaum et al. 2015; Soffritti et al. 2006). The proposed protocol uses SD rat strains that meet the requirement of the OECD 443 and 453 guidelines: “strains with low fecundity or a well-known high incidence of spontaneous developmental defects should not be used” (OECD 2011) and “using a strain of animal that has an acceptable survival rate for the long-term study” (OECD 2009).

There are known limits for this animal model for individual cancer end points. For example, SD rats represent an optimal model for breast cancer research (Teitelbaum et al. 2015), while the high prevalence of benign tumors of the pituitary gland and pheocromocytoma of the subrenal gland make SD rats an inappropriate model for tumors of these organs (Dinse et al. 2010).

Numbers of Animals

There is widespread agreement that the relatively small numbers of animals used in most standard toxicity tests is a serious issue in terms of sensitivity and reliability. On the other hand, there are social and ethical concerns about the number of animals used in these tests (Hartung and Rovida 2009). Inadequate tests are a main driver of additional testing, such that it can be argued that utilizing robust methods, with increased numbers of animals per test, will reduce overall animal testing. Current guidelines recommend study designs which encompass at least three treatment groups plus control. For the OECD TG 453 carcinogenicity and chronic toxicity protocol the minimal number of animals is 480; for the OECD TG 443, the minimal number is 1,760 and for the NTP MOG Reproduction Study, it is 3,200 animals (Table 1). But because only a limited number of end points are assessed in each of these tests, more animals are expected to be required to empower a broad-based toxicological evaluation for hazard and risk assessment. Performing these studies separately, as is current practice, would require up to 3,680 animals (Table 1).

In our proposal, breeders (virgin males and females) of about 10–15 weeks of age are matched in a single outbred mating, in a number adequate to obtain sufficient animals for the study. The objective of breeding is to generate animals in order to have no more than one sister and one brother for each control and exposed group (two sisters and two brothers in the carcinogenicity arm) in order to avoid any bias due to familial relationship.

Studying at least three exposure groups plus controls, the number for a comprehensive human equivalent hazard identification study is 1,720 animals (Figure 1 and Table 1). A higher number of exposed and control animals included in the studies better guarantees higher sensitivity of the model, sufficient statistical power, and overall saving animals that would be sacrificed in unnecessary repetition of the studies or performing uninformative underpowered studies (Hooijmans et al. 2010).

In compliance with the 3Rs, we suggest, whenever possible, to avoid the use of culling and to use all the pups generated during the experiment, avoiding unnecessary sacrifice of animals. It is our opinion that avoiding culling also would permit generally a more rigorous measure of litter mortality and simulate a human equivalent scenario, with more genetic variability and avoiding possible selection bias (for example selecting only healthy animals with higher birth weight). Nevertheless, the use of culling might be appropriate for studying suspect endocrine disruptor substances, as litter size can impact the weights and the growth rate of the pups, which can affect the timing of puberty. Puberty timing regulates other end points, so that the change in body weight from not equalizing litter size early on might have an inadvertent impact on the study.

Dose Ranges

Under current testing procedures (Maronpot et al. 2004), when toxicology studies are performed, relatively high doses of a chemical are given to animals, generally higher than the doses humans are exposed to. However this is not always the case, especially for various workplaces and occupations and high-dose drug and cancer chemotherapies. Toxicity testing is typically carried out with maximum tolerated dose (MTD), previously determined in shorter-term exposures experiments of 28–90 days. Toxicology studies of higher doses show that a chemical can be lethal (and needs to be avoided), or block or disrupt pregnancies, or induce birth defects. These high-dose effects may not always be observed at lower doses, which is why some assume that these are safe exposures, but there may be other end points affected, that cannot be detected by typical methods of a standard bioassay (Teitelbaum et al. 2015). Non-monotonic dose–response curves reveal such unexpected effects, especially for endocrine disrupting chemicals (EDCs) (e.g., plasticizers, pesticides, and other industrial chemicals) as shown by several toxicological and epidemiologic studies on noncancer end points that are relevant to metabolic disease (Thayer et al. 2005, 2006; Vandenberg et al. 2012). In the multitude of chemicals that have never been tested adequately at low doses but were already tested for carcinogenicity at high doses, we suggest testing doses in the range of actual highest human exposure, setting the LOAEL (lowest observed adverse effect level) from traditional toxicological studies as the highest dose, particularly in experiments designed to test endocrine-sensitive end points. For chemicals never tested for long-term carcinogenic effects, at least one high-dose group near the MTD should be included, obviating the problem of unnecessary repetition of the bioassay if the low dose protocol is not carcinogenic.

Estimation of daily intake of a test substance depends on knowledge of the toxicokinetics, including route of administration, distribution, metabolism, and excretion, which are not all readily available from the literature (Søeborg et al. 2014). If a range of doses is unavailable or unknown, we propose that a dose-range finding (DRF) should be performed before starting the experimental protocol in order to determine an optimal exposure concentration for each chemical selected as close as possible to the estimated human exposure; in particular, when novel food (European Commission 2017) or similar test compounds are studied, nutritional aspects and other relevant methodological aspects (e.g., bioavailability, food metabolism that might differ in rodents and humans, stability of the test compound) related to exposure should be studied (EFSA 2013). When conducting exposure studies with low doses [many orders of magnitude lower than the no observed adverse effect level (NOAEL)], a systematic dose-calibration study should be performed in an appropriate rodent model in order to identify the administered oral dose of the test substance that results in biomarker concentrations (e.g., urine, serum) comparable to the ones observed in human population (Teitelbaum et al. 2016). Of course other higher doses must be chosen to adequately challenge biological systems and to provide some observable indication of toxicity, without jeopardizing the health and well-being or the body weights and survival of exposed animals, as well as being optimally sensitive to adequately evaluate the potential carcinogenicity (Bucher 2000; Huff 1999; Melnick et al. 2008). Higher doses also increase a priori statistical power to detect noncancer effects using a relatively small number of animals, although remarkable exceptions exist particularly for endocrine effects (Vandenberg et al. 2012).

Timing of Exposure

Adult exposure to some chemicals is certainly an important factor in adverse health outcomes; however, increased focus on the fetus and neonate is of primary concern since developing organisms are extremely sensitive to perturbations by chemicals, especially those with hormone-like activity. Certain types of adverse effects may be more severe in developing organisms and occur at chemical concentrations that are in some instances below levels that would be considered harmful in adults (Tabb and Blumberg 2006). Few guidelines for testing environmental chemicals include prenatal or early-life exposures, and thus often do not provide information on risks of carcinogens related to early-life exposure (Rudel et al. 2011; Tabb and Blumberg 2006). Based on results of long-term carcinogenicity bioassays testing chemical and physical agents using rodents, there is ample evidence demonstrating that exposures during early developmental phases produce an overall increase of malignant tumors and increases of specific organ site neoplasms related to exposures to specific carcinogens as in the case of vinyl chloride and benzene (Maltoni et al. 1981, 1989; Huff et al. 2008; Soffritti et al. 2008). Early exposure to chemicals is particularly important in study designs if there is reason to believe human exposures begin in utero and that susceptibility may be greater during growth and early developmental stages (Rice et al. 1989).

For a clear understanding of this protocol, it should be considered that 16 weeks of age in adult rats roughly correspond to 10 years of age in human years (Sengupta 2013). In our proposal, animals belonging to the chronic toxicity and carcinogenicity study arm are observed until 130 weeks of age (corresponding to about 75–80 years of age in humans), starting exposure during fetal life (dams, 12th day of pregnancy), whereas OECD guidelines stipulate that animals should be killed and examined at 104 weeks of treatment (corresponding to about 60–65 years of age in humans) (Huff et al. 2008). Interim kills are also planned following the OECD TG 453 to provide information on the progression of non-neoplastic events and neoplastic changes and mechanistic information.

The Reproductive and Developmental Toxicity Study arm mimics human exposure during critical windows of development, and includes a) prenatal (F1) animals treated during embryonic life and sacrificed at postnatal day (PND) 21; b) postnatal (F1) animals treated through lactation, starting from birth (PND 1) and sacrificed at PND 21; c) prepubertal (F1) animals treated from PND 21 to PND 42; d) pubertal (F1) animals treated from PND 42 to PND 63; and e) adult parous and nulliparous (F1) female animals treated from PND 1 through lactation, until PND 181 (Figure 1). At 10–15 weeks, the parous group rats are mated (outbred), and chemical treatment is continued through pregnancy, delivery of pups (F2), and lactation. At the time of sacrifice of parous rats on PND 181, F2 pups had completed weaning.

In order to verify or elucidate effects in second generation, F2 offspring generated from F1 adult parous female rats are examined and sacrificed on PND 28.

During necropsy, frozen target tissues (including blood) and organs, together with paraffin-embedded tissues, are stored for histopathology and molecular biology studies, EDCs effects, neurotoxicity, biochemical and biohematological changes (metabolism), and toxic and preneoplastic lesions.

Additional End Points and Adverse Effects of the Test Compound

The aim of our integrated experimental design was to investigate all or a majority of possible health effects related to exposure to the studied agent and to minimizing the unnecessary use of experimental animals. Our design also avoids wasted time when doing sequential end point studies. End points assessed in traditional toxicology and carcinogenicity testing are food and water consumption, chemical exposure, weight loss and gain, clinical pathology, survival and mortality, changes in organ weight, preneoplastic and neoplastic diseases with histopathological analyses. However, many examined chemicals have shown to also cause complex effects in animals, affecting organ development and functional and behavioral changes (Vandenberg et al. 2012). To best evaluate these fundamental end points, we included in our protocol design several of the NTP MOG and OECD TG 443 end points for immunotoxicity, neurotoxicity, and developmental and reproductive toxicity. It should be noted that this protocol is easily scalable (e.g., additional groups can be added if appropriate, or specific arms can be amended if previously investigated) and simple changes are feasible and would permit to target specific end points or tissues (for example sperm aneuploidy) that are not described in this proposal.


In our proposed lifetime experimental design, we assess a range of adverse outcomes of interest using a relatively large population of animals (sufficient power), born at the same time after mating of outbred breeders, randomized and studied for dose-related effects, with the lowest possible risk of bias (blinding of assessors of outcomes, randomization, blinded assessment of pathological lesions by a minimum of two assessors). Typically, for studying all the previously mentioned parameters (WOS, fertility, development, toxicity, carcinogenicity), approximately 10–20 studies are performed, using more animals, in different laboratories, with different procedures. Our experimental model and design overcomes these deficiencies and permits more information to be gathered on toxic, mechanistic, and biological parameters, using the same but fewer overall animals in a large but unique experiment. In fact, in our experimental design, rats from the same generation are used for studying chronic toxicity and carcinogenicity outcomes and distributed in satellite parallel experiments (WOS), thus minimizing variables between different arms of the multi-end point investigation, for detecting also reproductive/developmental toxicity.

Our integrated experimental protocol requires 1,720 animals, with a reduction up to 53% in animal use as compared to using separate test protocols (Table 1), representing an opportunity for investigating multiple toxicological end points at once, sparing animal lives in accordance with the 3Rs. We also expected an important reduction in terms of time, because the realization of a single integrated experiment would take a shorter time for design, approval, performance, and analysis if compared with multiple and sequential ones, which, in turn, would reduce costs and improve the availability of data for risk assessment.

The protocol we suggest addresses several important issues in the application of toxicological research to human health risk assessment including information on different toxicological outcomes of exposures and health hazards of importance to human populations that are currently not completely covered by standard test protocols; earlier initiation and longer duration of exposure and observation of animals (130 weeks of age instead of 110) for a more comprehensive analysis of potential effects of chemical exposures and outcome assessment; enabling interim analyses and other strategies to examine specific outcomes over the lifespan. For increased efficiency, results of these tests can be shared among laboratories. Ideally the in vivo biophase should be the responsibility of one laboratory in order to favor consistency of methods and quality of long-term animal studies (Gift et al. 2013). After the biophase, various end points, parameters, findings, and information on each category might be evaluated by different topic-expert scientists and laboratories. Frozen tissue samples from target organs are stored in order to study mechanistic aspects of the toxic process. Other relevant evidence, including cellular and molecular analyses related to mechanisms can be included in experimental designs, as has been proposed for the forthcoming OECD and NTP integrated guidelines regarding long-term in vivo studies (Darzynkiewicz et al. 2011; Kissling et al. 2007; Recio et al. 2010; Witt et al. 2008).


This protocol represents a proposal to regulatory scientists and the scientific community in general.

Compared to other OECD and NTP guidelines, this protocol has the unique feature of integrating carcinogenicity, toxicity and reproductive and developmental toxicity end points in a single protocol, with animals of the same generation, exploring windows of susceptibility that are currently not addressed in the other guidelines design. The design and protocol discussed here requires validation in order to demonstrate that the combined test is feasible and is at least as good as the separate tests (OECD 2005). Experience in the application of this proposal will be required in order to reach the same level of confidence that has been achieved for the standard carcinogenicity bioassays (Huff 2010). A priori establishment of criteria and consensus on relevant end points of interest is also a good starting point for evidence-based evaluations and following systematic review of obtained results (Birnbaum et al. 2013; Mandrioli and Silbergeld 2016). This is clearly needed, for example, for testing endocrine-active substances with multiple end points, as well as modes and mechanisms of action, as the most reliably predictive animal model has yet to be identified. With this protocol, we aim to produce robust data sets that could also support the validation and discrimination of consensus criteria for evaluating noncancer outcomes, such as endocrine disruption.

We propose that conducting such integrated bioassays could enhance and expand scientific evidence for risk assessments, gathering sufficient and rapid information on several adverse effects in a unique protocol for protecting public health (Robinson 2012).


Bergman Å, Becher G, Blumberg B, Bjerregaard P, Bornman R, Brandt I, et al. 2015. Manufacturing doubt about endocrine disrupter science—a rebuttal of industry-sponsored critical comments on the UNEP/WHO report “State of the Science of Endocrine Disrupting Chemicals 2012.” Regul Toxicol Pharmacol 73:1007–1017.

Bergman Å, Heindel JJ, Jobling S, Kidd KA, Zoeller RT, eds. 2012. State of the Science of Endocrine Disrupting Chemical – 2012: An Assessment of the State of the Science of Endocrine Disruptors Prepared by a Group of Experts for the United Nations Environment Programme and World Health Organization.​f [accessed 10 April 2016].

Birnbaum LS. 2013. State of the science of endocrine disruptors. Environ Health Perspect 121:A107, doi: 10.1289/ehp.1306695.

Birnbaum LS, Thayer KA, Bucher JR, Wolfe MS. 2013. Implementing systematic review at the National Toxicology Program: status and next steps. Environ Health Perspect 121:A108–A109, doi: 10.1289/ehp.1306711.

Brash D, Cairns J. 2009. The mysterious steps in carcinogenesis. Br J Cancer 101:379–380.

Bucher JR. 2000. Doses in rodent cancer studies: sorting fact from fiction. Drug Metab Rev 32:153–163.

Bucher JR. 2002. The National Toxicology Program rodent bioassay: designs, interpretations, and scientific contributions. Ann N Y Acad Sci 982:198–207.

Darzynkiewicz Z, Smolewski P, Holden E, Luther E, Henriksen M, François M, et al. 2011. Laser scanning cytometry for automation of the micronucleus assay. Mutagenesis 26(1):153–161.

Dinse GE, Peddada SD, Harris SF, Elmore SA. 2010. Comparison of NTP historical control tumor incidence rates in female Harlan Sprague Dawley and Fischer 344/N Rats. Toxicol Pathol 38(5):765–775.

EFSA (European Food Safety Authority). 2013. Considerations on the applicability of OECD TG 453 to whole food/feed testing. EFSA Journal 11(7):3347, doi: 10.2903/j.efsa.2013.3347.

European Commission. 2017. Novel Food.​od_en [accessed 5 January 2017].

European Union. 2010. Directive 2010/63/EU of the European Parliament and of the Council of 22 September 2010 on the protection of animals used for scientific purposes. Official Journal of the European Union L 276/33.

Foster PM. 2014. Regulatory Forum opinion piece: new testing paradigms for reproductive and developmental toxicity—the NTP modified one generation study and OECD 443. Toxicol Pathol 42:1165–1167.

Gift JS, Caldwell JC, Jinot J, Evans MV, Cote I, Vandenberg JJ. 2013. Scientific considerations for evaluating cancer bioassays conducted by the Ramazzini Institute. Environ Health Perspect 121(11–12):1253–1263, doi: 10.1289/ehp.1306661.

Gray LE Jr, Wilson V, Noriega N, Lambright C, Furr J, Stoker TE, et al. 2004. Use of the laboratory rat as a model in endocrine disruptor screening and testing. ILAR J 45:425–437.

Hanahan D, Weinberg RA. 2011. Hallmarks of cancer: the next generation. Cell 144:646–674.

Hartung T. 2009. Toxicology for the twenty-first century. Nature 460:208–212.

Hartung T, Rovida C. 2009. Chemical regulators have overreached. Nature 460:1080–1081.

Haseman J, Melnick R, Tomatis L, Huff J. 2001. Carcinogenesis bioassays: study duration and biological relevance. Food Chem Toxicol 39:739–744.

Hooijmans CR, Leenaars M, Ritskes-Hoitinga M. 2010. A gold standard publication checklist to improve the quality of animal studies, to fully integrate the Three Rs, and to make systematic reviews more feasible. Altern Lab Anim 38:167–182.

Huff J. 1992. Design strategies, results and evaluations of long-term chemical carcinogenesis studies. Scand J Work Environ Health 18(suppl)1:31–37.

Huff J. 1996. Chemically induced cancers in hormonal organs of laboratory animals and of humans. Prog Clin Biol Res 394:77–102.

Huff J. 1999. Animal and human carcinogens [Letter]. Environ Health Perspect 107:A341–A342.

Huff J. 2010. Predicting chemicals causing cancer in animals as human carcinogens. Occup Environ Med 67:720, doi: 10.1136/oem.2009.054569.

Huff J, Boyd J, Barrett JC. 1996. Hormonal carcinogenesis and environmental influences: background and overview. Prog Clin Biol Res 394:3–23.

Huff J, Jacobson MF, Davis DL. 2008. The limits of two-year bioassay exposure regimens for identifying chemical carcinogens. Environ Health Perspect 116:1439–1442, doi: 10.1289/ehp.10716.

King-Herbert AP, Sills RC, Bucher JR. 2010. Commentary: update on animal models for NTP studies. Toxicol Pathol 38:180–181.

King-Herbert A, Thayer K. 2006. NTP workshop: animal models for the NTP rodent cancer bioassay: stocks and strains—should we switch? Toxicol Pathol 34:802–805.

Kissling GE, Dertinger SD, Hayashi M, MacGregor JT. 2007. Sensitivity of the erythrocyte micronucleus assay: dependence on number of cells scored and inter-animal variability. Mutat Res 634(1):235–240.

Maffini MV, Rubin BS, Sonnenschein C, Soto AM. 2006. Endocrine disruptors and reproductive health: the case of bisphenol-A. Mol Cell Endocrinol 254–255:179–186.

Maltoni C. 1976. Occupational carcinogenesis. Predictive value of carcinogenesis bioassays. Ann N Y Acad Sci 271:431–443.

Maltoni C, Ciliberti A, Cotti G, Conti B, Belpoggi F. 1989. Benzene, an experimental multipotential carcinogen: results of the long-term bioassays performed at the Bologna Institute of Oncology. Environ Health Perspect 82:109–124.

Maltoni C, Ciliberti A, Pinto C, Soffritti M, Belpoggi F, Menarini L. 1997. Results of long-term experimental carcinogenicity studies of the effects of gasoline, correlated fuels, and major gasoline aromatics on rats. Ann N Y Acad Sci 837:15–52.

Maltoni C, Conti B, Cotti G, Belpoggi F. 1985. Experimental studies on benzene carcinogenicity at the Bologna Institute of Oncology: current results and ongoing research. Am J Ind Med 7:415–446.

Maltoni C, Lefemine G, Ciliberti A, Cotti G, Carretti D. 1981. Carcinogenicity bioassays of vinyl chloride monomer: a model of risk assessment on an experimental basis. Environ Health Perspect 41:3–29.

Maltoni C, Soffritti M, Belpoggi F. 1999. The scientific and methodological bases of experimental studies for detecting and quantifying carcinogenic risks. Ann N Y Acad Sci 895:10–26.

Mandrioli D, Silbergeld EK. 2016. Evidence from toxicology: the most essential science for prevention. Environ Health Perspect 124:6–11, doi: 10.1289/ehp.1509880.

Manservisi F, Gopalakrishnan K, Tibaldi E, Hysi A, Iezzi M, Lambertini L, et al. 2015. Effect of maternal exposure to endocrine disrupting chemicals on reproduction and mammary gland development in female Sprague-Dawley rats. Reprod Toxicol 54:110–119.

Maronpot RR, Flake G, Huff J. 2004. Relevance of animal carcinogenesis findings to human cancer predictions and prevention. Toxicol Pathol 32(suppl 1):40–48.

McCormick DL. 2013. Preclinical evaluation of carcinogenicity using the rodent two-year bioassay. In: A Comprehensive Guide to Toxicology in Preclinical Drug Development. Amsterdam, Netherlands:Elsevier, 423–436.

Melnick R, Lucier G, Wolfe M, Hall R, Stancel G, Prins G, et al. 2002. Summary of the National Toxicology Program’s report of the endocrine disruptors low-dose peer review. Environ Health Perspect 110:427–431.

Melnick RL, Thayer KA, Bucher JR. 2008. Conflicting views on chemical carcinogenesis arising from the design and evaluation of rodent carcinogenicity studies. Environ Health Perspect 116:130–135, doi: 10.1289/ehp.9989.

NRC (National Research Council). 2014a. Review of the Formaldehyde Assessment in the National Toxicology Program 12th Report on Carcinogens. Washington, DC:National Academies Press.

NRC. 2014b. Review of the Styrene Assessment in the National Toxicology Program 12th Report on Carcinogens. Washington, DC:National Academies Press.

NTP (National Toxicology Program). 2011a. NTP’s Modified One-Generation Studies.​mog/index.html [accessed 10 February 2016].

NTP. 2011b. Specifications for the Conduct of Studies to Evaluate the Toxic and Carcinogenic Potential of Chemical, Biological and Physical Agents in Laboratory Animals for the National Toxicology Program (NTP). Research Triangle Park, NC:National Institutes of Health, National Institute of Environmental Health Sciences, NTP.​finalntp_toxcarspecsjan2011.pdf [accessed 5 January 2017].

NTP. 2016. Toxicology/Carcinogenicity.​artox/index.html [accessed 10 February 2016].

OECD (Organization for Economic Co-operation and Development). 2005. Guidance Document on the Validation and International Acceptance of New or Updated Test Methods for Hazard Assessment. OECD Series On Testing And Assessment 34. ENV/JM/MONO(2005)14. Paris, France:OECD Publishing.​s/feddocs/oecd/oecd-gd34.pdf [accessed 31 December 2016].

OECD. 2009. Test No. 453: Combined Chronic Toxicity/Carcinogenicity Studies. OECD Guidelines for the Testing of Chemicals, Section 4: Health Effects. Paris, France:OECD Publishing.​en [accessed 31 December 2016].

OECD. 2011. Test No. 443: Extended One-Generation Reproductive Toxicity Study. OECD Guidelines for the Testing of Chemicals, Section 4: Health Effects. Paris, France:OECD Publishing.​en [accessed 31 December 2016].

Recio L, Hobbs C, Caspary W, Witt KL. 2010. Dose–response assessment of four genotoxic chemicals in a combined mouse and rat micronucleus and comet assay protocol. J Toxicol Sci 35(2):149–162.

Rice JM, Rehm S, Donovan PJ, Perantoni AO. 1989. Comparative transplacental carcinogenesis by directly acting and metabolism-dependent alkylating agents in rodents and nonhuman primates. IARC Sci Publ 96:17–34.

Robinson C. 2012. Time to bring risk assessment into the real world. Research Europe. London, UK; 26 January 2012.​ire_Robinson3/publication/258412098_Time​_to_bring_risk_assessment_into_the_real_​world/links/00b495282726336f64000000.pdf​?origin=publication_detail [accessed 31 December 2016].

Rudel RA, Fenton SE, Ackerman JM, Euling SY, Makris SL. 2011. Environmental exposures and mammary gland development: state of the science, public health implications, and research recommendations. Environ Health Perspect 119:1053–1061, doi: 10.1289/ehp.1002864.

Schiffelers MJ, Blaauboer BJ, Bakker WE, Hendriksen CF, Krul C. 2015. Regulatory acceptance and use of the Extended One Generation Reproductive Toxicity Study within Europe. Regul Toxicol Pharmacol 71(1):114–124.

Sengupta P. 2013. The laboratory rat: relating its age with human’s. Int J Prev Med 4:624–630.

Silbergeld EK, Mandrioli D, Cranor CF. 2015. Regulating chemicals: law, science, and the unbearable burdens of regulation. Annu Rev Public Health 36:175–191.

Søeborg T, Frederiksen H, Andersson AM. 2014. Considerations for estimating daily intake values of nonpersistent environmental endocrine disruptors based on urinary biomonitoring data. Reproduction 147:455–463.

Soffritti M, Belpoggi F, Minardi F, Bua L, Maltoni C. 1999. Mega-experiments to identify and assess diffuse carcinogenic risks. Ann N Y Acad Sci 895:34–55.

Soffritti M, Belpoggi F, Minardi F, Maltoni C. 2002. Ramazzini Foundation cancer program: history and major projects, life-span carcinogenicity bioassay design, chemicals studied, and results. Ann N Y Acad Sci 982:26–45.

Soffritti M, Belpoggi F, Degli Esposti D. 2006. Cancer prevention: the lesson from the lab. In: Cancer Medicine at the Dawn of the 21st Century: The View from Bologna. Biasco G, Tanneberger S, eds. Bologna, Italy:Bononia University Press, 49–64.​ads/2008/03/Cancer-Prevention_the-lesson​-from-the-lab_2006.pdf [accessed 18 April 2016].

Soffritti M, Belpoggi F, Tibaldi E, Esposti DD, Lauriola M. 2007. Life-span exposure to low doses of aspartame beginning during prenatal life increases cancer effects in rats. Environ Health Perspect 115:1293–1297, doi: 10.1289/ehp.10271.

Soffritti M, Belpoggi F, Esposti DD, Falcioni L, Bua L. 2008. Consequences of exposure to carcinogens beginning during developmental life. Basic Clin Pharmacol Toxicol 102:118–124.

Tabb MM, Blumberg B. 2006. New modes of action for endocrine-disrupting chemicals. Mol Endocrinol 20:475–482.

Teitelbaum SL, Belpoggi F, Reinlib L. 2015. Advancing research on endocrine disrupting chemicals in breast cancer: expert panel recommendations. Reprod Toxicol 54:141–147.

Teitelbaum SL, Li Q, Lambertini L, Belpoggi F, Manservisi F, Falcioni L, et al. 2016. Paired serum and urine concentrations of biomarkers of diethyl phthalate, methyl paraben and triclosan in rats. Environ Health Perspect 124:39–45, doi: 10.1289/ehp.1409586.

Thayer KA, Melnick R, Burns K, Davis D, Huff J. 2005. Fundamental flaws of hormesis for public health decisions. Environ Health Perspect 113:1271–1276, doi: 10.1289/ehp.7811.

Thayer KA, Melnick R, Huff J, Burns K, Davis D. 2006. Hormesis: a new religion? [Letter]. Environ Health Perspect 114:A632–A633.

Tomatis L. 1979. The predictive value of rodent carcinogenicity tests in the evaluation of human risks. Annu Rev Pharmacol Toxicol 19:511–530.

Vandenberg LN, Maffini MV, Sonnenschein C, Rubin BS, Soto AM. 2009. Bisphenol-A and the great divide: a review of controversies in the field of endocrine disruption. Endocr Rev 30:75–95.

Vandenberg LN, Colborn T, Hayes TB, Heindel JJ, Jacobs DR Jr, Lee DH, et al. 2012. Hormones and endocrine-disrupting chemicals: low-dose effects and nonmonotonic dose responses. Endocr Rev 33:378–455.

vom Saal FS, Akingbemi BT, Belcher SM, Birnbaum LS, Crain DA, Eriksen M, et al. 2007. Chapel Hill bisphenol A expert panel consensus statement: integration of mechanisms, effects in animals and potential to impact human health at current levels of exposure. Reprod Toxicol 24:131–138.

Witt KL, Livanos E, Kissling GE, Torous DK, Caspary W, Tice RR, et al. 2008. Comparison of flow cytometry-and microscopy-based methods for measuring micronucleated reticulocyte frequencies in rodents treated with nongenotoxic and genotoxic chemicals. Mutat Res 649(1):101–113.

Yamagiwa K, Ichikawa K. 1918. Experimental study of the pathogenesis of carcinoma. Cancer Res 3(1):1–29.

Relative Contributions of Agricultural Drift, Para-Occupational, and Residential Use Exposure Pathways to House Dust Pesticide Concentrations: Meta-Regression of Published Data

Author Affiliations open
1Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Department of Human Health and Services, Bethesda, Maryland, USA; 2Department of Environmental Health Sciences, Yale School of Public Health, Yale University, New Haven, Connecticut, USA; 3Department of Biological Sciences, Center for Human Health and the Environment, North Carolina State University, Raleigh, North Carolina, USA; 4National Exposure Research Laboratory, U.S. Environmental Protection Agency, Research Triangle Park, North Carolina, USA; 5Division of Surveillance, Hazard Evaluations and Field Studies, National Institute for Occupational Safety and Health, Cincinnati, Ohio, USA; 6Epidemiology Branch, National Institute of Environmental Health Sciences, National Institutes of Health, Department of Health and Human Services, Research Triangle Park, North Carolina, USA

PDF icon PDF Version (254 KB)

  • Background: Increased pesticide concentrations in house dust in agricultural areas have been attributed to several exposure pathways, including agricultural drift, para-occupational, and residential use.

    Objective: To guide future exposure assessment efforts, we quantified relative contributions of these pathways using meta-regression models of published data on dust pesticide concentrations.

    Methods: From studies in North American agricultural areas published from 1995 to 2015, we abstracted dust pesticide concentrations reported as summary statistics [e.g., geometric means (GM)]. We analyzed these data using mixed-effects meta-regression models that weighted each summary statistic by its inverse variance. Dependent variables were either the log-transformed GM (drift) or the log-transformed ratio of GMs from two groups (para-occupational, residential use).

    Results: For the drift pathway, predicted GMs decreased sharply and nonlinearly, with GMs 64% lower in homes 250 m versus 23 m from fields (interquartile range of published data) based on 52 statistics from seven studies. For the para-occupational pathway, GMs were 2.3 times higher [95% confidence interval (CI): 1.5, 3.3; 15 statistics, five studies] in homes of farmers who applied pesticides more recently or frequently versus less recently or frequently. For the residential use pathway, GMs were 1.3 (95% CI: 1.1, 1.4) and 1.5 (95% CI: 1.2, 1.9) times higher in treated versus untreated homes, when the probability that a pesticide was used for the pest treatment was 1–19% and ≥ 20%, respectively (88 statistics, five studies).

    Conclusion: Our quantification of the relative contributions of pesticide exposure pathways in agricultural populations could improve exposure assessments in epidemiologic studies. The meta-regression models can be updated when additional data become available.

  • Citation: Deziel NC, Beane Freeman LE, Graubard BI, Jones RR, Hoppin JA, Thomas K, Hines CJ, Blair A, Sandler DP, Chen H, Lubin JH, Andreotti G, Alavanja MC, Friesen MC. 2017. Relative contributions of agricultural drift, para-occupational, and residential use exposure pathways to house dust pesticide concentrations: meta-regression of published data. Environ Health Perspect 125:296–305;

    Address correspondence to N.C. Deziel, Yale School of Public Health, 60 College St., New Haven, CT 06510 USA. Telephone: (203) 785-6062. E-mail:

    This work was supported by the Intramural Research Program of the National Institutes of Health, National Cancer Institute (NCI), Division of Cancer Epidemiology and Genetics (Z01CP010119), and the National Institute of Environmental Health Sciences (Z01-ES049030). N.C.D. was supported, in part, through NCI (HHSN261201400231P). The U.S. Environmental Protection Agency through its Office of Research and Development collaborated in the research described here. It has been subjected to agency review and approved for publication.

    The findings and conclusions in this report are those of the authors and do not necessarily represent the views of the National Institute for Occupational Safety and Health.

    N.C.D. declares that her spouse is employed by the Dow Chemical Company, a manufacturer of thousands of chemicals, including some pesticides.

    The other authors declare no actual or potential competing financial interests.

    Received: 7 December 2015
    Revised: 28 April 2016
    Accepted: 30 June 2016
    Published: 26 July 2016

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (95 KB)

    PDF icon Supplemental Code and Data Zip File (5 KB)

    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact Our staff will work with you to assess and meet your accessibility needs within 3 working days.

    PDF icon Supplemental Table of Contents PDF (17 KB)


Adults living in agricultural areas may be exposed to pesticides from several occupational and environmental sources and pathways (Arbuckle et al. 2006; Curl et al. 2002; Deziel et al. 2015b; Harnly et al. 2009; Ward et al. 2006). Understanding the contribution of these use and transport exposure pathways to overall exposure is necessary for developing exposure assessment approaches for epidemiologic studies, designing exposure studies in nonoccupationally exposed populations, and for developing effective risk mitigation strategies. To provide a surrogate for quantitative, long-term, multi-source indoor pesticide exposure, pesticide concentrations in house dust have been measured in many studies (Butte and Heinzow 2002; Deziel et al. 2013; Lioy et al. 2002; Colt et al. 2004), in part because many pesticide biomarkers have short half-lives (measured in hours) reflecting recent exposure (Barr et al. 2006). Our recent qualitative review of exposure studies in North American agricultural environments found that increased pesticide concentrations in house dust were associated with take-home exposure from closer distances between homes and treated fields (agricultural drift pathway), farm work by one or more house residents (para-occupational pathway), and greater residential use of pesticides to treat various home, garden, and yard insects and weeds (residential use pathway) (Deziel et al. 2015b). To quantify the magnitude of these effects, we analyzed the house dust pesticide concentrations reported in published studies to obtain a summary measure of effect for each pathway across multiple pesticides and studies.

The analysis of the published data, however, presented a statistical challenge because the published dust pesticide concentrations were reported as summary statistics (i.e., means) or ratios (i.e., predicted relative difference obtained from regression models) rather than individual measurements. As a result, we needed to account for both the number of measurements and their variability to obtain an accurate summary effect measure. Koh et al. (2014) recently demonstrated the utility of mixed-effects meta-regression models to handle this challenge in an analysis synthesizing published lead exposure data to obtain temporal trends in occupational lead exposure. Our primary aim was to quantify the relative magnitude of exposure differences in dust pesticide concentrations in relation to surrogates representing each of the agricultural drift (e.g., distance of house to fields), para-occupational (e.g., how frequently a household member applies pesticides agriculturally), and residential pesticide use (e.g., treatment of insects or weeds in the home, yard, or garden) exposure pathways in North American agricultural homes. We focused on relative, rather than absolute, differences in dust pesticide concentrations within a pathway so that we could model the relationships across multiple active ingredients, for which absolute concentrations varied by orders of magnitude. Our secondary aim, undertaken to address the statistical challenges encountered, was to extend a mixed-effects meta-regression modeling approach used previously for epidemiologic analyses and occupational exposure data to environmental exposure data. To our knowledge, this represents one of the first uses of meta-regression models to synthesize published environmental measurements across multiple studies (Bain et al. 2014; Shields et al. 2015).


Data Abstraction

We included publications reporting pesticide concentrations in house dust in relation to agricultural drift, para-occupational activities, or residential use of pesticides in North American agricultural homes from our prior literature review (Deziel et al. 2015b) and one study that was published subsequent to our review (Deziel et al. 2015a). The prior systematic search identified studies published through September 2013 mainly from a PubMed search with the following terms: “environmental exposure [MeSH] AND pesticides [MeSH] AND (home OR household OR indoor).” We also searched Scopus, Web of Science, and Google Scholar, and examined reference lists of relevant publications. For the current analysis, we selected studies that measured pesticide concentrations in house dust because dust measurements are used as proxies for long-term environmental exposure (Butte and Heinzow 2002; Deziel et al. 2013). Based on findings from our prior review, we excluded studies with only air, food, or water samples due to low pesticide detection rates for those measures, and we excluded biological measurements because the measured pesticide biomarkers tended to have low percent detection and limited variability, and generally reflected only recent exposure. We repeated the PubMed search in March 2015 and identified one additional publication meeting the above criteria. Overall, 10 studies with published house dust pesticide concentrations were included.

From each study related to the agricultural drift pathway, we abstracted summary statistics of the house dust pesticide concentrations and the distances between the homes and the nearest fields. For distances reported categorically, we assigned the midpoint of the category. We used units of feet, because it was the most commonly reported unit and coincides with the response categories in many U.S. studies, including the Agricultural Health Study.

From each study related to the para-occupational pathway, we abstracted summary statistics of the dust pesticide concentrations for independent groups with different exposure potential (“comparison groups”). We extracted data for farmers with high pesticide use (high use group) versus low pesticide use (reference group), based on the frequency and recentness of pesticide application. In three studies, the high use group was farmers who applied pesticides generally and the reference group was those who did not apply pesticides (Fenske et al. 2002; Lu et al. 2000; Simcox et al. 1995). For two studies, both the high use and reference groups included pesticide applicators; therefore the comparison groups were those who applied the pesticide of interest either within 7 or 30 days of sampling (recentness of application varied by the pesticide active ingredient) versus those who did not apply the pesticide of interest within 7 or 30 days of sampling (Curwin et al. 2005), or those who applied atrazine ≥ 2 days/season versus < 2 days/season (Golla et al. 2012).

From studies of the residential use pathway, we extracted data from agricultural households reporting specific pest treatments (high use groups) and households reporting no treatment for that pest (reference group). In these studies, homeowners reported the type of pest treatment but did not provide the active ingredients of those treatments. Therefore, we extracted the type of pest treatment (e.g., fleas/ticks, weeds) and then derived a probability assessment of whether the treatment type was associated with the measured active ingredient using the National Cancer Institute (NCI) Pesticide Exposure Matrix (​icide) (Colt et al. 2007). This publicly available tool uses product sales and label information to predict the probability that an active ingredient was used in 96 different scenarios [12 pest treatment types, whether the applicator was a general consumer or professional commercial applicator, and four time frames (1976, 1980, 1990, 2000)]. We assigned the probabilities from the time frame closest to that of the individual study and, if multiple scenarios were relevant, averaged their probabilities. For example, from studies of weed treatment of lawns, we averaged the probabilities from the “professional weeds” and “consumer weeds” scenarios. We categorized the probabilities as 0% (active ingredient not listed), 1–19%, and ≥ 20%.

From the 10 studies for the above-mentioned comparison groups, we abstracted the available information on dust pesticide concentrations. These data were predominantly reported as summary results [i.e., arithmetic means (AMs), geometric means (GMs), standard deviations (SDs), geometric standard deviations (GSDs), number of measurements (N)]. We also abstracted data on the ratios between two comparison groups predicted from multivariable regression models, rather than GMs. These data were usually reported in tables; however, we also extracted data from boxplots and other figures when necessary. For each set of summary statistics, we obtained reported ancillary data, including study years, pesticide active ingredient, pesticide type (e.g., herbicide, insecticide, fungicide), and crop type (e.g., corn, orchard fruit). If the same measurements were reported both in descriptive analyses and in multivariable regression models within the same paper, we abstracted the data only once. To best capture the independent contribution of a single pathway, we abstracted data that accounted for the other potential pathways through adjustment in multivariable regression models or stratification wherever possible.

Data Treatment

We obtained a GM and GSD for each set of published dust pesticide concentrations. When these summary statistics were not directly reported, we estimated them using available formulae presented in Equations 1 through 6, where AM is the arithmetic mean, SD is the standard deviation, max is the maximum value, min is the minimum value, p25 is the value at the 25th percentile, and p10 is the value at the 10th percentile (Hein et al. 2008; Hewett 2005; Koh et al. 2014; Lavoué et al. 2007; Aitchison and Brown 1963).

GM equals e to-the-power-of natural-logarithm left-parenthesis AM right-parenthesis minus 0.5 times natural-logarithm left-parenthesis 1 plus open left-parenthesis fraction SD over AM, right-parenthesis, squared right-parenthesis[1]

GM = median[2]

GM = e(ln(max) + ln(min))/2 [3]

GSD equals e to-the-power-of squareroot natural-logarithm left-parenthesis 1 plus open left-parenthesis fraction SD over AM, right-parenthesis, squared right-parenthesis[4]

GSD = (e(ln(p25) – ln(GM))/–0.68 + e(ln(p10) – ln(GM))/–1.282)/2 [5]

GSD = e(ln(max) – ln(min))/4 [6]

Two studies collected more than one sample per home (Curwin et al. 2005; Golla et al. 2012). We accounted for the repeated within-home measurements in these studies by adjusting the number of samples collected using a design effect (Equation 7) to calculate the effective sample size (Equation 8) (Kish 1965). We divided the between-home variance by the sum of the between- and within-home variances to obtain the intra-class correlation coefficient (ICC) for each active ingredient (Curwin et al. 2005). In these two studies, the calculated effective sample size replaced the total number of measurements (Nsamples) for each summary statistic, where Nhomes is the number of homes corresponding to each summary statistic.

design effect equals 1 plus open left-parenthesis open left-parenthesis fraction N sub samples, over N sub homes, right-parenthesis, minus 1 times ICC[7]

effective sample size = Nsamples/design effect [8]

The GMs of the data on the effect of agricultural drift were approximately log-normally distributed based on visual inspection and therefore were natural log-transformed prior to additional analyses. For these data, we calculated the variance of each log-transformed GM using Equation 9, which we derived for these analyses using the delta method:

var left-parenthesis natural-logarithm left-parenthesis GM right-parenthesis right-parenthesis equals open left-parenthesis fraction 1 over N sub samples, right-parenthesis, times left-parenthesis natural-logarithm left-parenthesis GSD right-parenthesis right-parenthesis squared[9]

For the para-occupational and residential use pathways, the data were often abstracted from multivariable regression models that examined the association between log-transformed exposure and various determinants of exposure. We interpreted the anti-log of a model parameter, exp(β), as the ratio between the GMs of the high use group and the reference group. Hence, to include these data, we assumed that β equaled ln(ratio) and the β’s standard error squared (SE2) equaled the variance of the ln(ratio). When the standard error was not reported, we extracted it from the lower and upper confidence limits (LCL and UCL) of the exp(β) using Equation 10. Note that Equation 10 assumes that the LCL and UCL were reported as exponentiated terms, as was the case in these studies, and thus required transformation back to the log-scale.

SE equals open left-parenthesis fraction natural-logarithm LCL minus natural-logarithm UCL over 2, right-parenthesis, solidus 1.96[10]

To combine these regression parameter statistics with the data that were abstracted as GMs required converting the GMs to ratios. For these two pathways, we calculated the ratio of the GMs of the high use group compared to the reference group (Equation 11). The ratios were assumed log-normally distributed based on visual inspection and log-transformed prior to additional analyses. For these data we calculated the variance of the ratio of the log-transformed GMs using Equation 12, which we derived using the delta method.

ratio equals fraction GM sub high, over GM sub reference[11]

var open left-parenthesis fraction natural-logarithm left-parenthesis GM sub high, right-parenthesis over natural-logarithm left-parenthesis GM sub reference, right-parenthesis, right-parenthesis, equals open left-parenthesis fraction 1 over N sub high, right-parenthesis, times left-parenthesis natural-logarithm left-parenthesis GSD sub high, right-parenthesis right-parenthesis squared plus open left-parenthesis fraction 1 over N sub reference, right-parenthesis, times left-parenthesis natural-logarithm left-parenthesis GSD sub reference, right-parenthesis right-parenthesis squared[12]

Meta-Regression Models

We developed separate mixed-effects meta-regression models for each of the agricultural drift, para-occupational, and residential use pathways using PROC MIXED (version 9.3; SAS Institute Inc., Cary, NC). In these models, each summary statistic was weighted by the inverse of its study-specific variance. Regression parameters and between-study variances were obtained using maximum likelihood estimation. Estimation of the between-study variance required a starting value for computation, which we set to one-half of the mean within-study variance (Konstantopoulos and Hedges 2004). Pathway-specific analyses are described below, with the SAS code for each pathway’s primary model provided in the Supplemental Material, Appendix 1. For each pathway, we examined how the relative magnitude of dust pesticide concentrations varied based on surrogate measures, such as the relative change in dust pesticide concentrations at varying distances of the house to fields for the agricultural drift pathway. Exposure comparisons were made within, but not across pathways.

For agricultural drift, the dependent variable was the log-transformed GM. Most studies included GMs at various distances from treated fields and the reference distance varied between studies. As a result, the agricultural drift model incorporated two random effects: one identified each unique combination of publication, active ingredient, and distance from field to weight each statistic by the inverse of its study-specific variance and a second identified each unique combination of publication and active ingredient to account for active ingredient- and study-specific differences in baseline pesticide concentrations. We identified the best parametric characterization between the log-transformed GM and distance by evaluating various forms, including linear distance, natural log-transformed distance, inverse distance, and inverse distance squared. The natural log-transformed distance provided the best model fit based on the Akaike information criterion that was also consistent with graphical evaluations (not shown). In preliminary models, we tested a random slope for the relationship between ln(GM) and ln(distance) to allow for study- and pesticide-specific differences in the slope; and found no differences in slopes, although data were sparse. The overall drift model included only ln(distance) and provided an estimate of the GM dust pesticide concentration at varying distances in ft (d) of the house from the fields using the regression coefficients for the intercept (βintercept) and ln(distance) (βslope) (Equation 13). The percent change in GMs between two specific distances (d1 and d2) is calculated using Equation 14. In sensitivity analyses, we also developed separate models for herbicides, insecticides, and chlorpyrifos (the most commonly measured insecticide).

Predicted GM = exp[βintercept + βslope × ln(d)] = dβslope eintercept) [13]

% change in GM between d1 and d2 = [(GMd1GMd2)/GMd1] × 100 = [1 – (d2/d1)βslope] × 100 [14]

For para-occupational and residential use, the data were often abstracted as ratios from multivariable regression models, and these models used the log-transformed ratios of GMs from the high use and reference groups as the dependent variable. Using the ratio had the added benefit of removing the active-ingredient-specific differences in baseline; thus these models incorporated a single random intercept, which identified each unique combination of publication and pesticide active ingredient to weight each observation by the inverse of its study-specific variance.

For para-occupational exposure, we calculated an overall summary of the effect of the take-home pathway. In sensitivity analyses, we also developed separate models for the two types of comparison groups (farmers who ever applied vs. never applied pesticides and farmers who applied the pesticides in specific time windows or with specific frequencies). Studies comparing farmers who ever applied versus never applied pesticides were mainly of insecticide applications to fruit orchards; whereas, the studies comparing groups with more specific timing or frequency of applications addressed herbicide applications to row crops like corn and soybeans. Thus, we could not disentangle the separate effects of comparison group, pesticide type, and crop type. In additional sensitivity analyses, we also developed separate models for atrazine and chlorpyrifos.

For residential use, we calculated an overall summary of the ratio in dust pesticide concentrations in homes reporting various home, garden, and yard pest treatments (high use group) versus those not reporting a given treatment (reference group). We also developed separate models for each active ingredient-pest treatment probability category, which were evaluated overall, using only the largest study (Deziel et al. 2015a), and using all studies but the largest. In addition, we evaluated the effect of the active ingredient probability category separately for herbicides and insecticides. The data stratified by probability category were too sparse to develop pesticide-specific models.


Agricultural Drift Pathway

We identified seven studies reporting concentrations in house dust of multiple pesticide active ingredients in homes at varying distances from fields (Table 1) from which we extracted 52 sets of estimates. The reported distances ranged from 10 to 3,690 ft (3 to 1,125 m) with 25th, 50th, and 75th percentiles of 75 ft (23 m), 300 ft (91 m), and 820 ft (250 m), respectively. GSDs ranged from 1.4 to 10. Overall, house dust pesticide concentrations decreased sharply and non-linearly with increasing house distance from treated fields that was linear on a log-log scale, as shown in Figure 1. The model predicted GMs that were 64% lower in homes of 820 ft compared to 75 ft [the interquartile range (IQR)] and 35% lower in homes of 820 ft compared to 300 ft (75th percentile and median). The magnitude of decrease varied by pesticide type, with a 78% decrease in predicted GMs across the IQR for herbicides and fungicides and 51% across the IQR for insecticides (Table 2; see also Figure S1). These magnitudes of decreases were statistically different (p-value = 0.049) in a model that included all data and an interaction term for pesticide type and ln-distance. The magnitude of decline for chlorpyrifos mirrored that of all insecticides (50% across the IQR).

Table 1. Select View Table (HTML Version) for a 508-conformant version

Table 1. Geometric means (GMs) of agricultural drift of dust pesticide concentrations in agricultural homes at varying distances from fields.

View Table (HTML Version)
View larger image (TIF File)

Figure 1. Scatterplot of geometric mean pesticide concentrations in house dust samples (micrograms per gram) (y-axis) according to distance from treated fields (in feet, x-axis) with the size of the symbol for each individual geometric mean indicating its relative precision, and an overlaid meta-regression line.

Figure 1. The GMs of concentrations of pesticide house dust decreased logarithmically with distance between home and treated fields. Solid line indicates predicted association from meta-regression models. Predicted GM at a given distance d in ft = dβslopee(βintercept) = d–0.43e0.15. Circles indicate Distance/Pesticide/Paper-specific GMs, with circle width = (logGSD)2/Nsamples.

View larger image (TIF File)

Table 2. Select View Table (HTML Version) for a 508-conformant version

Table 2. Agricultural drift: model parameters for predicting the GMs of dust pesticide concentrations in agricultural homes at varying distances from fields.

View Table (HTML Version)
View larger image (TIF File)

Para-Occupational Pathway

We identified five studies reporting pesticide concentrations in house dust that could be used to quantify the mean difference between homes of farmers with high vs. low pesticide use (Table 3). From these studies we derived 15 estimates of the ratio of GMs (GM Ratio) for homes of farmers in the high use versus reference group. The GM ratios varied from 0.57 to 31 and the variances of the ln-ratios varied across three orders of magnitude.

Table 3. Select View Table (HTML Version) for a 508-conformant version

Table 3. Para-occupational exposure: Ratios of GMs of dust pesticide concentrations between measurements taken in homes of farmers with high pesticide use compared to those with low pesticide use.

View Table (HTML Version)
View larger image (TIF File)

Overall, in a meta-regression model, we found that dust pesticide concentrations were 2.3 times higher [95% confidence interval (CI): 1.5, 3.3] in homes of farmers with high pesticide use versus the reference group (Table 4). Sensitivity analyses indicated higher ratios in studies of farmers who applied specific pesticides in specific time windows or frequencies (Ratio: 3.8, 95% CI: 1.6, 9.2) than in studies with more general comparisons between farmers who applied or did not apply pesticides (Ratio: 2.0, 95% CI: 1.3, 3.0). It is not clear if these differences were related to differences between the comparison group type or pesticide type. However, these differences between the ratios were not statistically significant in a model that included all data and that incorporated pesticide type as an explanatory variable (p-value = 0.14). Additionally, the ratios were higher for atrazine (Ratio: 4.7, 95% CI: 1.6, 13) than for chlorpyrifos (Ratio: 1.6, 95% CI: 1.1, 2.3) (p-value for comparison = 0.06); however, these comparisons were based on small numbers. The between-statistic variances estimated by these models were not statistically significantly different from zero.

Table 4. Select View Table (HTML Version) for a 508-conformant version

Table 4. Para-occupational exposure: Predicted ratios of dust pesticide concentrations of homes of farmers with high versus low pesticide use based on meta-regression models.

View Table (HTML Version)
View larger image (TIF File)

Residential Use Pathway

We identified five studies reporting pesticide concentrations in house dust in agricultural homes that were treated or not treated (i.e., high use vs. reference) for various insects and weeds in the home, garden, or yard (Table 5). From these studies we derived 88 estimates of the ratios of GMs between treated vs. non-treated homes. The GM ratios varied from 0.22 to 6.8. Overall, dust pesticide concentrations were 1.3 times higher (95% CI: 1.1, 1.4) in households that treated vs. did not treat their homes, gardens, or yards for insect or weeds (Table 6). The magnitude of the contribution increased with the probability of use of an active ingredient for the specific pest treatment. For probability categories 0%, 1–19%, and ≥ 20%, respectively, the dust pesticide concentrations were 1.0 (95% CI: 0.8, 1.3), 1.3 (95% CI: 1.1, 1.4), and 1.5 (95% CI: 1.2, 1.9) times higher in households that treated vs. did not treat. The magnitude of the effect by probability category was somewhat larger when Deziel et al. 2015a was excluded (Table 6). However, this effect was only statistically different at the 1–19% probability category in models that included all data and a fixed-effect term for data source (1 = Deziel et al. 2015a; 0 = all other studies; p-values of 0.5, 0.002 and 0.4 for probability categories 0%, 1–19%, and ≥ 20%, respectively; not shown). Stratified analyses also showed some differences in the magnitude of effect between herbicides and insecticides for the 1–19% probability category, but not the 0% and ≥ 20% categories (Table 6); this difference was not statistically significant (p-value = 0.2).

Table 5. Select View Table (HTML Version) for a 508-conformant version

Table 5. Residential use exposure: Ratios of GMs of dust pesticide concentrations between measurements taken in agricultural homes that were treated (high use) versus not treated (reference) for home, garden, or yard insects or weeds.

View Table (HTML Version)
View larger image (TIF File)

Table 6. Select View Table (HTML Version) for a 508-conformant version

Table 6. Residential use: Predicted ratio in GM dust pesticide concentrations in treated versus untreated agricultural homes.

View Table (HTML Version)
View larger image (TIF File)


To our knowledge, this is the first use of meta-regression models to summarize environmental pesticide concentrations reported in the published literature. This approach allowed us to estimate the average ratios in pesticide concentration in house dust from various pathways across multiple studies while accounting for both the study size and concentration variability. Overall, pesticide concentrations in house dust decreased rapidly with increasing distance, with predicted GMs decreasing 64% across the IQR of the published data. Pesticide concentrations in dust were also 2.3 times higher (95% CI: 1.5, 3.3) in homes where a resident had high versus low agricultural use of pesticides and 1.3 times higher (95% CI; 1.1, 1.4) in homes where pesticides were used in the home, garden, or yard versus not used for specific pests. These findings provide data-driven weights, with confidence intervals, that could be used in future exposure assessment efforts in epidemiologic studies. In addition, this study provides a framework for applying meta-regression models to analyze published data for other exposures and exposure determinants of interest.

The contribution of the residential use pathway increased with increasing probability of the pesticide treatment type including the active ingredient, but there was little evidence for subgroup differences in the other two pathways. Sensitivity analyses suggested that the magnitude of the contribution of each pathway may differ by pesticide type or active ingredient. At this time, we have insufficient evidence to confirm these differences, as sample sizes were generally small. For example, agricultural drift might be influenced by pesticide type or active ingredient (e.g., due to differences in the volatility of active ingredients), crop type, meteorology, and pesticide application method (Damalas and Eleftherohorinos 2011; Ward et al. 2006). However, in our comparisons, the strong correlations among potential explanatory variables prevented us from disentangling whether the observed differences were attributable to any of those factors, and important differences may have been missed because of sparse data. Because of our transparent approach, our results can be updated as more data becomes available.

Using a mixed-effects model framework provided an opportunity to systematically account for both the within-study and between-study variability for specific pesticides and the power of the study based on the number of measurements. However, differences may still have been masked. For instance, in the agricultural drift model, there is potential for aggregation bias because we evaluated only an overall trend, rather than pesticide/study-specific trends. Preliminary models that incorporated a random slope did not detect pesticide/study-specific trends, but differences may have been missed due to computational limitations and sparse data. Visual inspection showed that most pesticide/study-specific trends paralleled the overall trend (see Figure S2); for these the random intercept would be sufficient to capture the offset in the intercept. However, we may have missed important differences for the small number of trends that did not parallel the overall trend. This challenge was also encountered in a previous meta-regression analysis of occupational lead data, where industry-specific temporal trends were unable to capture differences related to variability in the jobs that were monitored (Koh et al. 2014).

These models estimated an average effect that provides an estimate of the median change in dust pesticide concentration between distances (drift) or comparison groups (para-occupational, residential use pathways). In epidemiologic studies, comparing arithmetic means may be of greater interest. The arithmetic mean (AM) can be approximated using the equation AM = exp(β + 0.5 * SE2), where β and SE are the model parameter and its standard error before exponentiating the terms. Here, for the three pathways, the AM and GM were identical at one decimal place (not shown). Estimating the ratio of the AMs between comparison groups, that is, AMhigh/AMreference instead of GMhigh/GMreference, is more challenging, because the log-normal statistical properties of the former’s exposure distribution are more difficult to obtain and incorporate into statistical models. More flexible modeling approaches, such as Bayesian approaches that can specify different exposure distributions for each parameter, may address this challenge.


There were several additional limitations to these analyses related to the coverage of the data and use of surrogates to represent these three exposure pathways. First, the magnitude of these differences may be overestimated due to publication bias because studies that observed no association between pesticide house dust levels and a particular pathway often did not report summary statistics or regression coefficients and could not be included here. Publication bias may account for differences in the contribution of the residential use pathway between the Deziel et al. (2015a) study and all other papers. The Deziel et al. (2015a) study included 74 of the 88 statistics and reported all possible comparisons between multiple pesticides and pest treatments, whereas other papers evaluating several pesticides generally reported only statistically significant findings. For the agricultural drift pathway, several studies stated that they did not observe an association between dust pesticide concentrations and distance from home to treated fields without providing the underlying summary statistics (Coronado et al. 2011; Curwin et al. 2005; McCauley et al. 2003). However, in these studies, the homes tended to be located very close to the fields, limiting the variability in distance categories. Second, as described above, the data were generally too sparse to identify whether differences in pesticide house dust concentrations varied by subgroups (e.g., pesticide type, crop type, application method, geographic location, or time period) and important distinctions may have been missed. Third, we used exposure surrogates to create our comparison groups; the exposure pathways may be better characterized with other metrics. For instance, compared to self-reported distance to treated fields, agricultural drift may be better captured using geographic information systems approaches that use satellite images, crop maps, historical farm records, and state pesticide use reporting databases to better classify exposure according to crop acreage or quantity of active ingredients applied near residences (Fenske 2005; Gunier et al. 2011; Harnly et al. 2009; Jones et al. 2014; Ritz and Rull 2008; Ward et al. 2000); Fourth, most of the studies were based in the northwestern United States (Washington and Oregon) and Iowa, and thus the results may not be generalizable to populations in other geographic regions. Lastly, the lack of reporting of active ingredient-specific information in the published studies of the residential use treatments, and the resulting use of group-level probability-based weights from the NCI pesticide exposure matrix, introduces uncertainty in the quantification of the contribution of the residential use pathway. This pesticide exposure matrix was last updated with market and usage data from the year 2000 and may have limited relevance for informing residential use of certain pesticides subsequent to that year.

There were also several limitations to the abstracted data and the modeling framework. First, it is difficult to disentangle the independent contribution of each pathway. Although we abstracted data that accounted for the other potential pathways through adjustment in multivariable regression models or stratification wherever possible, the estimates of the contributions of each pathway may be confounded by other pathways. Second, development of the richest data source possible required approximations, with varying errors, when GMs and GSDs were not directly reported. For example, we assumed the median was approximately equivalent to the GM. In addition, we visually extracted medians from graphs in four of the seven studies of agricultural drift, which introduced imprecision in the estimates. Similarly, based on visual inspection of the data, we assumed a lognormal distribution for both the dust pesticide concentrations and the ratios. Deviations from this assumption could affect the point estimates, p-values and confidence intervals. As a result, we presented results only to 2 significant figures and we use confidence intervals and p-values as guides and not definitive measures of scientific significance.

These findings provide insight into the contributions of these exposure pathways to the indoor dust pesticide concentrations; however, the impact of these differences to the pesticide exposure of adults remains uncertain because individual behaviors and characteristics also influence the amount of pesticide exposure and absorption (Hoppin et al. 2006). Pesticide concentrations in air, food, water, and biological specimens may also be used to represent adult exposure and dose. However, our prior review found that evaluations in media other than dust were rare, often had low detection rates, and for biomarkers represented only very recent exposure (Deziel et al. 2015b); as a result, these metrics were not included in these analyses. Previous studies that have compared concentrations or loadings of pesticides in bulk dust or wipes with concentrations of pesticide biomarkers in adults have observed weak to moderate correlations or associations (Thompson et al. 2014; Curwin et al. 2007; Arbuckle et al. 2006). However, making these comparisons is challenging because the varying media reflect different exposure windows, with dust samples representing a cumulative time window representing weeks, months, or years, and biomarkers often representing exposure in the hours to days prior to sample collection (Barr et al. 2006; Bouvier et al. 2005; Morgan et al. 2008). Future research with repeated biological measures would advance our understanding of the predictive value of pesticide house dust measurements for long-term exposures in adults. In addition, the framework used here can be expanded to other sample media as more data becomes available.


We used a novel application of meta-analysis to published pesticide exposure data to quantify the relative difference in dust pesticide concentrations in relation to surrogates representing three pesticide use and transport exposure pathways in agricultural populations. Our analyses found that homes near treated fields, homes of farmers who applied pesticides more frequently or recently, and homes of those who applied pesticides around the home, garden, and yard, had quantifiably higher pesticide concentrations in the dust compared to their reference groups. These results can inform the development of data-driven environmental exposure categorizations for epidemiologic studies. Our transparent meta-regression models can be updated when new data are available or further restricted or expanded based on the population of interest. Additionally, the framework developed for these analyses can be applied to other published exposure data.


Aitchison J, Brown JAC. 1963. The Lognormal Distribution: with Special Reference to Its Uses in Economics. Cambridge, UK:Cambridge University Press.

Arbuckle TE, Bruce D, Ritter L, Hall JC. 2006. Indirect sources of herbicide exposure for families on Ontario farms. J Expo Sci Environ Epidemiol 16:98–104.

Bain R, Cronk R, Wright J, Yang H, Slaymaker T, Bartram J. 2014. Fecal contamination of drinking-water in low- and middle-income countries: a systematic review and meta-analysis. PLoS Med 11(5):e1001644, doi: 10.1371/journal.pmed.1001644.

Barr DB, Thomas K, Curwin B, Landsittel D, Raymer J, Lu C, et al. 2006. Biomonitoring of exposure in farmworker studies. Environ Health Perspect 114:936–942, doi: 10.1289/ehp.8527.

Bouvier G, Seta N, Vigouroux-Villard A, Blanchard O, Momas I. 2005. Insecticide urinary metabolites in nonoccupationally exposed populations. J Toxicol Environ Health B Crit Rev 8:485–512.

Butte W, Heinzow B. 2002. Pollutants in house dust as indicators of indoor contamination. Rev Environ Contam Toxicol 175:1–46.

Colt JS, Cyr MJ, Zahm SH, Tobias GS, Hartge P. 2007. Inferring past pesticide exposures: a matrix of individual active ingredients in home and garden pesticides used in past decades. Environ Health Perspect 115:248–254, doi: 10.1289/ehp.9538.

Colt JS, Lubin J, Camann D, Davis S, Cerhan J, Severson RK, et al. 2004. Comparison of pesticide levels in carpet dust and self-reported pest treatment practices in four US sites. J Expo Anal Environ Epidemiol 14:74–83.

Coronado GD, Holte S, Vigoren E, Griffith WC, Barr DB, Faustman E, et al. 2011. Organophosphate pesticide exposure and residential proximity to nearby fields: evidence for the drift pathway. J Occup Environ Med 53:884–891.

Curl CL, Fenske RA, Kissel JC, Shirai JH, Moate TF, Griffith W, et al. 2002. Evaluation of take-home organophosphorus pesticide exposure among agricultural workers and their children. Environ Health Perspect 110:A787–A792.

Curwin BD, Hein MJ, Sanderson WT, Nishioka MG, Reynolds SJ, Ward EM, et al. 2005. Pesticide contamination inside farm and nonfarm homes. J Occup Environ Hyg 2:357–367.

Curwin BD, Hein MJ, Sanderson WT, Striley C, Heederik D, Kromhout H, et al. 2007. Urinary pesticide concentrations among children, mothers and fathers living in farm and non-farm households in Iowa. Ann Occup Hyg 51:53–65.

Damalas CA, Eleftherohorinos IG. 2011. Pesticide exposure, safety issues, and risk assessment indicators. Int J Environ Res Public Health 8:1402–1419.

Deziel NC, Colt JS, Kent EE, Gunier RB, Reynolds P, Booth B, et al. 2015a. Associations between self-reported pest treatments and pesticide concentrations in carpet dust. Environ Health 14:27, doi: 10.1186/s12940-015-0015-x.

Deziel NC, Friesen MC, Hoppin JA, Hines CJ, Thomas K, Beane Freeman LE. 2015b. A review of nonoccupational pathways for pesticide exposure in women living in agricultural areas. Environ Health Perspect 123:515–524, doi: 10.1289/ehp.1408273.

Deziel N, Ward M, Bell E, Whitehead T, Gunier R, Friesen M, et al. 2013. Temporal variability of pesticide concentrations in homes and implications for attenuation bias in epidemiologic studies. Environ Health Perspect 121:565–571, doi: 10.1289/ehp.1205811.

Fenske RA. 2005. State-of-the-art measurement of agricultural pesticide exposures. Scan J Work Environ Health 31(suppl 1):67–73.

Fenske RA, Lu C, Barr D, Needham L. 2002. Children’s exposure to chlorpyrifos and parathion in an agricultural community in central Washington State. Environ Health Perspect 110:549–553.

Golla V, Curwin B, Sanderson W, Nishioka M. 2012. Pesticide concentrations in vacuum dust from farm homes: variation between planting and nonplanting seasons. ISRN Public Health2012:539397, doi: 10.5402/2012/539397.

Gunier RB, Ward MH, Airola M, Bell EM, Colt J, Nishioka M, et al. 2011. Determinants of agricultural pesticide concentrations in carpet dust. Environ Health Perspect 119:970–976, doi: 10.1289/ehp.1002532.

Harnly ME, Bradman A, Nishioka M, McKone TE, Smith D, McLaughlin R, et al. 2009. Pesticides in dust from homes in an agricultural area. Environ Sci Technol 43:8767–8774.

Hein MJ, Waters MA, van Wijngaarden E, Deddens JA, Stewart PA. 2008. Issues when modeling benzene, toluene, and xylene exposures using a literature database. J Occ Environ Hyg 5:36–47.

Hewett P. 2005. Equations for calculating exposure manage ment objectives. Technical Report No. 05–02. Morgantown, WV:Exposure Assessment Solutions, Inc.​2016/03/TR-05-02-ContObj.pdf [accessed 22 December 2016].

Hoppin JA, Adgate JL, Eberhart M, Nishioka M, Ryan PB. 2006. Environmental exposure assessment of pesticides in farmworker homes. Environ Health Perspect 114:929–935, doi: 10.1289/ehp.8530.

Jones RR, Yu CL, Nuckols JR, Cerhan JR, Airola M, Ross JA, et al. 2014. Farm residence and lymphohematopoietic cancers in the Iowa Women’s Health Study. Environ Res 133:353–361, doi: 10.1016/j.envres.2014.05.028.

Kish L. 1965. Survey Sampling. New York:John Wiley & Sons.

Koh DH, Nam JM, Graubard BI, Chen YC, Locke SJ, Friesen MC. 2014. Evaluating temporal trends from occupational lead exposure data reported in the published literature using meta-regression. Ann Occup Hyg 58:1111–1125.

Konstantopoulos S, Hedges LV. 2004. Meta-analysis. In: The SAGE Handbook of Quantitative Methodology for the Social Sciences. Kaplan D, ed. Thousand Oaks, CA:SAGE Publications, Inc, 281–297.

Lavoué J, Bégin D, Beaudry C, Gérin M. 2007. Monte Carlo simulation to reconstruct formaldehyde exposure levels from summary parameters reported in the literature. Ann Occup Hyg 51:161–172.

Lioy PJ, Freeman NC, Millette JR. 2002. Dust: a metric for use in residential and building exposure assessment and source characterization. Environ Health Perspect 110:969–983.

Lu C, Fenske RA, Simcox NJ, Kalman D. 2000. Pesticide exposure of children in an agricultural community: evidence of household proximity to farmland and take home exposure pathways. Environ Res 84:290–302.

McCauley LA, Lasarev MR, Higgins G, Rothlein J, Muniz J, Ebbert C, et al. 2001. Work characteristics and pesticide exposures among migrant agricultural families: a community-based research approach. Environ Health Perspect 109:533–538.

McCauley LA, Michaels S, Rothlein J, Muniz J, Lasarev M, Ebbert C. 2003. Pesticide exposure and self reported home hygiene: practices in agricultural families. AAOHN J 51:113–119.

Morgan MK, Sheldon LS, Thomas KW, Egeghy PP, Croghan CW, Jones PA, et al. 2008. Adult and children’s exposure to 2,4-D from multiple sources and pathways. J Expo Sci Environ Epidemiol 18:486–494.

Ritz B, Rull RP. 2008. Assessment of environmental exposures from agricultural pesticides in childhood leukaemia studies: challenges and opportunities. Radiat Prot Dosimetry 132:148–155.

Shields KF, Bain RES, Cronk R, Wright JA, Bartram J. 2015. Association of supply type with fecal contamination of source water and household stored drinking water in developing countries: a bivariate meta-analysis. Environ Health Perspect 123:1222–1231, doi: 10.1289/ehp.1409002.

Simcox NJ, Fenske RA, Wolz SA, Lee IC, Kalman DA. 1995. Pesticides in household dust and soil: exposure pathways for children of agricultural families. Environ Health Perspect 103:1126–1134.

Thompson B, Griffith WC, Barr DB, Coronado GD, Vigoren EM, Faustman EM. 2014. Variability in the take-home pathway: farmworkers and non-farmworkers and their children. J Expo Sci Environ Epidemiol 24:522–531.

Ward MH, Lubin J, Giglierano J, Colt JS, Wolter C, Bekiroglu N, et al. 2006. Proximity to crops and residential exposure to agricultural herbicides in Iowa. Environ Health Perspect 114:893–897, doi: 10.1289/ehp.8770.

Ward MH, Nuckols JR, Weigel SJ, Maxwell SK, Cantor KP, Miller RS. 2000. Identifying populations potentially exposed to agricultural pesticides using remote sensing and a geographic information system. Environ Health Perspect 108:5–12.

Biomarker Levels of Toxic Metals among Asian Populations in the United States: NHANES 2011–2012

Author Affiliations open
1University of Texas Health Science Center at Houston School of Public Health, Houston, Texas, USA; 2CH2M HILL, Inc., Houston, Texas, USA

PDF icon PDF Version (188 KB)

  • Introduction: The Centers for Disease Control and Prevention (CDC) recently found that Asians have considerably higher biomarker levels of cadmium, lead, mercury, and arsenic than whites, blacks, Mexican Americans, and other Hispanics in the United States.

    Objective: Our goal was to further evaluate the higher metal biomarker levels among Asians.

    Methods: Biomarker data (blood cadmium, blood lead, blood mercury, urinary total arsenic, and urinary dimethylarsinic acic) from individuals ≥ 6 years of age were obtained from the 2011–2012 National Health and Nutrition Examination Survey (NHANES). We compared geometric mean levels of these five metal biomarkers in Asians with those of four other NHANES race/ethnic groups (white, black, Mexican American, and other Hispanic), and across three Asian subgroups (Chinese, Asian Indian, and other Asian). We also evaluated associations between biomarker levels and sociodemographic, physical, dietary, and behavioral covariates across the Asian subgroups.

    Results: Asians had significantly higher levels of all five metal biomarkers than other race/ethnic groups (p < 0.05), regardless of sociodemographic, physical, dietary, behavioral, or geographic characteristics. We also found variations in biomarker levels across the Asian subgroups. In general, Asian Indians had lower levels than the other two Asian subgroups, except for blood lead. The following characteristics were found to be significant predictors of several biomarker levels: sex, age, education, birthplace, smoking, and fish consumption.

    Conclusions: Overall, the Asian group had the highest geometric mean biomarker levels for all of the five metal variables. Furthermore, we provided evidence that significant variations in the biomarker levels are present across the Asian subgroups in the United States.

  • Citation: Awata H, Linder S, Mitchell LE, Delclos GL. 2017. Biomarker levels of toxic metals among Asian populations in the United States: NHANES 2011–2012. Environ Health Perspect 125:306–313;

    Address correspondence to H. Awata, c/o CH2M, 14701 St. Mary’s Lane, Suite 300, Houston, Texas 77079 USA. Telephone: (281) 721-8487. E-mail:

    The data analysis of the restricted data at the Research Data Center (RDC) in Atlanta was supported by department research funds from the University of Texas Health Science Center School of Public Health.

    The findings and conclusions in this paper are those of the authors and do not necessarily represent the views of the RDC, the National Center for Health Statistics, or the Centers for Disease Control and Prevention.

    The authors declare they have no actual or potential competing financial interests.

    Received: 24 February 2016
    Revised: 10 June 2016
    Accepted: 18 July 2016
    Published: 12 August 2016

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (229 KB)

    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact Our staff will work with you to assess and meet your accessibility needs within 3 working days.

    PDF icon Supplemental Table of Contents PDF (96 KB)


Cadmium, lead, mercury, and arsenic are among the most toxic environmental contaminants. The International Agency for Research on Cancer (IARC) classifies arsenic and cadmium as human carcinogens (Group 1), and lead and mercury (methylmercury) as possibly carcinogenic to humans (Group 2B) (IARC 2013). Although levels of exposure to these metals/metalloids (hereafter, collectively referred to simply as “metals”) have been generally decreasing in the United States, various adverse health effects, such as cardiovascular and developmental effects, damage to the nervous system, and kidney failure, have been associated with exposure to these metals at the current, relatively low, environmental exposure levels (Ferraro et al. 2010; Lebel et al. 1996; McLaine et al. 2013; Moon et al. 2013). The health effects of low-level exposures are also important because some of the effects have been regarded to have no safe exposure threshold (Anderson 1983; Jakubowski 2011). Therefore, exposure to these toxic metals still poses a significant public health risk, and it is vital to reduce overall exposure and subsequently health risks, especially for those highly exposed subpopulation groups.

Asian populations have considerably higher blood and urinary levels of these metals than other racial/ethnic groups (i.e., whites, blacks, and Hispanics) in the United States (CDC 2014; McKelvey et al. 2007). For example, based on a recent analysis of biomarker data by the Centers for Disease Control and Prevention (CDC) in 2014, the geometric mean blood mercury levels among Asians (1.86 μg/L) is four times greater than that of Mexican Americans (0.48 μg/L) (CDC 2014). Asian populations in the United States include multiple ethnic subgroups that are culturally, religiously, historically, and geographically diverse. Hence, the differences in these characteristics across subgroups may affect biomarker levels of these metals. However, this was not examined in the original CDC analysis.

The National Health and Nutrition and Examination Survey (NHANES) is a national population-based survey program conducted by the National Center for Health Statistics (NCHS) that assesses the health and nutritional status of the civilian noninstitutionalized general U.S. population. The NCHS collects data continuously and releases data every 2 years in a 2-year data cycle. An important addition to the most recent data cycle (i.e., NHANES 2011–2012) was that Asian populations were oversampled, and data for Asians were reported in a separate race category as opposed to being included in the “other” race category (NCHS 2013). Because studies evaluating the health and nutrition status among Asians on a national level are relatively scarce, the addition of the Asian category should allow researchers to investigate the health and nutrition status of this race group. Further, evaluation of exposure characteristics across Asian subgroups could help identify highly exposed subpopulations and also their potential exposure sources.

The objective of the present study was to expand the CDC’s analysis of biomarker data and further evaluate the higher metal biomarker levels among Asians by comparing the biomarker levels of four metals (cadmium, lead, mercury, and arsenic) in Asians with those of other racial and ethnic groups in the United States. We examined variations in biomarker levels of metals in the major Asian subgroups (Chinese and Asian Indian) in the United States and the association of biomarker levels with various demographic, socioeconomic, physical, dietary, behavioral, and geographical characteristics within the subgroups.


Data Source

NHANES data available through the CDC were used as the data source. NHANES recruits approximately 5,000 participants annually, using a complex, multistage, probability sampling design. The multistage sampling procedure includes sampling from four stages of geographical units (county, city block, household, and individual), where subsequent sampling occurs within the unit selected in the prior stage. Multiple samples can be drawn from the same unit (e.g., multiple individuals from one household). Self-reported demographic, socioeconomic, dietary, and health-related information is collected through interview and questionnaire, whereas medical examination and collection of biological specimens (blood/urine) for laboratory tests are administered by health professional and qualified staff at the mobile examination center. The NHANES data collection procedures are described in detail elsewhere (Johnson et al. 2014).

The majority of the NHANES data are publicly available and were obtained directly from the CDC web site (CDC 2015). Access to certain data sets is restricted to protect study participant confidentiality. The restricted data used in this study (i.e., Asian ancestry and geographical information of the participants) were accessed and analyzed at the CDC Research Data Center (RDC), following a strict NCHS protocol (NCHS 2012a). Data collection for NHANES was approved by the NCHS Research Ethics Review Board (ERB). Analysis of de-identified data from the survey is exempt from federal regulations for the protection of human research participants. Analysis of restricted data through the NCHS RDC was also approved by the NCHS ERB.

Study Population

For this study, the study population was the general U.S. population (≥ 6 years of age), including both males and females and all racial and ethnic groups, except those categorized as “other” (i.e., Pacific Islanders, Native Americans/Alaskan Natives, and multiracial individuals). The “other” race group was excluded because of its small sample size and the heterogeneous nature of the group. Additionally, the non-Hispanic Asian group [Far East Asia, Southeast Asia, or South Asia/the Indian subcontinent (NCHS 2013)] was subdivided into Chinese (Chinese and Taiwanese), Asian Indian (Asian Indian, Bengalese, Bharat, Dravidian, East Indian, and Goanese), and Other Asians based on the answer to DMQ.336 in the NHANES’s survey questionnaire. When a participant selected multiple Asian ancestries (e.g., Chinese and Filipino), they were categorized into the “Other Asian” subgroup. Chinese and Asian Indians were selected because they are the two largest Asian subgroups. Each subgroup accounts for approximately 20% of the Asian population (Hoeffel et al. 2012). There was no oversampling of the specific subgroups within the Asian population performed in NHANES 2011–2012.

Biomarker Data

We evaluated five biomarkers: blood cadmium (B-Cd), blood lead (B-Pb), blood mercury (B-Hg), urinary total arsenic (U-tAs) and urinary dimethylarsinic acid (U-DMA). Study participants age ≥ 1 year were eligible for collection of blood samples, whereas urinary samples were obtained from a randomly selected one-third subset of the participants (≥ 6 years old). Arsenic acid, arsenous acid, monomethylarsonic acid (MMA), and DMA are metabolites of inorganic arsenic. Although methylated species such as MMA and DMA can be metabolites of less harmful organic arsenic, these five inorganic arsenic metabolites are often summed to represent inorganic arsenic exposure. Because inorganic arsenic metabolites other than DMA typically have low frequency of detection (< 40%), we only evaluated biomarker levels of U-DMA in our study. Similar to the CDC study of metal biomarkers (CDC 2014), urinary metal concentrations were adjusted using the concentration of creatinine in urine to account for the effect of urinary dilution:

Creatinine-corrected urinary concentration (μg/g) = [100(Lmg/dLg) × metal concentration in urine (μg/L)] ÷ [creatinine in urine (mg/dL)]

For samples with biomarker levels below the limit of detection (LOD), NHANES uses “fill values” (LOD divided by the square root of 2). In accordance with the Fourth National Report on Human Exposure to Environmental Chemicals (CDC 2014), we used these fill values in our analyses. The LOD for biomarker parameters used to establish the fill values were as follows: B-Cd, 0.16 μg/L; B-Pb, 0.25 μg/dL; B-Hg, 0.16 μg/L; U-tAs, 1.25 μg/L; U-DMA, 1.80 μg/L. The detection frequency of B-Cd ranged from 63% among Mexican Americans to 87% among Asians; for U-DMA, the detection frequency ranged from 73% among whites to 91% among Asians. The biomarker levels of three other metal variables presented a relatively high frequency of detection in all groups: B-Pb (≥ 98%), B-Hg (≥ 91%), U-tAs (≥ 91%). The biomarker data were log-transformed to reduce skewness. Detailed information about laboratory procedures including sample collection, storing, and handling of specimens, quality control, and instrument and equipment used for the chemical analyses can be found elsewhere (NCHS 2011a, 2011b, 2012b).


The covariates included in the analyses were sex, age, education, household income, birthplace, poverty–income ratio (PIR) according to the Department of Health and Human Services poverty guidelines (dichotomized based on the median value of 1.63) (DHHS 2013), body mass index (BMI) (underweight, < 18.5 kg/m2; normal weight, 18.5–< 25 kg/m2; overweight, 25–< 30 kg/m2; obese, ≥ 30 kg/m2), smoking (based on the tertile of serum cotinine level), fish consumption, urbanization classification based on 2013 NCHS urban–rural classification scheme for counties (Ingram and Franco 2014), and U.S. Census region. BMI was included based on the association between lower BMI and high B-Hg levels observed in previous studies (Buchanan et al. 2015; Rothenberg et al. 2015). For participants < 20 years of age, education level of the household reference person (frequently, the adult owner/renter of the residence) was used. BMI category was determined based on the CDC’s sex-specific 2000 BMI for-age growth charts for the age group < 20 years (underweight, < 5th percentile; normal weight, 5th–< 85th percentile; overweight, 85th–< 95th percentile; obese, ≥ 95th percentile). Table 1 provides details on the breakdown and response categories of each of these covariates.

Table 1. Select View Table (HTML Version) for a 508-conformant version

Table 1. Characteristics of study participants [n (%) or %] with weighted percentage, NHANES 2011–2012.

View Table (HTML Version)
View larger image (TIF File)

Statistical Analysis

All statistical analyses were performed using SAS-callable SUDAAN version 11.0.1 (RTI International, Research Triangle Park, NC, USA) installed as an add-on to SAS software version 9.3 or higher (SAS Institute Inc., Cary, NC, USA). We accounted for the NHANES’s complex sample design and applied appropriate strata, cluster, and weights, as described in the NHANES documentation (CDC 2015), in all the statistical analyses.

We stratified the data by five NHANES race/ethnic groups: non-Hispanic white, non-Hispanic black, Mexican American, other Hispanic and Asian subgroups (Chinese, Asian Indian, and Other Asian), and computed weighted statistics for biomarker levels by each covariate. The statistics included the geometric mean and its 95% confidence interval (CI), as well as the 50th and 95th percentiles based on the Taylor series linearization method (RTI International 2012). Summary statistics were presented for five biomarker variables [B-Cd, B-Pb, B-Hg, U-TAs (creatinine-corrected), and U-DMA (creatinine-corrected)]. In accordance with the Fourth National Report on Human Exposure to Environmental Chemicals, the geometric mean concentration was not calculated when the level for a biomarker was below the LOD in > 40% of the samples (CDC 2014). For the protection of study participants’ confidentiality, analyses using geographical covariates (urbanization and census region) were not conducted for Asian subgroups.

We compared geometric means of biomarker levels for each covariate category across five NHANES race/ethnic groups and then compared geometric means of biomarker levels across three Asian subgroups, using analysis of variance (ANOVA). Further, differences in geometric means within each covariate were assessed using ANOVA, stratified by NHANES race/ethnic group and Asian subgroup. For all analyses, p < 0.05 was considered statistically significant.


Sample Characteristics

Table 1 presents the study participants’ characteristics by racial/ethnic group. The final number of samples included in the analysis was 6,951 out of 9,756; approximately one-third (2,427) were used for urinary biomarker analyses.

Since differences in biomarker levels may reflect group characteristics such as socioeconomic status and dietary patterns, we first examined the comparability of the various racial/ethnic groups and subgroups by the covariates. The distribution of age groups varied across the racial/ethnic groups. The Asian group had a distribution similar to those of blacks and other Hispanics and tended to be younger than whites and older than Mexican Americans. The Asian group had the highest percentage of college graduates or above. Socioeconomic status (denoting household income and PIR) of the Asian group mirrored that of the white group, with these two groups having higher percentages of the highest income category (> $75,000) and above median PIR than the other three groups. Asians had the lowest percentage of U.S.-born participants (24.8%), compared with > 90% of the white and black populations having been born in the United States. The distributions of recent fish consumers (those who had eaten fish in the 30 days before the study) were generally comparable across the five groups. Large geographical variations existed across the groups. Asians as well as Hispanics and Mexican Americans tended to live in urban areas, with the largest populations of Asian and Mexican-American participants being found in the West.

The weighted percentages of the Asian subgroup samples (Chinese and Asian Indians) were roughly proportional to those observed in the 2010 U.S. Census data (Hoeffel et al. 2012). In general, age groups were distributed similarly. Education and economic status among Chinese and Asian Indians was higher than those of Other Asians. Asian Indians had an approximately 10% lower percentage of U.S.-born individuals than other two subgroups. The proportion of individuals with a normal BMI was highest among Chinese. There was a noticeably higher rate of recent fish consumers in the Chinese and Other Asian subgroups (> 80%) than that of Asian Indians (56.4%).

Analysis of Biomarker Data

Weighted summary statistics of biomarker data (geometric mean and 50th and 95th percentile) are provided in Tables S1–S5 for the five groups and in Tables S6–S10 for the three Asian subgroups.

Overall Comparison across Racial/Ethnic Groups and Asian Subgroups

For all biomarkers, the geometric mean value in Asians was significantly (p < 0.05) higher than that in each of the other racial/ethnic groups (Table 2). This observation was consistent in nearly all of the comparisons performed within subsets of data based on the various demographic, socioeconomic, physical, dietary, behavioral, and geographical characteristics. Biomarker levels among Asians were significantly lower than those of other groups in only two cases: the comparisons of B-Cd and B-Pb levels in U.S.-born individuals (see Tables S1 and S2). For all other comparisons, biomarker levels among Asians are either the highest (mostly significantly) or not significantly different from those of other race/ethnic groups with higher biomarker levels.

Table 2. Select View Table (HTML Version) for a 508-conformant version

Table 2. Comparison of weighted geometric mean biomarker levels across NHANES racial and ethnic group.

View Table (HTML Version)
View larger image (TIF File)

Across the Asian subgroups, biomarker levels were generally similar between the Chinese and Other Asian subgroups (Table 3). The Asian-Indian subgroup had lower biomarker levels than those of the other two Asian subgroups, with the exception of B-Pb. Although the differences in B-Pb levels were not significant, Asian Indians had the highest overall geometric mean B-Pb across the three Asian subgroups. In comparisons made within Asian subgroups, B-Pb levels were significantly higher among Asian Indians for adolescents (12–19 years old) (0.90 μg/dL), older adults (≥ 60 years old) (2.19 μg/dL), those with household income ≥ $75,000 (1.33 μg/dL), and above-mean PIR (1.37 μg/dL) categories than those in the other two Asian subgroups.

Table 3. Select View Table (HTML Version) for a 508-conformant version

Table 3. Comparison of weighted geometric mean biomarker levels across Asian subgroup.

View Table (HTML Version)
View larger image (TIF File)

Predictors of Biomarker Levels in Asian Subgroups

Cadmium. Sex was significantly associated with B-Cd levels in two of the three Asian subgroups. Females had higher B-Cd levels than males across all subgroups (Table 3). A general trend of increasing B-Cd with age was observed. There was an apparent inverse trend with socioeconomic status (education, income, and PIR) and B-Cd levels. B-Cd levels were significantly higher in individuals born outside of the United States, compared with those born in the United States in all of the Asian subgroups. A clear trend of B-Cd levels increasing with cotinine levels was observed in all subgroups.

Lead. B-Pb levels were significantly associated with sex. B-Pb levels were significantly higher among males than females in all three Asian subgroups (Table 3). B-Pb level generally increased with age. There was a general trend of decreasing B-Pb levels with higher educational status. Individuals born outside of the United States had higher B-Pb levels than those born in the United States across all of the Asian subgroups. A clear trend of B-Pb levels increasing with cotinine levels was observed in all subgroups.

Mercury. A general trend of increasing B-Hg levels with age was observed, with the exception of the Asian-Indian subgroup (Table 3). Significant differences in B-Hg across BMI categories were observed among Chinese and Other Asian subgroups, although no consistent pattern of B-Hg was seen between these two subgroups. Recent fish consumers had higher B-Hg levels than non-consumers in all three Asian subgroups.

Arsenic, total. The general patterns of the U-tAs levels across age groups were similar in all Asian subgroups (Table 3). U-tAs levels decreased from the youngest group (6–11 years) to the second youngest age group (12–19 years) and then generally increased with age after childhood (≥ 12 years). U-tAs levels were significantly higher among recent fish consumers than non-consumers in all three Asian subgroups.

DMA. The patterns of the U-DMA levels across age groups were similar to those of the U-tAs (Table 3). U-DMA levels were often higher among the youngest age group (6–11 years) than those among other age groups. Across the age groups (≥ 12 years), there was a general trend of increasing U-DMA levels with age. Recent fish consumers had higher U-DMA levels than non-consumers in all three Asian subgroups.


Our study confirmed there are racial/ethnic differences in the biomarker levels of toxic metals—cadmium, lead, mercury, and arsenic—in the United States. Overall, biomarker levels among Asians were higher than in other racial/ethnic groups regardless of sociodemographic, physical, behavioral, dietary, and geographic characteristics (see Tables S1–S5). Asians had significantly lower biomarker levels than other groups in only two comparisons: a) The B-Cd among U.S.-born blacks was significantly higher than that among U.S.-born Asians, and b) U.S.-born whites and blacks had significantly higher B-Pb levels than U.S.-born Asians. Across the Asian subgroups, the lowest biomarker levels were generally observed among Asian Indians, except for B-Pb levels. Although no significant difference was observed in the overall comparison of B-Pb levels across Asian subgroups (≥ 6 years old), significantly higher B-Pb levels among Asian Indians were found in adolescents (12–19 years old), older adults (≥ 60 years old), people in the highest income category (≥ $75,000), and people above the median PIR. The elevated B-Pb levels in Asian Indians may be associated with their spice and cosmetic use, since elevated levels of lead have been found in turmeric (Gleason et al. 2014), a main ingredient of curry, and in eye makeup, such as surma or kohl, that are often used in Indian communities (Goswami 2013).

In general, biomarker levels among Asians in the United States were lower than the levels reported in studies conducted in Asian countries. Ding et al. (2014) evaluated the B-Cd and B-Pb levels of the general population in China, based on randomly selected study participants aged 6–60 years old (n = 18,120) from 24 districts in eight provinces in China between 2009 and 2010. Geometric mean B-Cd and B-Pb levels from this study were 0.49 μg/L and 3.49 μg/dL, respectively, compared with the geometric mean B-Cd (0.45 μg/L) and B-Pb (1.22 μg/L) levels observed among the Chinese subgroup in the present study (Table 4). Geometric mean blood biomarker levels (2011) reported in the Korea NHANES (Seo et al. 2015), a Korean national health survey similar to the CDC’s NHANES, were slightly higher, but comparable with the levels observed among the Other Asian subgroup, which is assumed to consist mainly of Filipino, Vietnamese, Korean, and Japanese according to the 2010 Census (Hoeffel et al. 2012). The geometric mean blood biomarker levels among those Koreans ≥ 19 years were 0.86 μg/L (B-Cd), 1.99 μg/dL (B-Pb), and 3.08 μg/L (B-Hg) (Table 4). In our study, the ranges of the geometric mean of B-Cd, B-Pb, and B-Hg levels in the corresponding age group (≥ 20 years old) of Other Asians were 0.42–0.74 μg/L (B-Cd), 0.92–1.53 μg/dL (B-Pb), and 2.18–3.80 μg/L (B-Hg). Urinary arsenic levels in Koreans were noticeably higher than the levels observed in the present study. Geometric mean U-tAs levels reported in the Korea NHANES (2008–2009) ranged from 90.6 μg/g-creatinine (20–39 years old) to 157.6 μg/g-creatinine (≥ 60 years old) (Rhee et al. 2013), whereas U-tAs levels observed in our study were 24.21 μg/g-creatinine (20–39 years old) to 52.85 μg/g-creatinine (≥ 60 years old) among the Other Asian subgroup (Table 4).

Table 4. Select View Table (HTML Version) for a 508-conformant version

Table 4. Comparison of geometric mean biomarker levels of Asian subgroups in the U.S. to those reported in Asian countries.

View Table (HTML Version)
View larger image (TIF File)

Except for lead, the exposure pathway of the metals we evaluated is known to be predominantly food consumption for the general population. Seafood is the major source of dietary exposure to mercury (methylmercury) and arsenic (total) (ATSDR 1999, 2007a). In addition to seafood, cereal grains (including rice) and poultry are the major contributors to dietary arsenic exposure in the United States (Tsuji et al. 2007; Vogt et al. 2012; Xue et al. 2010). Smoking is the main source of cadmium exposure (ATSDR 2012), though exposure to cadmium for nonsmokers occurs mostly through diet, such as consumption of vegetables and cereal grains (Egan et al. 2007; He et al. 2013). Sources of lead exposure include environmental exposure through lead-containing dust and soil from hazardous waste sites, highways, and old fruit orchards; smoking; drinking water from old plumbing systems; inhalation or direct contact with lead-based paint; and ingestion of food from lead-glazed potteries or dishes (ATSDR 2007b). In this study, recent fish consumption was a significant predictor for B-Hg and U-tAs levels. In addition, positive dose–response relationships were found for cotinine levels (an indicator of smoking) and both B-Cd and B-Pb in each of the Asian subgroups.

Further, our study found that several other characteristics are important predictors of biomarker levels. Sex and age differences in biomarker levels were generally consistent across Asian subgroups. Females had higher B-Cd and lower B-Pb levels than males. Biomarker levels generally increased with age. A higher level of U-tAs and U-DMA were observed in the youngest age group (6–11 years). This may be attributable to greater arsenic exposure and/or age-dependent toxicokinetc characteristics (e.g., efficient absorption or poor excretion of arsenic) of this age group. Additionally, we found birthplace to be an important predictor of biomarker levels: consistently higher biomarker levels (albeit not always significant) were observed among Asians born outside of the United States compared with Asians born in the United States. Although higher, the biomarker levels among non–U.S.-born Asians are less than the levels reported in their countries of origins described in the previous paragraph. Further, as discussed earlier, within the comparisons among U.S.-born individuals, Asians had significantly lower B-Cd and B-Pb than those of other racial/ethnic groups. A further characterization of metal exposure depending on birthplace and its relationship with biomarker levels will be warranted in future studies. These patterns of biomarker levels based on sex, age, and birthplace among Asians agreed with the results reported in previous studies based on the general U.S. population (Caldwell et al. 2009; Mortensen et al. 2014; Peters et al. 2014). In contrast, there appear to be different patterns of B-Hg and U-tAs among Asians for the covariates representing socioeconomic status. A general trend of increasing B-Hg and U-tAs with increasing educational and socioeconomic status was observed among the racial/ethnic groups other than Asians, with this trend being more pronounced in the white group. This result was consistent with the results of previous studies (Buchanan et al. 2015; McKelvey et al. 2007; Mortensen et al. 2014). It is typically explained that individuals with higher incomes and/or educational achievement can afford to add larger fish (e.g., tuna, swordfish), which tend to have higher mercury content, to their diet (Hightower and Moore 2003; Mortensen et al. 2014). However, this trend was reversed among the Asian population. One possible explanation for this difference is that Asians of lower socioeconomic status may consume fish containing higher levels of mercury. For example, some economically disadvantaged Asian subgroups may be more likely to engage in subsistence fishing and consume locally harvested fish that have higher levels of environmental contaminants.

There are several limitations associated with the present study. First, because of the cross-sectional design of the NHANES, the data represent only a snapshot of biomarker levels on the day of examination. Similarly, some of the covariates (smoking based on cotinine levels, and fish consumption) only reflect the participants’ living environment or food consumption patterns immediately before the survey, and may not represent their long-term exposure. Second, the toxic metals evaluated in this study have different half-lives in the human body. Cadmium is not readily excreted and has a long biological half-life (as long as 38 years) (ATSDR 2008). Although the biological half-life of lead is approximately 30 days, it tends to accumulate in the bones and soft tissues over a long time and is released very slowly (ATSDR 2010). Mercury, predominantly present in the blood as methylmercury, has a half-life of approximately 2 months (ATSDR 1999). Therefore, the blood biomarkers for cadmium and lead may be indicators suitable for the body burden after long-term exposure. The biological half-life of arsenic is fairly short, roughly 2–3 days (ATSDR 2007a). Because urinary biomarkers have short half-lives and reflect short-term exposure, they tend to vary more depending on the study participants’ food consumption, living environment, and occupational exposure immediately before the sampling. Also, because the information related to fish consumption is self-reported, it is subject to recall bias. Further, following the CDC’s analytical approach, urinary biomarker data corrected using urinary creatinine level were used in our study. Urinary creatinine levels vary depending on various factors such as sex, muscle mass, diet, and health conditions. Our supplemental comparisons (Tables S1–S10) of urinary creatinine levels across the racial/ethnic groups indicate lower levels of urinary creatinine among Asians than the other groups. These differences are also attributable to the higher U-tAs and U-DMA among Asians observed in this study. Our analysis evaluated the association between biomarker levels and a limited number of covariates representing study participants’ demographic, socioeconomic, physical, behavioral, and dietary characteristics. Covariates characterizing food consumption patterns were limited to fish intake; we did not include other important food sources of metal exposures. For instance, a significant association between biomarker levels of arsenic (both total and inorganic) and rice consumption has been reported (Davis et al. 2012; Wei et al. 2014); we did not analyze this. Further, we did not include covariates representing study participants’ living environment or occupational exposure in our study. A recent study based on NHANES data suggests that occupation is a significant predictor of blood lead and blood cadmium levels (Peters et al. 2014). Lead paint and use of lead-containing pottery may be important sources of environmental lead exposure. Inclusion of these covariates may have improved our characterization of metal exposure. Furthermore, there may be race/ethnicity specific differences in frequency of genetic variants that influence absorption, distribution, metabolism, elimination/excretion processes and such differences could also be related to differences in biomarker levels of metals across groups.

Another uncertainty associated with the current study is how representative our sample was of the Asian population. Asians typically have a lower participation rate in national surveys than other racial/ethnic groups, and the NHANES response rate among Asian in 2011 was approximately 10–20% lower than that of other groups (Broitman 2012). Because of potential response bias, the NCHS performed an analysis of nonresponders by comparing the demographic and socioeconomic characteristics of responders and nonresponders (NCHS 2013). Based on this analysis, the NCHS concluded that, although a potential for nonresponse bias may exist, weight adjustment lessens the bias. Our analysis used appropriate sample weights; however, it still remains uncertain to what extent this potential bias may have remained and distorted the results. Furthermore, we used biomarker levels of Asians from one NHANES data cycle. The Asian group was divided into three subgroups, and the results are based on a relatively small number of samples. Therefore, some of our results may be statistically unreliable and should be viewed with caution. Because oversampling of the Asian population continues in the next NHANES data cycle (2013–2014), the findings of this study should be verified with the larger data set in future studies.

This study also had several strengths. We evaluated differences in biomarker levels of five metals across different racial/ethnic groups in the United States, with a specific interest in the Asian population, due to previously reported elevated concentrations of metal biomarkers in this group. The NHANES 2011–2012 is the first data cycle to include a specific Asian race category, and to the best of our knowledge, this is one of the first studies to investigate biomarker levels in this historically less-studied racial group using nationally representative data. We evaluated biomarker levels of three subgroups of the Asian population: Chinese, Asian Indian, and Other Asian. Although NHANES is not designed to evaluate small sample groups and the results are not nationally representative, our study was able to assess general biomarker patterns among subgroups of Asians, which have rarely been evaluated, especially on a national scale.

According to the 2010 U.S. Census (U.S. Census Bureau 2013), Asians were the fastest-growing race/ethnic group in the United States with an increase of 43.2% between 2000 and 2010. As this study demonstrated, there are considerable variations in sociodemographic, behavioral, and exposure characteristics between Asians and other racial/ethnic groups and also between Asian subgroups. As the Asian population in the United States continues to grow, more studies are warranted to improve our understanding of the health and nutritional status of this minority group.


Asian populations were found to have the highest levels of B-Cd, B-Pb, B-Hg, U-tAs, and U-DMA across the five racial/ethnic groups assessed in the NHANES. Generally, this observation did not change when data were further examined by various demographic, socioeconomic, physical, dietary, behavioral, and geographical characteristics. Within the Asian group, considerable variations in biomarker levels are present across the Chinese, Asian Indian, and Other Asian subgroups. Biomarker levels of toxic metals, except B-Pb, are generally lowest among Asian Indians. Sex, age, education, birthplace, smoking, and fish consumption were found to be significant predictors of biomarker levels for certain metals.


Anderson EL. 1983. Quantitative approaches in use to assess cancer risk. Risk Anal 3(4):277–295.

ATSDR (Agency for Toxic Substances and Disease Registry). 1999. Toxicological Profile for Mercury.​6.pdf [accessed 23 October 2014].

ATSDR. 2007a. Toxicological Profile for Arsenic.​.pdf [accessed 23 October 2014].

ATSDR. 2007b. Toxicological Profile for Lead.​3.pdf [accessed 23 October 2014].

ATSDR. 2008. Case Studies in Environmental Medicine (CSEM). Cadmium Toxicity. WB 1096.​cs/cadmium.pdf [accessed 20 November 2015].

ATSDR. 2010. Case Studies in Environmental Medicine (CSEM). Lead Toxicity. WB 1105.​lead.pdf [accessed 17 September 2015].

ATSDR. 2012. Toxicological Profile for Cadmium.​.pdf [accessed 12 March 2015].

Broitman L. 2012. Ongoing Activities to Support the Asian Oversample in NHANES.​14_broitman.pdf [accessed 9 October 2014].

Buchanan S, Anglen J, Turyk M. 2015. Methyl mercury exposure in populations at risk: analysis of NHANES 2011–2012. Environ Res 140:56–64.

Caldwell KL, Jones RL, Verdon CP, Jarrett JM, Caudill SP, Osterloh JD. 2009. Levels of urinary total and speciated arsenic in the US population: National Health and Nutrition Examination Survey 2003–2004. J Expo Sci Environ Epidemiol 19(1):59–68.

CDC (Centers for Disease Control and Prevention). 2014. Fourth National Report on Human Exposure to Environmental Chemicals. Updated Tables. FourthReport_UpdatedTables_Feb2015.pdf [accessed 27 July 2014].

CDC. 2015. National Health and Nutrition Examination Survey. [accessed 10 October 2015].

Davis MA, Mackenzie TA, Cottingham KL, Gilbert-Diamond D, Punshon T, Karagas MR. 2012. Rice consumption and urinary arsenic concentrations in U.S. children. Environ Health Perspect 120:1418–1424, doi: 10.1289/ehp.1205014.

DHHS (U.S. Department of Health and Human Services). 2013. Poverty Guidelines, Research, and Measurement. Washington, DC:DHHS.​ines [accessed 23 September 2015].

Ding C, Pan Y, Zhang A, Wu B, Huang H, Zhu C, et al. 2014. Study of distribution and influencing factors of lead and cadmium in whole blood and urine among population in 8 provinces in China [in Chinese]. Zhonghua Yu Fang Yi Xue Za Zhi 48(2):91–96.

Egan SK, Bolger PM, Carrington CD. 2007. Update of US FDA’s Total Diet Study food list and diets. J Expo Sci Environ Epidemiol 17(6):573–582.

Ferraro PM, Costanzi S, Naticchia A, Sturniolo A, Gambaro G. 2010. Low level exposure to cadmium increases the risk of chronic kidney disease: analysis of the NHANES 1999–2006. BMC Public Health 10:304, doi: 10.1186/1471-2458-10-304.

Gleason K, Shine JP, Shobnam N, Rokoff LB, Suchanda HS, Ibne Hasan MOS, et al. 2014. Contaminated turmeric is a potential source of lead exposure for children in rural Bangladesh. J Environ Public Health 2014:730636, doi: 10.1155/2014/730636.

Goswami K. 2013. Eye cosmetic ‘surma’: hidden threats of lead poisoning. Indian J Clin Biochem 28(1):71–73.

He P, Lu Y, Liang Y, Chen B, Wu M, Li S, et al. 2013. Exposure assessment of dietary cadmium: findings from Shanghainese over 40 years, China. BMC Public Health 13(1):590, doi: 10.1186/1471-2458-13-590.

Hightower JM, Moore D. 2003. Mercury levels in high-end consumers of fish. Environ Health Perspect 111:604–608, doi: 10.1289/ehp.5837.

Hoeffel EM, Rastogi S, Kim MO, Shahid H. 2012. The Asian Population: 2010. 2010 Census Briefs. C2010BR-11.​fs/c2010br-11.pdf [accessed 21 October 2014].

IARC (International Agency for Research on Cancer). 2013. IARC Monographs on the Evaluation of Carcinogenic Risks to Humans: Agents Classified by the IARC Monographs, Volumes 1–117.​tion/ [accessed 20 December 2013].

Ingram DD, Franco SJ. 2014. 2013 NCHS urban–rural classification scheme for counties. Vital Health Stat 2 166:1–73.

Jakubowski M. 2011. Low-level environmental lead exposure and intellectual impairment in children—the current concepts of risk assessment. Int J Occup Med Environ Health 24(1):1–7.

Johnson CL, Dohrmann SM, Burt VL, Mohadjer LK. 2014. National Health and Nutrition Examination Survey: sample design, 2011–2014. Vital Health Stat 2 162:1–33.

Lebel J, Mergler D, Lucotte M, Amorim M, Dolbec J, Miranda D, et al. 1996. Evidence of early nervous system dysfunction in Amazonian populations exposed to low-levels of methylmercury. Neurotoxicology 17(1):157–167.

McKelvey W, Gwynn RC, Jeffery N, Kass D, Thorpe LE, Garg RK, et al. 2007. A biomonitoring study of lead, cadmium, and mercury in the blood of New York City adults. Environ Health Perspect 115:1435–1441, doi: 10.1289/ehp.10056.

McLaine P, Navas-Acien A, Lee R, Simon P, Diener-West M, Agnew J. 2013. Elevated blood lead levels and reading readiness at the start of kindergarten. Pediatrics 131(6):1081–1089.

Moon KA, Guallar E, Umans JG, Devereux RB, Best LG, Francesconi KA, et al. 2013. Association between exposure to low to moderate arsenic levels and incident cardiovascular disease. A prospective cohort study. Ann Intern Med 159(10):649–659.

Mortensen ME, Caudill SP, Caldwell KL, Ward CD, Jones RL. 2014. Total and methyl mercury in whole blood measured for the first time in the U.S. population: NHANES 2011–2012. Environ Res 134:257–264.

NCHS (National Center for Health Statistics). 2011a. Laboratory Procedure Manual. Arsenobetaine, Arsenocholine, Trimethylarsine Oxide, Monomethylarsonic Acid, Dimethylarsinic Acid, Arsenous (III) Acid, Arsenic (V) Acid. Hyattsville, MD:NCHS.

NCHS. 2011b. Laboratory Procedure Manual. Creatinine. Hyattsville, MD:NCHS.​es_11_12/biopro_g_met_creatinine.pdf [accessed 15 October 2015].

NCHS. 2012a. Disclosure Manual. Preventing Disclosure: Rules for Researchers. Hyattsville, MD:NCHS.​emanual.pdf [accessed 16 July 2015].

NCHS. 2012b. Laboratory Procedure Manual. Cadmium, Lead, Manganese, Mercury, and Selenium. Hyattsville, MD:NCHS.​es_11_12/pbcd_g_met_blood-metals.pdf [accessed 15 October 2015].

NCHS. 2013. National Health and Nutrition Examination Survey: Analytic Guidelines, 2011–2012. Hyattsville, MD:NCHS.​ytic_guidelines_11_12.pdf [accessed 15 October 2015].

Peters JL, Fabian MP, Levy JI. 2014. Combined impact of lead, cadmium, polychlorinated biphenyls and non-chemical risk factors on blood pressure in NHANES. Environ Res 132:93–99.

Rhee SY, Hwang YC, Woo JT, Chin SO, Chon S, Kim YS. 2013. Arsenic exposure and prevalence of diabetes mellitus in Korean adults. J Korean Med Sci 28(6):861–868.

Rothenberg SE, Korrick SA, Fayad R. 2015. The influence of obesity on blood mercury levels for U.S. non-pregnant adults and children: NHANES 2007–2010. Environ Res 138:173–180.

RTI International. 2012. SUDAAN 11. Language Manual. 2012; 1 and 2 (Release 11). Research Triangle Park, NC:Research Triangle Institute.

Seo JW, Kim BG, Kim YM, Kim RB, Chung JY, Lee KM, et al. 2015. Trend of blood lead, mercury, and cadmium levels in Korean population: data analysis of the Korea National Health and Nutrition Examination Survey. Environ Monit Assess 187(3):146, doi: 10.1007/s10661-015-4348-2.

Tsuji JS, Yost LJ, Barraj LM, Scrafford CG, Mink PJ. 2007. Use of background inorganic arsenic exposures to provide perspective on risk assessment results. Regul Toxicol Pharmacol 48(1):59–68.

U.S. Census Bureau. 2013. Asians Fastest-Growing Race Or Ethnic Group in 2012, Census Bureau Reports. CB13-112.​eases/2013/cb13-112.html [accessed 13 June 2013].

Vogt R, Bennett D, Cassady D, Frost J, Ritz B, Hertz-Picciotto I. 2012. Cancer and non-cancer health effects from food contaminant exposures for children and adults in California: a risk assessment. Environ Health 11:83, doi: 10.1186/1476-069X-11-83.

Wei Y, Zhu J, Nguyen A. 2014. Rice consumption and urinary concentrations of arsenic in US adults. Int J Environ Health Res 24(5):459–470.

Xue J, Zartarian V, Wang SW, Liu SV, Georgopoulos P. 2010. Probabilistic modeling of dietary arsenic exposure and dose and evaluation with 2003–2004 NHANES data. Environ Health Perspect 118:345–350, doi: 10.1289/ehp.0901205.

Association of Dietary Intake and Biomarker Levels of Arsenic, Cadmium, Lead, and Mercury among Asian Populations in the United States: NHANES 2011–2012

Author Affiliations open
1University of Texas Health Science Center at Houston School of Public Health, Houston, Texas, USA; 2CH2M HILL, Inc., Houston, Texas, USA

PDF icon PDF Version (1.1 MB)

  • Background: We have recently shown that biomarker levels of selected metals are higher in Asians than in other U.S. ethnic groups, with important differences within selected Asian subgroups. Much of this difference may be dietary in origin; however, this is not well established.

    Objective: We evaluated dietary intake of toxic metals as a source of increased biomarker levels of metals among U.S. Asians.

    Methods: We estimated daily food consumption and dietary intake of arsenic, cadmium, lead, and mercury by combining 24-hr dietary intake recall data from the 2011–2012 National Health and Nutrition Examination Survey (NHANES) with data from the USDA Food Composition Intake Database and FDA Total Dietary Study. We analyzed associations between dietary metal intake and biomarker levels of the metals using linear regression. Further, estimated food consumption and metal intake levels were compared between Asians and other racial/ethnic groups (white, black, Mexican American, and other Hispanic) and within three Asian subgroups (Chinese, Indian Asian, and other Asians).

    Results: Significant associations (p < 0.05) were found between biomarker levels and estimated dietary metal intake for total and inorganic arsenic and mercury among Asians. Asians had the highest daily fish and rice consumption across the racial/ethnic groups. Fish was the major contributor to dietary mercury and total arsenic intake, whereas rice was the major contributor to inorganic arsenic dietary intake. Fish consumption across the Asian subgroups varied, with Asian Indians having lower fish consumption than the other Asian subgroups. Rice consumption was similar across the Asian subgroups.

    Conclusions: We confirmed that estimated dietary intake of arsenic (total and inorganic) and mercury is significantly associated with their corresponding biomarkers in U.S. Asians, using nationally representative data. In contrast, estimated dietary intake of cadmium and lead were not significantly associated with their corresponding biomarker levels in U.S. Asians.

  • Citation: Awata H, Linder S, Mitchell LE, Delclos GL. 2017. Association of dietary intake and biomarker levels of arsenic, cadmium, lead, and mercury among Asian populations in the United States: NHANES 2011–2012. Environ Health Perspect 125:314–323;

    Address correspondence to H. Awata, c/o CH2M, 14701 St Mary’s Lane, Suite 300, Houston, TX 77079 USA. Telephone: (281) 721-8487.

    The authors thank dietetic specialists J. Chang and A. Hostler for their contribution to the study.

    The data analyses of the restricted data at the Research Data Center (RDC) in Atlanta were supported by department research funds from the University of Texas Health Science Center School of Public Health.

    The findings and conclusions in this paper are those of the author and do not necessarily represent the views of the RDC, the National Center for Health Statistics, or the Centers for Disease Control and Prevention.

    The authors declare they have no actual or potential competing financial interests.

    Received: 24 February 2016
    Revised: 10 June 2016
    Accepted: 18 July 2016
    Published: 2 September 2016

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (520 KB)

    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact Our staff will work with you to assess and meet your accessibility needs within 3 working days.

    PDF icon Supplemental Table of Contents PDF (93 KB)


Biomarker levels of toxic metals/metalloids (hereafter, simply referred to as “metals”), such as arsenic, cadmium, lead, and mercury, are higher among the Asian population than other racial/ethnic groups in the United States. A publication from the Centers for Disease Control and Prevention (CDC) reported that Asians had biomarker levels of these metals up to four times higher than other racial/ethnic groups (CDC 2014). For instance, the geometric mean blood mercury (total) level among Asians was 1.86 μg/L as compared with 0.48 μg/L in Mexican Americans.

Arsenic, cadmium, lead, and mercury are well known toxic environmental contaminants. U.S. and international environmental and public health agencies have classified inorganic arsenic and cadmium as human carcinogens (IARC 2013; U.S. EPA 2015). Further, exposure to these metals has been associated with a number of adverse health effects, including developmental and nervous system damage (ATSDR 1999, 2007a, 2007b, 2012). Hence, these elevated biomarker levels reflect a potentially increased health risk among the Asian population.

Food consumption is considered one of the predominant exposure pathways of these toxic metals. These metals are bioaccumulative and ubiquitous in the environment. Although mitigation efforts in the United States over the past few decades have largely succeeded in controlling their release into the environment, they are still detectable in many foods. Several of these foods are consumed by Asian Americans in large amounts. For instance, elevated levels of mercury and arsenic (total) in seafood and arsenic (inorganic) in grains (e.g., rice) have been reported (FDA 2014a, 2014b). These foods are staples of the Asian diet. However, studies characterizing dietary intake levels of these metals among the Asian populations (i.e., the populations that appear to be at highest risk of exposure based on biomarker studies) in the United States were conducted mostly in cohorts from geographic areas with high Asian populations. Consequently, our understanding of dietary exposure characteristics of Asians on a national scale is fairly limited.

To fill this gap, we evaluated dietary intake of these metals in the United States, based on nationally representative data. We evaluated the association between dietary metal intake and biomarker levels across various racial/ethnic groups (Asian, white, black, Mexican American, and other Hispanic). In addition, because Asians in the United States comprise several different ethnic subpopulations that may have different dietary patterns, we evaluated these associations across two major Asian subgroups (Chinese and Asian Indian). Finally, we examined variations in food consumption and dietary metal intake across these same groups and subgroups to identify the foods that contribute most to their overall dietary metal intake.


Study Population

The National Health and Nutrition Examination Survey (NHANES) was used as the primary data source for this study. The NHANES is a national population-based survey program assessing the health and nutritional status of the civilian noninstitutionalized general U.S. population. Health and nutrition data are collected each year from approximately 5,000 survey participants, selected using a complex, multistage, probability sampling design (Johnson et al. 2014). The multistage sampling procedure is comprised of four stages of geographical unit selection. It starts with a selection of the primary sampling units (typically at the county level) and then selects smaller geographical units (city blocks and then households) within the units at each subsequent stage. At the final stage, more than one individual is often drawn from a single household. The NHANES data are released every 2 years.

This study used the data from the most recent data cycle (2011–2012) because it was the first NHANES cycle to oversample Asians. The non-Hispanic Asian category includes individuals with self-reported origins in the Far East Asia, Southeast Asia, or South Asia (the Indian subcontinent) (NCHS 2013). We further subcategorized Asians into the two largest Asian subgroups: Chinese (Chinese and Taiwanese) and Asian Indian (Asian Indian, Bengalese, Bharat, Dravidian, East Indian, and Goanese), and combined all the remaining Asians into an “Other Asian” subgroup, to investigate variations across Asian subgroups. No specific subgroups within the Asian population were oversampled in NHANES 2011–2012.

Sociodemographic, dietary, and biomarker data from the NHANES 2011–2012 cycle are publicly available and were obtained directly from the CDC website. Because access to data on Asian ancestry and geographical information is restricted, analyses of these variables were conducted at the CDC Research Data Center (RDC) in Atlanta, Georgia (NCHS-RDC 2012) following review and approval by the NCHS Research Ethics Review Board.

Estimation of Food Consumption and Dietary Metal Intake

We used three data sets to estimate dietary metal intake. A brief description of each data set is provided below and is summarized in Table 1.

Table 1. Select View Table (HTML Version) for a 508-conformant version

Table 1. Summary of data sources used in the estimation of dietary metal intake.

View Table (HTML Version)
View larger image (TIF File)

Consumption data. The NHANES food consumption data were used to estimate the types and amounts of food consumed by study participants (USDA 2014). These data were collected using an interviewer-administered questionnaire that included a 24-hr dietary recall instrument. The interview was administered on two nonconsecutive survey days, 3–10 days apart. During the interview, food items were recorded “as consumed” (e.g., meat lasagna), rather than on an individual food component basis (e.g., tomato). We limited our analyses to data from individuals with body weight and food-consumption data available for days 1 and 2. Specific information about collection and processing of the food consumption data is provided online (USDA 2014).

Composition data. The composition of each food item was determined using the Food Commodity Intake Database (FCID) created by the U.S. Environmental Protection Agency (U.S. EPA) and U.S. Department of Agriculture (USDA) (U.S. EPA 2014). This database provides the amount of each individual food ingredient, hereafter referred to as “food commodity,” included in 100 g of each food reported specifically in NHANES. For instance, the FCID food commodities in 100 g of “meat lasagna” include 20.9 g of tomato, 16.2 g of wheat flour, and 7.9 g of beef.

The most recent FCID (FCID 2005–2010) has been updated to the previous NHANES data cycle (2009–2010) but does not include food items added to the 2011–2012 NHANES dietary data. Thus, we identified food items in the current FCID that most closely represent the new food items in terms of food description and composition (U.S. EPA 2014) and used their composition data for these newly added food items. We also used the USDA’s cross-reference information—which presents the changes in the food coding due to expansion, consolidation, and renumbering of coding system between current and previous NHANES data cycles—for this selection process (USDA 2015). For a small number of the food items (< 1% of food items reported in the dietary consumption data) for which a representative food item was not identified, new composition data were created based on the existing data for similar food items and/or the USDA’s Food and Nutrient Database for Dietary Studies (food composition data for nutritional studies) (USDA 2015). We assigned no composition data to food items that were not of interest to the present study (e.g., water, energy and alcoholic drinks, condiments).

Chemical data. The U.S. Federal Drug Administration (FDA) Total Dietary Study (TDS) (FDA 2013) was used as our main source of chemical concentration data. The TDS is a continuous food-safety monitoring program, in which food samples are collected using an approach called “market basket.” The data are collected in three cities from each of four regions of the nation, with one region sampled each quarter (spring in the South, summer in the Northeast, autumn in the North Central region, and winter in the West) (FDA 2015b). Samples of food, collected directly from retail stores and fast-food restaurants in each region, are compiled to create a market basket representing the average U.S. diet.

For this study we used 2006–2011 TDS data. For food commodities that were not included in the 2006–2011 TDS data set, we used TDS data from previous years (1991–2005). The TDS’s effort to analyze mercury in food is targeted at fish and other seafood; therefore, mercury data in other food groups are fairly limited. Specific information about laboratory procedures used in the TDS is presented elsewhere (FDA 2015a).

The TDS does not include data on inorganic arsenic. Consequently, data for inorganic arsenic were obtained from Schoof et al. (1999), who applied a modified market basket survey approach and collected inorganic arsenic data from 40 foods that were expected to contribute to at least 90% of dietary inorganic intake in the general U.S. population (Schoof et al. 1999).

Two dietetic specialists linked food commodities between the composition and chemical data. To focus this effort, food commodities consumed in the largest quantities were determined for each of the five NHANES racial/ethnic groups. Food commodities found to make up 95% of the diet for at least one racial/ethnic group were identified as the target food commodities. Initially, each dietitian linked one-half of the target food commodities. These linkages were then reviewed by both dietitians, with the final linkage based on their consensus decision.

Estimation approach. We estimated daily food consumption and dietary metal intake, generally following the approach presented in the study conducted by Yost et al. (2004):

  • We translated NHANES’s food consumption data, presented “as consumed,” into food commodities using the food composition database.
  • Total daily consumption of each food commodity was estimated as the sum of all meals (including snacks) in a 24-hr period. The estimated daily food commodity consumption was divided by survey participant’s body weight to obtain body weight-adjusted daily commodity consumption. Food commodity consumption was calculated for both day 1 and day 2, and the 2-day average was used for daily commodity consumption of each survey participant.
  • We then estimated food-category consumption by summing the calculated commodity-specific food consumption, obtained in step 2, by 14 major food categories (vegetables, fruits, mushroom, nuts, herbs and spices, cereal grains, beef, pork, poultry, other meat, fish, dairy, egg, and oil) and additional subcategories under cereal grains (white rice and brown rice) and fish (freshwater fish, saltwater fish, and shellfish).
  • Based on the chemical data, we estimated daily dietary metal intake by multiplying the daily commodity consumption, obtained in step 2, by metal concentration in the food commodity. In accordance with the Fourth National Report on Human Exposure to Environmental Chemicals (CDC 2014), the level of detection (LOD) divided by the square root of 2 was used as the “fill value” for subjects with non-detected results.
  • Similar to step 3, we estimated food-category dietary metal intake by summing the calculated commodity-specific, daily, dietary metal intake, obtained in step 4, by 14 major categories and sub-categories.
  • Last, we estimated total individual dietary metal intake by summing all of the calculated commodity-specific daily dietary metal intakes, obtained in step 4, for each person.

Biomarker Data

Biomarker data for blood cadmium (B-Cd), blood lead (B-Pb), and blood mercury (B-Hg), as well as urinary total arsenic (U-tAs) and urinary dimethylarsinic acid (U-DMA) were obtained from NHANES. Biomarker data were collected on day 1 of the 2 nonconsecutive food consumption survey days. Blood biomarker samples were collected from survey participants age ≥ 1 year, whereas urinary biomarker samples were collected from a randomly selected one-third of participants, ages ≥ 6 years. Urinary biomarker data were adjusted for creatinine to address the effect of urinary dilution, as computed in the CDC document (CDC 2014). Inorganic arsenic is excreted as inorganic arsenic and methylated metabolites (e.g., monomethylarsonic acid, DMA). Although these methylated arsenic species can also be metabolites of organic arsenic, a sum of these metabolites is commonly used to represent inorganic arsenic exposure (Davis et al. 2012; Wei et al. 2014). As DMA is the predominant detectable inorganic arsenic metabolite, U-DMA concentrations were used in the evaluation of dietary inorganic arsenic intake data. The detection frequency of biomarker data as follows: B-Cd (73%); B-Pb (99%); B-Hg (94%); U-tAs (≥ 96%); U-DMA (≥ 79%). For samples with non-detectable biomarker levels, we used the LOD divided by the square root of 2, as reported in the NHANES data. More information regarding laboratory procedures used for the chemical analyses are presented in the National Center for Environmental Health’s Laboratory Procedure Manuals (NCHS 2011a, 2011b, 2012).

Statistical Analyses

We performed all statistical analyses using SAS-callable SUDAAN version 11.0.1 (RTI International, Research Triangle Park, NC, USA). SUDAAN was installed as an add-on to SAS software version 9.3 or higher (SAS Institute Inc., Cary, NC, USA). The data were stratified by five NHANES racial/ethnic groups (Asian, white, black, Mexican American, and other Hispanic) and three Asian subgroups (Chinese, Asian Indian, and Other Asian), and the results were presented for each group. All the statistical analyses accounted for the NHANES’s complex sample design and weighting.

Analysis of associations between biomarker levels and dietary metal intake. We evaluated the association between biomarker levels and dietary metal intake using linear regression, with biomarker level as the dependent variable and dietary metal intake level as an independent variable. Both biomarker levels and dietary metal intake data were log-transformed (base of 10) and included in the model as continuous variables. Two models were constructed for the analysis: a) bivariate regression model and b) multivariate regression model (“full” model) adjusting for all covariates (as indicated below). The association was evaluated for each metal and each of the racial/ethnic groups and subgroups. For the regression analysis, we restricted the data to those individuals who had complete data for all the covariates used in the analysis.

Descriptive statistics of food consumption and dietary metal intake. We computed weighted summary statistics (arithmetic mean, and 50th and 95th percentiles) for body weight (BW)-adjusted food consumption (in units of g-food/kg-BW/day) and BW-adjusted dietary metal intake (in units of μg-metal/kg-BW/day) across 14 major food categories and additional subcategories under cereal grains and fish. Dietary metal intake was calculated for total arsenic, inorganic arsenic, cadmium, lead, and mercury.

We also calculated weighted summary statistics for dietary metal intake within subgroups of each race/ethnicity defined by the following sociodemographic and geographic covariates: sex, age, education, household income, birthplace (United States/non–United States), and urban–rural classification, based on the 2013 NCHS urban–rural classification scheme for counties (Ingram and Franco 2014), and U.S. census region. Because of protections intended to preserve study participants’ confidentiality, we were unable to analyze Asian subgroups using geographical covariates (urbanization and census region). For children (6–19 years), the education level of the household reference person (typically the adult owner or renter of the residence) was used.

We performed pairwise comparisons of arithmetic mean dietary intake for each combination of the five NHANES racial/ethnic groups (e.g., mean intake in Asian females vs. non-Hispanic white females). In addition, we compared each combination of the three Asian subgroups (e.g., mean intake in Chinese females vs. Asian Indian females) to assess variability across the Asian subgroups. Further, differences in arithmetic means of dietary metal intake within each covariate (e.g., Asian males vs. Asian females) were determined using analysis of variance (ANOVA). Statistical significance was determined by p-value < 0.05.


Sample Characteristics

After excluding children < 6 years of age, “other” race groups, and participants without body weight or dietary data from both interview days, data from 6,099 NHANES participants were included in our analyses (Figure 1). The characteristics of the study population are presented in Table 2.

Figure 1. Flow diagram of eligibility for analyses.

Figure 1. Schematic diagram of inclusion criteria and sample counts.

Notes: B-Cd, blood cadmium; B-Hg, blood mercury; B-Pb, blood lead; NHANES, National Health and Nutrition Examination Survey; U-DMA, urinary dimethylarsinic acid; U-tAs, urinary total arsenic.

View larger image (TIF File)

Table 2. Select View Table (HTML Version) for a 508-conformant version

Table 2. Characteristics of study participants [n (%) or %] with weighted percentage, NHANES 2011–2012.

View Table (HTML Version)
View larger image (TIF File)

The age distribution of Asians was similar to those of the black and other Hispanic groups. A higher educational status was evident among Asians, compared with other racial/ethnic groups. Household income in Asians closely corresponded to that of whites and was higher than those of other racial/ethnic groups: More than 40% of whites and Asians reported their annual household income was > $75,000. More than 70% of Asians were born outside the United States, whereas the majority of whites and blacks (> 90%) were born in the United States. Geographic variations across the groups were fairly large. Approximately 80% of Asians lived in metro areas, and tended to concentrate mainly in the western region of the United States.

Across the Asian subgroups, the age distribution was generally similar, although the distribution of Asian Indians tended to be slightly shifted to younger ages than those of other two subgroups. Across the Asian subgroups, a higher educational status was observed among Chinese and Asian Indians than the Other Asian subgroup. The percentage of U.S.-born individuals was considerably lower in Asian Indians than in Chinese and Other Asian subgroups. The relative distribution of the Asian subgroups in the study population was similar to that reported in the 2010 Census data (Hoeffel et al. 2012).

Data Preparation of Dietary Intake

The food consumption data from 6,099 individuals comprised 5,273 unique food items. The food items reported in the consumption data were converted into 386 individual food commodities. Among these items, we identified 123 food commodities as target food commodities based on consumed quantities (i.e., commodities making up 95% of the diet in at least one of the racial/ethnic groups). We were able to assign chemical data to approximately 80% of the target food commodities for their total arsenic, cadmium, and lead content. No chemical data were available for the remaining 20%, but these commodities each comprised < 0.5% of the total food consumed. Because the TDS’s mercury analysis in food is focused on fish and other seafood, we were able to assign mercury content only to 28 out of the 123 target food commodities (roughly 20%).

Regression Analyses

Table 3 presents the results of linear regression analyses predicting biomarker levels as a function of estimated dietary intake. The results of the bivariate and multivariate models for total and inorganic arsenic were similar. In general, total dietary arsenic intake (DI-tAs) and dietary inorganic arsenic intake (DI-iAs) were significant predictors of U-tAs and U-DMA, respectively (p < 0.05). Standardized regression coefficients between total and inorganic arsenic were similar across the racial/ethnic groups and Asian subgroups, ranging from 0.24 to 0.41 (excluding nonsignificant results) for total arsenic, and from 0.26 to 0.59 for the inorganic arsenic model. For both the total and inorganic arsenic models, a higher standardized regression coefficient was observed among the Asian group and Asian subgroups compared with those of other racial/ethnic groups.

Table 3. Select View Table (HTML Version) for a 508-conformant version

Table 3. Association between dietary intake of metals and biomarker levels: multiple linear regression results.

View Table (HTML Version)
View larger image (TIF File)

In multivariate models, dietary cadmium intake (DI-Cd) was not a significant predictor of B-Cd levels in either the main racial/ethnic groups or the Asian subgroups. A significant correlation between B-Pb levels and dietary lead intake (DI-Pb) was found only among Mexican Americans.

Dietary mercury intake (DI-Hg), on the other hand, was a significant predictor of B-Hg levels among all racial/ethnic groups and subgroups, except Asian Indians. The Chinese subgroup had the highest standardized regression coefficient value for the regression model between DI-Hg and B-Hg levels.

Comparisons of Dietary Metal Intake

Table 4 presents overall mean dietary metal intake across the five NHANES racial/ethnic groups. Mean dietary metal intake by sociodemographic covariates for the Asian subgroups is shown in Table 5.

Table 4. Select View Table (HTML Version) for a 508-conformant version

Table 4. Comparison of weighted mean dietary metal intake (μg-metal/kg-BW/day) by NHANES racial and ethnic group.

View Table (HTML Version)
View larger image (TIF File)

Table 5. Select View Table (HTML Version) for a 508-conformant version

Table 5. Comparison of weighted mean dietary metal intake (μg-metal/kg-BW/day) by Asian subgroup.

View Table (HTML Version)
View larger image (TIF File)

Arsenic, total. The Asian group had the highest overall mean DI-tAs across the five racial/ethnic groups (Table 4). In general, higher DI-tAs among the Asian group was consistently observed, independent of the various sociodemographic and geographical characteristics. DI-tAs among the Asian group was often more than twice that of other racial/ethnic groups. The majority of DI-tAs originated from fish (> 85%), regardless of racial/ethnic group. The Asian group had the highest contribution from fish (92.6%) (Figure 2). Furthermore, Asians had the highest percentage of fish consumers and the highest arithmetic mean fish consumption (see Table S1). On average, fish consumption among Asians was roughly twice that of other racial/ethnic groups.

Figure 2. Stacked bar graph showing percentage contribution of fish versus other dietary components to total dietary arsenic (y-axis) according to race/ethnicity (x-axis).

Figure 2. Food category–specific percent contribution to dietary arsenic (total) intake by race/ethnicity.

Legend: A, Asian; AI, Asian Indian; B, black; C, Chinese; H, other Hispanic; MA, Mexican American; O, Other Asian; W, white.

View larger image (TIF File)

Among the three Asian subgroups, variations in DI-tAs were minimal (Table 5). There were no apparent associations of DI-tAs with sociodemographic covariates. The observation that DI-tAs originated predominantly from fish did not vary across Asian subgroups. However, substantial variations in fish consumption patterns did exist among these subgroups (see Table S2). The Asian-Indian subgroup had a considerably lower percentage of fish consumers and a lower average fish consumption, compared with the Chinese and Other Asian subgroups.

Arsenic, inorganic. Similar to the DI-tAs results, the overall mean DI-iAs was highest among the Asian group (Table 4). Also, Asians had significantly higher DI-iAs as compared with all other racial/ethnic groups in nearly all of the comparisons performed within the sociodemographic and geographic covariate categories. The contribution from cereal grains was the highest across different food categories, ranging from 67% (white) to 82.1% (Asians) (Figure 3). Rice made up most of the DI-iAs from cereal grains among Asians, whereas the contribution of rice to overall DI-iAs from cereal grains was lower among other racial/ethnic groups. Asians consumed more rice (white and brown) than any other racial/ethnic group, in terms of rice consumption percentage and mean rice consumption (see Table S1).

Figure 3. Stacked bar graph showing percentage contribution of rice, other cereal grains, and other dietary components, respectively, to total dietary inorganic arsenic (y-axis) according to race/ethnicity (x-axis).

Figure 3. Food category–specific percent contribution to dietary arsenic (inorganic) intake by race/ethnicity.

Legend: A, Asian; AI, Asian Indian; B, black; C, Chinese; H, other Hispanic; MA, Mexican American; O, Other Asian; W, white.

View larger image (TIF File)

DI-iAs was similar among Asian subgroups. There was no apparent pattern of differences in DI-iAs based on the sociodemographic covariates, except by age groups, where there were some significant differences. Children (6–11 years) had the highest DI-iAs but no noticeable differences were observed across the next four older age groups. Unlike the fish consumption results, no apparent differences in rice consumption was observed across the three Asian subgroups (see Table S2).

Cadmium. Asians had the highest overall mean DI-Cd (Table 4). Vegetables were the largest source of DI-Cd, accounting for > 56% of the total DI-Cd, followed by cereal grains (~ 15–20%), fruits (~ 6–8%), and dairy (~ 5–6%) (see Figure S1). Although the general makeup of DI-Cd sources was similar across the racial/ethnic groups, the contribution of rice to overall DI-Cd from cereal grains among Asians was two to six times higher than those of other racial/ethnic groups.

There was little variation in DI-Cd across the Asian subgroups (Table 5). A similar pattern of age-related differences that was observed for DI-iAs was also seen in DI-Cd. There was a trend toward increasing DI-Cd levels with greater educational status in all three Asian subgroups. Asian Indians and Other Asians born outside the United States had significantly lower DI-Cd than those born in the United States. The source of DI-Cd, was similar across the Asian subgroups (see Figure S1).

Lead. Overall DI-Pb was highest among Mexican-Americans, but not significantly higher than among Asians (Table 4). The degree and statistical significance of difference in DI-Pb between Asians and other racial/ethnic groups were the least of the five metals. DI-Pb was more widely distributed among different food categories than other metals. The four largest DI-Pb contributors were vegetables, fruits, cereal grains, and dairy, with each of these contributing 14–24% of total DI-Pb, depending on the racial/ethnic group (see Figure S2).

As was the case with DI-Cd, there was little variation in DI-Pb among Asian subgroups (Table 5). The patterns of the association of DI-Pb with age, education, and birthplace were similar to those observed for DI-Cd. Further, the sources of DI-Pb were similar across Asian subgroups, except for a higher DI-Pb contribution from dairy sources among Asian-Indians (see Figure S2).

Mercury. Although chemical data on mercury in the TDS were limited due to their specific data collection focus on fish and other seafood, we also estimated DI-Hg. Asians had significantly higher overall mean DI-Hg than the other main racial/ethnic groups (Table 4); however, the differences were not as pronounced as those seen in DI-tAs with regard to their degree and statistical significance. Dairy was the highest source of DI-Hg for whites, Mexican Americans, and other Hispanics, accounting for approximately 34–41% of DI-Hg. Among blacks and Asians, however, the largest DI-Hg contribution was from fish (see Figure S3).

As with other metals, variations in DI-Hg across the three Asian subgroups were minimal (Table 5). There was no apparent association of DI-Hg with sociodemographic covariates, except for age-related differences similar to those observed for DI-iAs, DI-Cd, and DI-Pb. Moreover, a higher DI-Hg contribution from fish was observed among the Other Asian subgroup, whereas there was generally little difference in the sources of DI-Hg between Chinese and Indian Asian subgroups (see Figure S3).


The daily food consumption and dietary metal intake estimated in our study were in general agreement with the results presented in previous studies. We estimated a mean daily consumption of seafood ranging from 0.36 (white) to 0.84 g/kg/day (Asians). The range of the mean per capita consumption of seafood (finfish and shellfish combined) based on the analysis of the 2003–2006 NHANES data (all ages) by the U.S. EPA was 0.23 (Mexican American) to 0.45 g/kg/day for the “other” ethnic group including Asians (U.S. EPA 2011). Additionally, the study of adults’ seafood consumption (≥ 18 years old) from 10 Asian American and Pacific Islander ethnic groups in King County, Washington, estimated the mean consumption of all seafood to be 1.89 g/kg/day (Sechena et al. 1999). Furthermore, the estimated daily consumption of rice in our study was 0.27 (white) to 1.23 g/kg/day (Asians). The U.S. EPA estimated the per capita consumption of rice to be between 0.2 (white) and 0.8 g/kg/day for the “other” ethnic group based on the 2003–2006 NHANES data (U.S. EPA 2011).

Further, the dietary metal intake levels estimated in the present study generally agreed with the results presented previously. Xue et al. (2010) computed Di-tAs and Di-iAs based on NHANES 2003–2004 using a probabilistic exposure model. The ranges of the estimated mean DI-tAs and DI-iAs in various age groups ≥ 6 years of age were 0.25–0.37 and 0.03–0.05 μg/kg/day, respectively. Our estimations of the mean DI-tAs and DI-iAs were slightly higher than those estimated by Xue et al. (2010), with respective mean intake ranges of 0.61–2.0 and 0.05–0.11 μg/kg/day. An average DI-Cd based on the national representative food consumption data (ages ≥ 1 year) in the United States between 1989 and 1991 was 0.2 μg/kg/day (Dougherty et al. 2000). We estimated DI-Cd to be 0.08–0.11 μg/kg/day in our study. A probabilistic analysis of DI-Hg based on NHANES 1999–2006 estimated DI-Hg across different age groups ≥ 6 years of age to be 0.01–0.05 μg/kg/day for the “other” race group including Asians and to be 0.01–0.02 μg/kg/day for the rest of racial/ethnic groups combined (white, black, and Mexican American) (Xue et al. 2012). Our estimated mean DI-Hg was 0.09 μg/kg/day for Asians and 0.05–0.07 μg/kg/day for the rest of the four racial/ethnic groups. The estimated DI-Pb previously reported was similar to the levels estimated in our study. The estimated mean DI-Pb among population-based samples from the EPA Region V (Midwest states including Indiana, Illinois, Michigan, Minnesota, Ohio, and Wisconsin) by Thomas et al. (1999) was 0.25 μg/kg/day. We estimated the average DI-Pb to be approximately 0.1 μg/kg/day.

Using nationally representative data, our study confirmed that DI-tAs, DI-iAs, and DI-Hg are key pathways of arsenic and mercury exposures and are significantly associated with their corresponding biomarker levels among the Asian populations in the United States. Despite the high fish consumption rate in the Chinese subgroup, the regression model for total arsenic in this subgroup had the lowest standardized regression coefficient and highest p-value, suggesting that there may be other arsenic exposure sources or different levels of confounding (e.g., smoking) in this subgroup. The standardized regression coefficients for arsenic (total and inorganic) among Asians were greater than those of other racial/ethnic groups, suggesting that other factors, which differ across racial/ethnic groups, may influence these associations (e.g., more efficient absorption, poorer elimination). In comparison to other metals, the difference in the estimated DI-tAs and DI-iAs between Asians and the other racial/ethnic groups was greater (often two times higher) and statistically significant. The significant difference was most pronounced for DI-iAs. We also confirmed that fish (for total arsenic and mercury) and rice (for inorganic arsenic) are the predominant contributors to the dietary intake of metals among Asians. Previously, this had only been inferred from the data of the aggregated race group (i.e., “other” racial group in the NHANES which was comprised of small minority populations such as Native Americans, Pacific Islanders, and multiple racial individuals) (Wei et al. 2014; Xue et al. 2012). Although arsenic consumed through fish is considered to be mostly the less harmful organic forms of arsenic, the higher DI-tAs in Asians are worth noting.

Unlike arsenic exposure, there was no compelling evidence that estimated dietary intake is an important exposure pathway for cadmium and lead exposure among Asians. Although the estimated DI-Cd was generally the highest within the Asian population, no significant association between B-Cd level and DI-Cd was observed. There appeared to be less evidence supporting the hypothesis that DI-Pb contributed to B-Pb among Asians. Although not always significant, Mexican Americans had higher mean DI-Pb than Asians. DI-Pb was not a significant predictor of B-Pb levels among Asians, but it was among Mexican Americans. These results suggest that contributions from nondietary sources may be important for cadmium and lead exposures among Asian populations, which is consistent with our understanding of the exposure characteristics of cadmium and lead in the general U.S. population (ATSDR 2007b, 2012). Smoking is the main exposure route for cadmium, followed by food consumption. Likewise, exposure to lead can originate from various environmental and occupational sources. Adjusting data for these exposure sources may have improved our evaluation of dietary contributions.

Aside from these findings, there is another important consideration when interpreting the results: The metals evaluated in our study have different half-lives and toxicokinetics characteristics in the human body. For instance, cadmium in blood exhibits the first component of elimination with a half-life of 3–4 months, followed by a slow component with a half-life of 10 years (Järup et al. 1983); therefore, B-Cd may be a reflective of body burden from long-term exposure. On the other hand, arsenic has a shorter half-life (~ 2–3 days), and its biomarker levels may be a better representation of short-term exposure (ATSDR 2015). This may be another reason we observed a positive relationship between the U-tAs and U-DMA levels and their estimated dietary levels, because the dietary data we used were obtained from 24-hr recall rather than long-term food consumption surveys.

The limitations associated with the present study stemmed mainly from two sources: the estimation of metal concentrations in food and the application of the NHANES food consumption data. The estimation of metal concentrations has some limitations. Concentrations in food commodities were estimated based on a single representative concentration and did not account for variations in chemical concentrations across different food types, geographical locations of cultivation (Meharg et al. 2009; Williams et al. 2005, 2007), and growing methods (Barański et al. 2014), among other factors within a single food commodity. For instance, the commodity “saltwater fish” includes a wide variety of fish species that can have different mercury content (FDA 2014b). Additionally, we were not able to assign chemical data to all of the target food commodities due to a lack of data. Seaweed is a good example; it may have an elevated metal content (Almela et al. 2006; Rose et al. 2007) and can be an important source of dietary metal intake (Lee et al. 2006). Therefore, omission of such a food commodity in the estimation of metal concentrations will underestimate overall dietary metal intake. Further, the composition of food commodities was assumed to be the same, although there may be variations in food recipe and preparations. Moreover, we used an LOD-based “fill value” for the nondetected results in the estimation of metal concentrations in food. The use of the fill value may have diluted the importance of food with high metal contents in the estimation of overall dietary intake and likely weakened the association between biomarker levels and dietary metal intake, especially for mercury. Also, there may be some uncertainties associated with the use of inorganic arsenic data from the study by Schoof et al. which were collected in an older time period (i.e., 1997).

Other limitations of the study are attributable to the NHANES consumption data. We estimated the daily amount of food consumed, based on 24-hr recall dietary data. Therefore, the data may only represent a snapshot of study participants’ dietary consumption and may not reflect their long-term food consumption patterns. In addition, these data are subject to recall bias. Further, the relatively small sample sizes of Asian subgroups from one data cycle may have produced statistically unreliable results that should be viewed with caution. The results of our study should be verified based on a larger data set, appending the data from the continuous oversampling of Asian populations in the 2012–2014 NHANES in the future.

The major strengths of this study are attributed to use of national representative data of Asian populations from NHANES. We believe that this is one of the first works to investigate the dietary consumption and dietary metal intake of Asians on a national scale. Currently, studies of food consumption and dietary metal intake in Asian population in the United States are limited mainly to those ethnic groups in Far East Asia (such as Chinese, Japanese, and Koreans) and to cohorts from geographic areas with high Asian populations (e.g., New York City, Hawaii, and California). As fractions of Asian ethnic groups (e.g., Asian Indians) originating from regions other than the Far East are becoming larger, and residences of Asians in the United States have become more geographically diverse in the past decade (U.S. Census Bureau 2013), our study provides more comprehensive characteristics of Asian populations in the United States than previous studies. Other advantages of use of the NHANES data are its relatively large sample size and ability to account for study participants’ various sociodemographic and geographic characteristics in our data analysis. Furthermore, our study evaluated comprehensive dietary metal intake estimated based on a large number of target food commodities, rather than food consumption or consumption frequency of limited food items that were often used as bases in the previous studies.


To our knowledge, this is one of the first studies evaluating dietary intake as a potential cause of the elevated biomarker levels of four metals (arsenic, cadmium, lead, and mercury) previously seen among Asian populations in the United States. We confirmed that dietary intake is an important exposure pathway for total and inorganic arsenic and mercury. Our study also confirmed that fish (for total arsenic and mercury) and rice (for inorganic arsenic) are important dietary sources of their arsenic and mercury exposures. The results of cadmium and lead were not as conclusive as those of arsenic and mercury, indicating contributions from nondietary sources may be important for cadmium and lead.


Almela C, Clemente MJ, Vélez D, Montoro R. 2006. Total arsenic, inorganic arsenic, lead and cadmium contents in edible seaweed sold in Spain. Food Chem Toxicol 44(11):1901–1908.

ATSDR (Agency for Toxic Substances and Disease Registry). 1999. Toxicological Profile for Mercury.​.asp?id=115&tid=24 [accessed 23 October 2014].

ATSDR. 2007a. Toxicological Profile for Arsenic.​.asp?id=22&tid=3 [accessed 23 October 2014].

ATSDR. 2007b. Toxicological Profile for Lead.​.asp?id=96&tid=22 [accessed 23 October 2014].

ATSDR. 2012. Toxicological Profile for Cadmium.​.asp?id=48&tid=15 [accessed 12 March 2015].

ATSDR. 2015. Case Studies in Environmental Medicine: Taking an Exposure History.​y/docs/exposure_history.pdf [accessed 20 November 2015].

Barański M, Srednicka-Tober D, Volakakis N, Seal C, Sanderson R, Stewart GB, et al. 2014. Higher antioxidant and lower cadmium concentrations and lower incidence of pesticide residues in organically grown crops: a systematic literature review and meta-analyses. Br J Nutr 112(5):794–811.

CDC (Centers for Disease Control and Prevention). 2014. Fourth National Report on Human Exposure to Environmental Chemicals. Updated Tables, July, 2014. Atlanta, GA:CDC.

Davis MA, Mackenzie TA, Cottingham KL, Gilbert-Diamond D, Punshon T, Karagas MR. 2012. Rice Consumption and Urinary Arsenic Concentrations in U.S. Children. Environ Health Perspect 120:1418–1424, doi: 10.1289/ehp.1205014.

Dougherty CP, Henricks Holtz S, Reinert JC, Panyacosit L, Axelrad DA, Woodruff TJ. 2000. Dietary exposures to food contaminants across the United States. Environ Res 84(2):170–185.

FDA (U.S. Food and Drug Administration). 2013. Total Diet Study.​rch/TotalDietStudy/default.htm [accessed 10 October 2013].

FDA. 2014a. Arsenic in Rice and Rice Products.​Contaminants/Metals/ucm319870.htm [accessed 7 February 2015].

FDA. 2014b. Mercury Levels in Commercial Fish and Shellfish (1990–2010).​Contaminants/Metals/ucm115644.htm [accessed 3 March 2015].

FDA. 2015a. Elemental Analysis Manual (EAM) for Food and Related Products.​rch/LaboratoryMethods/ucm2006954.htm [accessed 31 August 2015].

FDA. 2015b. Study Design.​rch/TotalDietStudy/ucm184232.htm [accessed 21 October 2015].

Hoeffel EM, Rastogi S, Kim MO, Shahid H. 2012. The Asian Population: 2010. 2010 Census Briefs. C2010BR-11.​fs/c2010br-11.pdf [accessed 21 October 2014].

IARC (International Agency for Research on Cancer). 2013. IARC Monographs on the Evaluation of Carcinogenic Risks to Humans: Agents Classilfied by the IARC Monographs, Volumes 1–117.​tion/ [accessed 20 December 2013].

Ingram DD, Franco SJ. 2014. NCHS Urban–Rural Classification Scheme for Counties. Vital Health Stat 2 166:1–73.

Järup L, Rogenfelt A, Elinder CG, Nogawa K, Kjellström T. 1983. Biological half-time of cadmium in the blood of workers after cessation of exposure. Scand J Work Environ Health 9(4):327–331.

Johnson CL, Dohrmann SM, Burt VL, Mohadjer LK. 2014. National Health and Nutrition Examination Survey: sample design, 2011–2014. Vital Health Stat 2 162:1–33.

Lee HS, Cho YH, Park SO, Kye SH, Kim BH, Hahm TS, et al. 2006. Dietary exposure of the Korean population to arsenic, cadmium, lead and mercury. J Food Compost Anal 19:S31–S37.

Meharg AA, Williams PN, Adomako E, Lawgali YY, Deacon C, Villada A, et al. 2009. Geographical variation in total and inorganic arsenic content of polished (white) rice. Environ Sci Technol 43(5):1612–1617.

NCHS (National Center for Health Statistics). 2011a. Laboratory Procedure Manual. Analyte: Arsenobetaine, Arsenocholine, Trimethylarsine Oxide, Monomethylarsonic Acid, Dimethylarsinic Acid, Arsenous (III) Acid, Arsenic (V) Acid. Hyattsville, MD:NCHS.

NCHS. 2011b. Laboratory Procedure Manual. Analyte: Creatinine. Hyattsville, MD:NCHS.

NCHS. 2012. Laboratory Procedure Manual. Analytes. Cadmium, Lead, Manganese, Mercury, and Selenium. Hyattsville, MD:NCHS.

NCHS (National Center for Health Statistics). 2013. National Health and Nutrition Examination Survey: Analytic Guidelines, 2011–2012. Hyattsville, MD:NCHS.​ytic_guidelines_11_12.pdf [accessed 15 October 2015].

NCHS-RDC (National Center for Health Statistics Research Data Center). 2012. Disclosure Manual. Preventing Disclosure: Rules for Researchers. Hyattsville, MD:NCHS.​emanual.pdf [accessed 16 July 2015].

Rose M, Lewis J, Langford N, Baxter M, Origgi S, Barber M, et al. 2007. Arsenic in seaweed—Forms, concentration and dietary exposure. Food Chem Toxicol 45(7):1263–1267.

Schoof RA, Yost LJ, Eickhoff J, Crecelius EA, Cragin DW, Meacher DM, et al. 1999. A market basket survey of inorganic arsenic in food. Food Chem Toxicol 37(8):839–846.

Sechena R, Nakano C, Liao S, Polissar N, Lorenzana R, Truong S, et al. 1999. Asian and Pacific Islander Seafood Consumption Study. EPA 910/R-99-003. Washington, DC:U.S. EPA.

Thomas KW, Pellizzari ED, Berry MR. 1999. Population-based dietary intakes and tap water concentrations for selected elements in the EPA region V National Human Exposure Assessment Survey (NHEXAS). J Expo Anal Environ Epidemiol 9(5):402–413.

U.S. Census Bureau. 2013. Asians Fastest-Growing Race Or Ethnic Group in 2012, Census Bureau Reports. Releases no. CB13-112.​eases/2013/cb13-112.html [accessed 13 June 2013].

USDA (U.S. Department of Agriculture). 2014. Agricultural Research Service. WWEIA/NHANES Overview. 2014;​cid=13793# [accessed 2 October 2014].

USDA (U.S. Department of Agriculture). 2015. Agricultural Research Service. USDA Food and Nutrient Database for Dietary Studies 2011–2012. Food Surveys Research Groups.​beltsville-md/beltsville-human-nutrition​-research-center/food-surveys-research-g​roup/docs/fndds-download-databases/ [accessed 3 September 2015].

U.S. EPA. 2011. Exposure Factors Handbook: 2011 Edition (Final). EPA/600/R-09/052F. Washington, DC:U.S. EPA.

U.S. EPA. 2014. Food Commodity Intake Database: What We Eat in America. [accessed 6 October 2014].

U.S. EPA. 2015. Integrated Risk Information System (IRIS). [accessed 25 August 2015].

Wei Y, Zhu J, Nguyen A. 2014. Rice consumption and urinary concentrations of arsenic in US adults. Int J Environ Health Res 24(5):459–470.

Williams PN, Price AH, Raab A, Hossain SA, Feldmann J, Meharg AA. 2005. Variation in arsenic speciation and concentration in paddy rice related to dietary exposure. Environ Sci Technol 39(15):5531–5540.

Williams PN, Raab A, Feldmann J, Meharg AA. 2007. Market basket survey shows elevated levels of As in South Central U.S. processed rice compared to California: consequences for human dietary exposure. Environ Sci Technol 41(7):2178–2183.

Xue J, Zartarian VG, Liu SV, Geller AM. 2012. Methyl mercury exposure from fish consumption in vulnerable racial/ethnic populations: probabilistic SHEDS-Dietary model analyses using 1999–2006 NHANES and 1990–2002 TDS data. Sci Total Environ 414:373–379.

Xue J, Zartarian V, Wang SW, Liu SV, Georgopoulos P. 2010. Probabilistic modeling of dietary arsenic exposure and dose and evaluation with 2003–2004 NHANES data. Environ Health Perspect 118:345–350, doi: 10.1289/ehp.0901205.

Yost LJ, Tao SH, Egan SK, Barraj LM, Smith KM, Tsuji JS, et al. 2004. Estimation of dietary intake of inorganic arsenic in U.S. children. Hum Ecol Risk Assess 10(3):473–483.

Passive Sampling for Indoor and Outdoor Exposures to Chlorpyrifos, Azinphos-Methyl, and Oxygen Analogs in a Rural Agricultural Community

Author Affiliations open
1Department of Occupational and Environmental Health, University of Iowa College of Public Health, Iowa City, Iowa, USA; 2Department of Environmental and Occupational Health Sciences, University of Washington School of Public Health, Seattle, Washington, USA

PDF icon PDF Version (437 KB)

  • Background: Recent studies have highlighted the increased potency of oxygen analogs of organophosphorus pesticides. These pesticides and oxygen analogs have previously been identified in the atmosphere following spray applications in the states of California and Washington.

    Objectives: We used two passive sampling methods to measure levels of the ollowing organophosphorus pesticides: chlorpyrifos, azinphos-methyl, and their oxygen analogs at 14 farmworker and 9 non-farmworker households in an agricultural region of central Washington State in 2011.

    Methods: The passive methods included polyurethane foam passive air samplers deployed outdoors and indoors and polypropylene deposition plates deployed indoors. We collected cumulative monthly samples during the pesticide application seasons and during the winter season as a control.

    Results: Monthly outdoor air concentrations ranged from 9.2 to 199 ng/m3 for chlorpyrifos, 0.03 to 20 ng/m3 for chlorpyrifos-oxon, < LOD (limit of detection) to 7.3 ng/m3 for azinphos-methyl, and < LOD to 0.8 ng/m3 for azinphos-methyl-oxon. Samples from proximal households (≤ 250 m) had significantly higher outdoor air concentrations of chlorpyrifos, chlorpyrifos-oxon, and azinphos-methyl than did samples from nonproximal households (p ≤ 0.02). Overall, indoor air concentrations were lower than outdoors. For example, all outdoor air samples for chlorpyrifos and 97% of samples for azinphos-methyl were > LOD. Indoors, only 78% of air samples for chlorpyrifos and 35% of samples for azinphos-methyl were > LOD. Samples from farmworker households had higher indoor air concentrations of both pesticides than did samples from non-farmworker households. Mean indoor and outdoor air concentration ratios for chlorpyrifos and azinphos-methyl were 0.17 and 0.44, respectively.

    Conclusions: We identified higher levels in air and on surfaces at both proximal and farmworker households. Our findings further confirm the presence of pesticides and their oxygen analogs in air and highlight their potential for infiltration of indoor living environments.

  • Citation: Gibbs JL, Yost MG, Negrete M, Fenske RA. 2017. Passive sampling for indoor and outdoor exposures to chlorpyrifos, azinphos-methyl, and oxygen analogs in a rural agricultural community. Environ Health Perspect 125:333–341;

    Address correspondence to J.L. Gibbs, Department of Occupational and Environmental Health, University of Iowa College of Public Health, 145 North Riverside Dr., Iowa City, IA 52242 USA. Telephone: (319) 335-4405. E-mail:

    We would like to thank the staff at the University of Washington Environmental Health Laboratory and Trace Organics Analysis Center for their consultation on method development. In addition, we thank B. Thompson, I. Islas, and E. Carosso from the Fred Hutchinson Cancer Research Center (FHCRC) in Seattle, Washington; the Yakama Nation Environmental Protection Program (YNEP) in Sunnyside, Washington; and G. Hoogenboom at Washington State University for assisting in identification of community sampling locations.

    The field work was financially supported by the Center for Child Environmental Health Risks Research, the National Institute of Environmental Health Sciences/National Institutes of Health (grant P01 ES009601), and the U.S. Environmental Protection Agency (grant RD-83451401). Passive sampling method development was funded by the Pacific Northwest Agricultural Safety and Health (PNASH) Center (National Institute for Occupational Safety and Health; grant 2 U50 OH07544).

    The authors declare they have no actual or potential competing financial interests.

    Received: 22 January 2016
    Revised: 18 July 2016
    Accepted: 19 July 2016
    Published: 12 August 2016

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (136 KB)

    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact Our staff will work with you to assess and meet your accessibility needs within 3 working days.

    PDF icon Supplemental Table of Contents PDF (101 KB)


Organophosphorus Pesticides and Oxygen Analogs

In the Yakima Valley region of Washington State, there are more than a thousand orchards (e.g., apples, pears, cherries) covering over 100,000 acres. Washington is the lead producer of apples and cherries in the United States, and 10–12 billion apples are picked each year (USDA 2009) (Figure 1 shows a map of the region). The region is also home to many farmworker families and more than half of the population is Hispanic/Latino (U.S. Census Bureau Geography Division 2010). Most of this population is involved in tree fruit production—harvesting, pruning, thinning, and applying agricultural chemicals (Thompson et al. 2008). In 2011, chlorpyrifos (CPF) and azinphos-methyl (AZM) were some of the most commonly applied organophosphorus (OP) pesticides in tree fruit and vegetable production (Baker and Stone 2015). Both pesticides are often applied in aerosolized form to tree fruits using a large sprayer attached to a tractor. In 2012, the U.S. Environmental Protection Agency (EPA) banned the use of AZM in apple production. Prior to the ban, when this study was conducted, AZM and CPF were commonly sprayed with application rates averaging 0.5 kg/acre and 1 kg/acre active ingredient, respectively (USDA 2008).

Figure 1. Map of Washington State showing the location of the Yakima Valley study area, with an inset showing the location of Washington State within the United States.

Figure 1. Map of Yakima Valley, Washington State study region.

View larger image (TIF File)

The use of OP pesticides in Yakima Valley has long been a health concern of local residents to due to potential human exposures resulting from off target volatilization and drift. In 2008, the state of Washington funded a study to examine off-target movement of OP pesticides and potential risk to bystanders (Fenske et al. 2009). In the 2009 study, CPF, AZM, and their oxygen analogs were identified in the outdoor air of the surrounding agricultural communities, indicating direct atmospheric transformation. Other studies have also reported these compounds in air (Armstrong et al. 2013b; CARB 1998; CDPR 2006, 2009).

Toxicology studies have focused on the relative potency of combined OP pesticides and their oxygen analogs in animal models (Costa et al. 2005), and acknowledge transformation to the oxygen analog in vivo as a metabolic product through breakdown mechanisms involving cytochrome p450 enzymes. Chlorpyrifos-oxon (CPF-O) is poses a special risk for genetically susceptible individuals who have lower levels of the paraoxonase (PON-1–/–) enzyme (Shih et al. 1998), and children may be susceptible to the CPF-O due to differences in their metabolic functioning during development (Barr et al. 2004; Costa et al. 2005). Therefore, it is important to consider the presence of oxygen analogs in the air when measuring human exposure.

In several studies over a decade ago, we found that levels of OP pesticide metabolites in the urine of farmworker children were significantly higher than the levels in the urine of non-farmworker children in the same region (Loewenherz et al. 1997; Lu et al. 2000): These relatively high levels were later confirmed by comparison with national biomonitoring data (Fenske et al. 2005). We also found that pesticide levels in household dust (including AZM and CPF) were higher in farmworker homes than in non-farmworker homes in the same region (Lu et al. 2000; Fenske et al. 2002).

CPF and AZM are both semivolatile compounds, and they exist as both vapor and particle-bound forms in air. This phase-partitioning is highly dependent on a combination of the timing of pesticide application and meteorological factors (Howard 1991). Both compounds can persist for days to weeks outdoors, and for several months indoors (Lewis 2005; Wauchope et al. 1992). There is very little scientific data regarding the long-term atmospheric transport of CPF, AZM, CPF-O, and azinphos-methyl-oxon (AZM-O), and even less is known about their ability to infiltrate indoor environments.

Passive Sampling for Pesticides

To date, many studies have focused on short and long term human health outcomes associated with OP pesticides (Roberts et al. 2012; Quandt et al. 2006), although very few have incorporated long-term air and surface exposure measurements for OP pesticides and oxygen analogs due to the high costs and invasive procedures associated with residential sampling. The oxygen analogs (CPF-O and AZM-O) are relatively new phenomena, and, to our knowledge, no studies have measured them in residences.

Although active air sampling is useful for examining daily fluctuations or collecting a personal sample over the course of a work shift, it involves frequent collection of sampling media, uses electricity, and requires space for the sampling pumps. In a previous study (Armstrong et al. 2013a), we identified artificial transformation from CPF to CPF-O during active air sampling with OVS/XAD-2 tubes (NIOSH 1994) in a controlled laboratory environment. In response, we developed a polyurethane foam (PUF) passive air sampling (PAS) method that was able to sample for OP pesticides and their oxygen analogs at rates similar to active air sampling at 2 L/min (Armstrong et al. 2014b). CPF and AZM are both suitable for passive air sampling because they have ideal chemical properties, including octanol-air partition coefficients (log KOA values) that fall somewhere between polychlorinated biphenyls (PCBs) and polybrominated diphenyl ethers (PBDEs) (Table 1).

Table 1. Select View Table (HTML Version) for a 508-conformant version

Table 1. Chemical properties of CPF, AZM, and suitability for passive sampling with PUF-PAS.

View Table (HTML Version)
View larger image (TIF File)

Another common indoor passive sampling method involves deposition plates to collect settled particulate. Since the deposition method collects larger diameter particles (as opposed to gases), it is a useful measure of particle-bound phase and dust settling. Deposition plates for indoor OP pesticides have been used in previous studies using polyethylene, chromatography paper, and double-layer gauze pads backed by aluminum foil (Keenan et al. 2010; Lu and Fenske 1998).

Our overall aim of this study was to use passive sampling methodologies to measure airborne and surface deposition levels of CPF, AZM, CPF-O, and AZM-O outside and inside of households in a rural agricultural region. Our secondary aim was to compare the levels between proximal and non-proximal and between farmworker and non-farmworker households to determine if certain groups were at higher risk of exposure.

Study Methods

Sampling Plan

We conducted the residential sampling during three seasons in 2011: a) the spring pre-thinning season for CPF and CPF-O, b) the summer thinning season for AZM and AZM-O, and c) winter dormancy season for CPF, CPF-O, AZM, and AZM-O (Figure 2 shows timeline). The pre-thinning application, thinning application, and winter dormancy seasons were defined using CPF and AZM product information from Washington State University’s Decision Aid System (, which uses meteorologic and entomologic data to predict optimal pesticide application times for tree fruit producers. In addition, we contacted local Washington State agricultural extension agents to inform us about field activity.

Figure 2. Diagram showing the timing of agricultural activities (Pre-thinning: March–May; Thinning: June–August; Dormant: November–December) and the timing of pesticide sampling (CPF and CPF-O: March–April; AZM and AZM-O: June–August; CPF, CPF-O, AZM, and AZM-O: November–December).

Figure 2. Sampling time-line. Sampling occurred in 2011 during the spring application season for CPF and CPF-O, during the summer application season for AZM and AZM-O, and during the winter dormancy season for CPF, CPF-O, AZM, and AZM-O.

View larger image (TIF File)

Twenty-three sampling locations were selected a priori to be equally grouped as proximal (≤ 250 m of any nearest tree fruit field) and nonproximal (> 250 m). Of these, 20 participants were recruited from households enrolled with the Para Niños Saludables project. This is a community-based research project led by researchers at the Fred Hutchinson Cancer Research Center involving a cohort of 60 farmworker and 40 non-farmworker families. We defined farmworker households as having one or more current farmworkers (temporary or full time), and non-farmworker households as having no farmworkers living in the household (employment status was obtained from the 2011 Para Niños Saludables household survey). Details about the project and population have been previously reported by Thompson et al. (2008). Overall, there were 6 proximal farmworker, 2 proximal non-farmworker, 7 nonproximal farmworker, and 5 nonproximal and non-farmworker households (see Table S1).

The three remaining sampling locations were outdoor community air monitoring sites (managed by the Yakama Nation Environmental Protection Program) within 100 m of the nearest Para Niños Saludables residence. During the study, these community locations were required to support replicate sampling and side-by-side comparisons with the active air sampling methods for quality assurance purposes. In addition, for data analysis, at these locations the outdoor measurements were used as surrogates for the nearest household. We used proximity and farmworker employment data from the nearest household and checked to make sure surrogate community locations and participant residences were equidistant from tree fruit fields. One of the community sites was located in a rural area near a proximal farmworker household. The other two community sites were urban, near nonproximal non-farmworker households.

We plotted all locations in ArcGIS (version 10.0; ESRI, Redlands, CA) using GPS coordinates collected with a GPS Map 60CS handheld unit (Garmin, Inc., Olathe, KS). We identified tree fruit fields using a Cropland Data Layer from USDA CropScape. The Cropland Data Layer is a geo-referenced, crop-specific land cover data layer created annually using satellite imagery and extensive agricultural ground cover (USDA-NASS 2012). We checked to make sure surrogate community locations and participant residences were equidistant from tree fruit fields. Since such a large portion of the rural community was involved in agriculture, it was challenging to identify non-farmworker households that were also proximal (only two households were recruited).

We collected a total of 66 outdoor air samples (CPF and CPF-O, n = 36; AZM and AZM-O, n = 30), and 53 indoor air and surface deposition samples (CPF and CPF-O, n = 27; AZM and AZM-O, n = 26) during the application seasons. These numbers include duplicate and triplicate samples deployed in the same location at the same time for quality control purposes (see Table S1 for description of replicate samples). We deployed 7 outdoor air samples and 7 indoor air and surface deposition samples at six locations during the winter dormant season as a control. These winter locations were chosen for optimal geospatial distribution across the region.

This study followed protocols approved by the Fred Hutchinson Cancer Research Center Institutional Review Board. Written informed consent (in Spanish or English) was obtained for all households in the study. A field industrial hygienist scheduled a meeting with the promotora and the household members to set up the samplers. Outdoors, we placed the PUF-PAS away from children’s play areas, buffers (≥ 8 m from trees and buildings), livestock, and other high foot traffic areas. Indoors, we placed the PUF-PAS and deposition plates in the living room or kitchen to capture an area of the house where family members spend a large amount of time. This location was placed ≥ 1 m high on a shelf or desk to minimize interference or contact with other surfaces (e.g., walls, windows, doors). Monthly sampling periods ranged from 24 to 32 days. At each household, outdoor and indoor samples were deployed and collected on the same day. During the time of collection, we obtained qualitative participant feedback about the passive samplers.

Sampling Materials

The PUF-PAS device uses properties of atmospheric diffusion to collect contaminants without the use of a pump and sampling rate is controlled by diffusivity (Hourani and Underhill 1988; Shoeib and Harner 2002). The PUF-PAS method for measurement of OP pesticides and oxygen analogs was previously tested in both laboratory and field environments by Armstrong et al. (2014b) using depuration compounds and side-by-side comparisons with more traditional active sampling methods (U.S. EPA 1999). We derived average air concentration (Cair, ng/m3) from the sampling rate (RPUF-PAS, m3/day) and the mass of pesticide on the matrix (Mpas, ng), where t = time in days (Equation 1):

Cair = Mpas/(RPUF-PAS × t) [1]

Prior to deployment, we spiked each outdoor PUF-PAS with depuration compounds [210 ng of CPF-methyl-D6 (99%, 100 μg/mL in acetonitrile; EQ Laboratories, Atlanta, GA) and 450 ng of AZ-ethyl-D10 (98.5%, 1,000 μg/mL in toluene; EQ Laboratories)] with a 50 μL Hamilton positive displacement syringe. Depuration compounds were not used indoors to ensure safety of residents. We calculated outdoor sampling rates, or RPUF-PAS, using the loss of depuration compounds from the PUF matrix and by calibration with side-by-side active air sampling (AAS). All procedures and calculation of sampling rates have been described by Armstrong et al. (2014b).

Outdoors, we placed the PUF-PAS disk (Tisch Environmental, 14 cm in diameter, 1.3 cm thick, surface area 370 cm2) in a stainless steel, domed chamber (22 cm diameter) to protect from wind, precipitation, and sunlight (Shoeib and Harner 2002; Schuster et al. 2012; Tuduri et al. 2006). Air was allowed to flow over the PUF disks through a 1.5 cm gap between chamber encasements. The sampling housing was hooked to a steel sampling mast at 1.5 m height. After collection, the PUF sample media was sealed in a glass Petri dish, and stored in a –20°C freezer.

Indoors, the shape and surface area of the PUF-PAS was cylindrical (7 × 3 cm diameter, 74 cm2 surface area), and similar to the mini-PUF introduced by Bohlin et al. (2010). We hung the cylinder from a 22 cm tall free-standing hook. Next to the indoor PUF-PAS, a small surface deposition plate consisting of a Petri dish (6 cm diameter, 89000-300 VWR) lined with a polypropylene (PP) filter (5 μm pore, 17.3 cm2 surface area, Whatman) collected deposited particulate. A temperature logger (LogTag TRIX-8) was placed near both passive sampling devices. After collection we wrapped indoor PUF-PAS cylinders in aluminum foil and stored them in zipper-sealed bags, covered and sealed deposition plates, and stored both sample types similarly to outdoor samples. Indoor air concentrations (Cair, ng/m3) were derived using the same calculation (Equation 1) as for outdoors. Since depuration compounds were not used indoors, indoor sampling rates (RPUF-PAS, m3/day) were estimated using the KA (air-side mass transfer coefficient) and the surface area (Sarea) of the indoor PUF cylinder (74 cm2) (Equation 2). We determined KA from the average loss of depuration compounds in previous laboratory tests at 25°C (Armstrong et al. 2014b). KA was adjusted for average indoor temperatures recorded by the indoor temperature logger. The calculation, below, has also been described by Shoeib and Harner (2002) (Equation 2):

Indoor RPUF-PAS = KA × Sarea [2]

For the surface deposition samples, we divided the mass of pesticide (Mpp) by the surface area (Sarea) to obtain a mass loading (Sload, ng/cm2) (Butte and Heinzow 2002) (Equation 3):

Sload = Mpp/(Sarea) [3]

Chemical Analysis

Preparation and storage of PUF-PAS matrices followed similar procedures used in other studies (Bohlin et al. 2010; Shoeib and Harner 2002). We rinsed Petri dishes and aluminum foil with solvent during extraction. PUF-PAS and PP filter matrices were sonicated for 1.5 hr at room temperature (20–23°C) in 10–50 mL acetonitrile solution containing stable-isotope labeled internal standards and then evaporated to 1.5 mL. Large particulate was filtered with a PTFE syringe filter (13 mm, 0.2 μm porosity). Sample analysis was conducted using the liquid chromatography tandem mass spectrometry (LC-MS/MS) method with internal standards (Armstrong et al. 2014a, 2014b). Instrument limits of detection (LOD) were 1 ng/sample for CPF and CPF-O, and 1 ng/sample AZM, and 5 ng/sample for AZM-O. The instrument LOD for all depuration compounds was 1 ng/sample. After accounting for the volume of PUF-PAS and surface area of deposition plates, this corresponded to PUF-PAS method Limit of Quantification (LOQ) ranging from 0.01 to 0.02 ng/m3 for CPF/CPF-O and 0.02 to 0.03 ng/m3 for AZM/AZM-O; and a surface deposition plate method LOQ of 0.03 ng/cm2 for CPF/CPF-O and 0.17 ng/cm2 for AZM/AZM-O.

Coefficients of variation (CV) were ≤ 19% for CPF, ≤ 9% for CPF-O, ≤ 37% for AZM, and ≤ 10% for AZM-O in outdoor air samples; and ≤ 6% for CPF indoor air samples. For the surface deposition plates, CVs were ≤ 15% for CPF and all replicate samples for AZM were below the LOD. We were unable to calculate CVs for AZM, CPF-O, and AZM-O from indoor air samples and surface deposition plates because replicate samples were < LOD. All field blanks were below the LOD for CPF, CPF-O, AZM, and AZM-O. Storage stability and spike fortification recovery results were 80–120% for all measured compounds.

Data Analysis

For air samples below the LOD, we assigned a substitute value for Mpas and Mpp by taking LOD divided by the square root of 2 and divided by the effective air sampling volume (Equation 1 and 2), or surface area (Equation 3), respectively. We calculated the mean, standard deviation, and range for outdoor and indoor air concentrations (ng/m3) and indoor surface deposition (ng/m2) among household types (i.e., proximal farmworker, proximal non-farmworker, non-proximal farmworker, and non-proximal non-farmworker). We compared group results using a 2-way non-parametric Friedman test (α = 0.10), which is similar to a parametric repeated measure ANOVA (Zimmerman and Zumbo 1993). Next, we compared outdoor and indoor air concentrations and indoor surface deposition in proximal vs. non-proximal and farmworker vs. non-farmworker variables using a non-parametric Kruskal–Wallis one way ANOVA test (α = 0.05). Replicate samples were included in these calculations.

Since both outdoor and indoor air samples were collected simultaneously, we calculated indoor/outdoor mean ratios for each household by dividing the indoor air concentration by the outdoor air concentration (the mean of replicate samples was used when necessary). We then calculated the mean of these ratios by household type. A ratio greater than 1 indicates higher indoor pesticide concentrations, whereas a ratio less than 1 indicates higher outdoor pesticide concentrations. We compared indoor and outdoor mean ratios in proximal versus non-proximal and farmworker versus non-farmworker households using a non-parametric Kruskal–Wallis one way ANOVA test (α = 0.05).

Finally, we calculated the Spearman’s correlation coefficient (Rs) between air concentrations and surface deposition indoors. Since replicate sampling can influence correlation results, the mean of replicate samples was used for this calculation. All statistical calculations were performed in Stata (version 11.2; College Station, TX). We did not compare group results of outdoor and indoor air concentrations or indoor surface deposition for the winter samples due to limitations of small sample size. We deployed samples in the winter to primarily test for presence of OP pesticides during a dormant season.


Outdoor Air Concentrations

We present the results of outdoor air concentrations by household type in Table 2. All air samples yielded detectable CPF and CPF-O. During the spring, cumulative residential air concentrations of CPF ranged from 9.2 to 199 ng/m3, and concentrations of CPF-O ranged from 0.03 to 20 ng/m3. We identified the highest levels of CPF (3 of 36 samples > 100 ng/m3) at proximal farmworker households within 100 m of apple, peach, corn, or wheat fields. We identified the highest levels of CPF-O (3 of 36 samples > 13 ng/m3) at both proximal farmworker and proximal non-farmworker households within 100 m of apple, peach, corn, or wheat fields.

Table 2. Select View Table (HTML Version) for a 508-conformant version

Table 2. Summary of outdoor and indoor air concentrations (ng/m3).

View Table (HTML Version)
View larger image (TIF File)

Although 29 of 30 (97%) air samples yielded detectable AZM, only 10 of 30 (33%) samples had detectable AZM-O. During the summer, cumulative air concentrations of AZM and AZM-O were lower than for CPF and CPF-O. Air concentrations of AZM ranged from < LOD to 7.3 ng/m3 and AZM-O ranged from < LOD to 0.8 ng/m3. We identified the highest levels of AZM (3 of 30 samples > 4 ng/m3) and AZM-O (3 of 30 samples > 0.3 ng/m3) at proximal farmworker households within 200 m of apple, peach, and cherry fields.

There were significant differences in outdoor air concentrations of CPF, CPF-O, and AZM between proximal-farmworker, proximal non-farmworker, non-proximal farmworker, and non-proximal non-farmworker households (Table 2, 2-way Friedman’s test, p < 0.10). Proximal households had higher mean outdoor air concentrations of CPF, CPF-O, and AZM than non-proximal households; and farmworker households also had significantly higher mean outdoor air concentrations of CPF and CPF-O than non-farmworker households (Table 3, Kruskal–Wallis test, p < 0.05).

Table 3. Select View Table (HTML Version) for a 508-conformant version

Table 3. Comparisons between proximal versus non-proximal and farmworker versus non-farmworker for outdoor air concentrations (ng/m3), indoor air concentrations (ng/m3), and surface deposition (ng/cm2).

View Table (HTML Version)
View larger image (TIF File)

Indoor Air Concentrations

We also present the results of indoor air concentrations by household type in Table 2. Overall, cumulative indoor air concentrations were lower than outdoor concentrations. For example, 21 of 27 (78%) indoor air samples yielded detectable levels of CPF, and only 7 of 27 (26%) had detectable levels of CPF-O. During the spring, indoor air concentrations of CPF ranged from < LOD to 18 ng/m3, and all concentrations of CPF-O were ≤ 0.6 ng/m3. We identified the highest levels of indoor CPF (4 of 27 samples > 9 ng/m3) in proximal and non-proximal farmworker households. Overall, farmworker households had higher indoor air concentrations of CPF than non-farmworker households (Table 3, Kruskal–Wallis test, p < 0.05).

During the summer, indoor air concentrations of AZM were lower than CPF, ranging from < LOD to 0.8 ng/m3 (Table 2). For example, 9 of 26 (35%) indoor air samples yielded detectable levels of AZM, and only 3 of 26 (12%) had detectable levels of AZM-O. There were no significant differences in indoor air concentrations of AZM or AZM-O between farmworker and non-farmworker and proximal and non-proximal households (Table 3). We identified the highest levels of indoor AZM (2 of 26 samples > 0.2 ng/m3) in non-proximal farmworker households. All indoor AZM air samples in non-farmworker households were < LOD.

Indoor/Outdoor Concentration Ratios

We present the mean indoor/outdoor ratios by household type in Table 4. All households reported CPF and CPF-O indoor/outdoor ratios < 1, except for one proximal farmworker household that reported a CPF indoor/outdoor ratio of 1.3. The overall indoor/outdoor ratio during spring was 0.17 and 0.05 for CPF and CPF-O, respectively. This indicated higher concentrations outdoors as compared to indoors. Farmworker households reported higher indoor/outdoor ratios of CPF than non-farmworker households (Kruskal–Wallis test, p < 0.001).

Table 4. Select View Table (HTML Version) for a 508-conformant version

Table 4. Indoor/outdoor air concentration ratios by household type.

View Table (HTML Version)
View larger image (TIF File)

Most households reported AZM indoor/outdoor ratios less than 1, except for two non-proximal farmworker households that reported indoor/outdoor ratios of 2.1 and 2.5. The overall indoor/outdoor ratio during the summer was 0.44 and 0.72 for AZM and AZM-O, respectively. Many of the reported ratios for CPF-O and AZM-O included substitute values for measurements below the LOD.

Indoor Surface Deposition

We present the results of surface deposition by household type in Table 5. Surface deposition measurements for CPF ranged from < LOD to 5.7 ng/cm2, and 15 of 27 (55%) measurements were < LOD. Surface deposition measurements for AZM were lower than CPF, ranging from < LOD to 1.6 ng/cm2, and 7 of 26 (27%) surface deposition measurements were < LOD for AZM. Overall, proximal households had higher levels of CPF on surfaces than non-proximal households (Table 3, Kruskal–Wallis test, p < 0.05). We observed very low levels of oxygen analogs in surface deposition samples (all ≤ 0.3 ng/cm2). We identified the highest deposition levels of CPF (4 of 27 samples ≥ 1 ng/cm2) at two proximal farmworker and two non-proximal farmworker households. We identified the highest deposition levels of AZM (4 of 26 samples ≥ 0.5 ng/cm2) at one proximal farmworker and three non-proximal farmworker households. The correlation was stronger for indoor CPF (Rs = 0.83, p < 0.001) than for AZM (Rs = 0.49, p < 0.04). We do not report correlations for oxygen analogs due to the large number of samples below the LOD.

Table 5. Select View Table (HTML Version) for a 508-conformant version

Table 5. Summary of indoor surface deposition (ng/cm2).

View Table (HTML Version)
View larger image (TIF File)

Winter Season (Control) Results

During the winter, outdoor air concentrations of CPF ranged from < LOD to 5.8 ng/m3, and CPF-O ranged from < LOD to 0.4 ng/m3 (Table 2). All air samples for AZM and AZM-O were below the LOD. Two proximal farmworker households had detectable indoor air concentrations of CPF ranging from 0.02 to 0.9 ng/m3; and all other indoor air samples were below the LOD for CPF-O, AZM, and AZM-O. All indoor surface deposition samples were below the LOD. During the winter, the overall indoor/outdoor ratio for CPF was 0.06.


Our study is the first to use simultaneous passive sampling methods to measure outdoor air concentrations, indoor air concentrations, and surface deposition of OP pesticides and their oxygen analogs in a residential setting. The passive methods captured monthly exposure estimates of CPF, CPF-O, AZM, and AZM-O with agreement between replicate samples and relatively low limits of detection when compared to more traditional active air sampling methods (NIOSH 1994; U.S. EPA 1999). In addition, the passive methods were minimally invasive to research participants. For example, two participants stated that they “hardly noticed the sampler was there.” There is great potential for the use of more passive sampling methods (such as PUF-PAS) in future epidemiology studies, particularly those being conducted in rural areas with limited outdoor electricity. The PUF-PAS provides good means of comparison because larger numbers of samples can be deployed over time, providing useful information for geographical information systems. We found the passive devices to be relatively low cost (e.g., the passive matrices were approximately 1% of the overall cost of daily active air sampling matrices required for the same time period). Since the samplers report cumulative exposures over the course of an entire month, researchers no longer need to exclusively rely on producer application reporting. However, there are some limitations to passive sampling. Since passive samplers report monthly averages, it is not possible to specify ‘peak’ exposure days. We have also found that sampling rates are highly influenced by meteorological factors, such as wind velocity (Armstrong et al. 2014b). In this study, we were able to control for such factors by using depuration compounds, since their rate of loss is also affected by temperature and wind velocity (Tuduri et al. 2006).

Overall, we found that outdoor and indoor air concentrations and surface deposition results for CPF and CPF-O during the spring were 5–10 times higher than AZM and AZM-O during the summer. We continued to measure low levels of airborne CPF and CPF-O (< 6 ng/m3) in a subset (k = 6) of locations during the dormant winter season. Since the ban on the use of AZM occurred in the year following this study (2012), it is possible that during the summer of 2011 tree fruit producers had already begun to use alternative products. This may have resulted in lower levels of AZM and AZM-O than we expected.

All reported outdoor air concentrations were within the range of concentrations reported in previous studies in California and Washington states (CARB 1998; CDPR 2006, 2009; Fenske et al. 2009). The indoor CPF air concentrations were within or below the range of concentrations reported in a 2004–2005 residential study in New York City conducted by Columbia University (Whyatt et al. 2007). In the present study, the levels of indoor air concentrations of CPF were 0.3–17.5 ng/m3, as compared to 0.4–177 ng/m3 in the Columbia study. However, CPF was used residentially in New York City for treatment of pests in homes and apartment buildings until 2002. In the present study, CPF was used primarily for outdoor agricultural purposes.

Outdoor air concentrations were higher for households in close proximity to tree fruit fields and households with farmworkers than outdoor concentrations at non-proximal households and non-farmworker households, respectively (Table 3). Proximal households (< 250 m from the nearest tree fruit field) had significantly higher mean outdoor air concentrations of CPF, CPF-O, and AZM (p = 0.02, 0.01, and < 0.001, respectively). Various studies have previously demonstrated associations between proximity and higher residential OP pesticide levels in air, dust, and in biomarkers of near-by residents (Loewenherz et al. 1997; Lu et al. 2000; Fenske et al. 2002). However, we defined proximal household by distance (in meters) to only tree fruit fields, and this definition was limited. First, it was unknown if the tree fruit field had been applied with OP pesticides during the sampling period. Second, during the course of the study, we learned that the highest levels of CPF air concentrations were measured at three proximal farmworker households that were within 100 m of corn and wheat fields—in addition to tree fruit fields. In the future, proximity to grain fields should also be considered in geographical regions where corn and grain is more widespread, as CPF is used to control worms, corn borer, and aphid pests in corn and wheat (Gomez 2009).

We found that air concentrations of CPF were lower indoors as compared to outdoors. The trend was similar for AZM, but it was not statistically significant. Inside the home, very little CPF-O or AZM-O was detected. This was expected, as there is less photolysis (via ultraviolet light) to break down parent compounds.

We identified higher indoor air concentrations of CPF in households with close proximity to tree fruit fields (p = 0.03) and farmworker status (p = 0.01) when compared to households that were non-proximal and did not have farmworkers (Table 3). These findings are similar to other studies that have identified farming households as more contaminated (Simcox et al. 1995; Bradman et al. 1997).

Overall, the indoor/outdoor ratios were lower for CPF than for AZM. During this study, we noted another important factor affecting indoor infiltration. At the end of the study period, the promotora asked household members if they remembered opening the windows during the spring, summer, and winter seasons. During the spring season (while sampling for CPF), only 2 of the households indicated opening windows due to colder weather; whereas during the summer season (while sampling for AZM), 10 of the households indicated opening the windows rather than using air conditioning. During the winter season, no households reported opening windows. The open windows may have contributed to the difference in indoor/outdoor ratios by allowing more AZM to come indoors due to higher air exchange rates (Laumbach et al. 2015). Nevertheless, we found that farmworker households reported higher indoor/outdoor ratios for CPF and CPF-O. Therefore, the potential source of indoor pesticides in non-proximal farmworker households may be more attributable to take home pesticide exposure rather than from outdoor infiltration. To test this theory, future studies should include more factors influencing indoor/outdoor ratios, such as open and closed windows, number of people living in the home, number and type of farmworkers in the home, and type of air conditioning and heating units.

Although indoor surface depositions of CPF and AZM were higher in proximal households than non-proximal households, there were no statistically significant differences observed between farmworker and non-farmworker household deposition samples (Table 3). There was good correlation between indoor surface deposition measurements and air concentrations (Rs = 0.83 and 0.49 for CPF and AZM, respectively).

There were some limitations to this study. First, we relied on very simple non-parametric statistical test methods rather than multivariable modeling because we were very limited by small sample size. In particular, we found it difficult to identify non-farmworker households that were also proximal since such a large portion of the population is involved in agriculture. Since this was our first attempt to deploy the PUF-PAS samplers for pesticides in a residential setting, we refrained from conducting a larger study in more households. Second, many indoor air and surface deposition samples were below limits of detection, and we had to rely on substituted values for analysis. For future studies using indoor sampling methods for OP pesticides, we suggest using sampling periods of 3–6 months rather than only 1 month. Third, we did not account for the non-independence of replicate samples from the same location and time period, although we deployed replicate samples across all household groups (see Table S1). Finally, although our ideal sampling period was 1 month, the sampling periods ranged from 24 to 32 days, since we had to coordinate the deployment schedule with household members. Although there is variation in pesticide use within a season, it was unlikely that this variation contributed to differences in pesticide levels, as there were no significant differences in sampling deployment periods between household groups.


We demonstrated the use of passive sampling methods for measuring long-term (1 month) exposures to OP pesticides and oxygen analogs in a remote agricultural area, and encourage others researchers to explore the use of passive sampling devices (like the PUF-PAS) in their region. Exposure data is currently lacking for sub-chronic and chronic epidemiological investigations in rural communities.

We have used passive sampling methods to identify higher outdoor and indoor air concentrations and surface deposition of OP pesticides and their oxygen analogs at both proximal (< 250 m of a tree fruit field) and farmworker households. This study has further confirmed our previous findings on the presence of OP pesticide oxygen analogs in air. On a residential level, human exposures to these oxygen analogs seem to be a greater concern outdoors than indoors. We have found that both proximal and farmworker households have higher levels of exposure to these airborne compounds. When considering cumulative and aggregate effects of human exposure to OP pesticides, the inclusion of oxygen analogs in future risk assessments will be necessary—especially if spending large quantities of time outdoors in rural agricultural areas near applied fields. More research is required to describe the community transport of these pesticide mixtures and how oxygen analogs are formed in outdoor environments.


Armstrong JL, Dills RL, Yu J, Yost MG, Fenske RA. 2014a. A sensitive LC-MS/MS method for measurement of organophosphorus pesticides and their oxygen analogs in air sampling matrices. J Envrion Sci Health B 49:102–108.

Armstrong JL, Fenske RA, Yost MG, Galvin K, Tchong-French M, Yu J. 2013a. Presence of organophosphorus pesticide oxygen analogs in air samples. Atmos Environ (1994) 66:145–150.

Armstrong JL, Fenske RA, Yost MG, Tchong-French M, Yu J. 2013b. Comparison of polyurethane foam and XAD-2 sampling matrices to measure airborne exposures to organophosphorus pesticides and their oxygen analogs in an agricultural community. Chemosphere 92:451–457.

Armstrong JL, Yost MG, Fenske RA. 2014b. Development of a passive air sampler to measure airborne organophosphorus pesticides and oxygen analogs in an agricultural community. Chemosphere 111:135–143.

Baker NT, Stone WW. 2015. Estimated Annual Agricultural Pesticide Use for Counties of the Conterminous United States, 2008–12. U.S. Geological Survey Data Series 907.​df [accessed 1 April 2016].

Barr DB, Bravo R, Weerasekera G, Caltabiano LM, Whitehead RD Jr, Olsson AO, et al. 2004. Concentrations of dialkyl phosphate metabolities of organophosphorus pesticides in the U.S. population. Environ Health Perspect 112:186–200, doi: 10.1289/ehp.6503.

Bohlin P, Jones KC, Levin JO, Lindahl R, Strandberg B. 2010. Field evaluation of a passive personal air sampler for screening of PAH exposure in workplaces. J Environ Monit 12:1437–1444.

Bradman MA, Harnly ME, Draper W, Seidel S, Teran S, Wakeham D, et al. 1997. Pesticide exposure to children from California’s Central Valley: results of a pilot study. J Expo Anal Environ Epidemiol 7:217–234.

Butte W, Heinzow B. 2002. Pollutants in house dust as indicators of indoor contamination. Rev Environ Contam Toxicol 175:1–46.

CARB (California Air Resources Board). 1998. Report for the Application and Ambient Air Monitoring of Chlorpyrifos (and oxon analogue) in Tulare County During Spring/Summer 1996. Sacramento, CA:California Air Resources Board.

CDPR (California Department of Pesticide Regulation). 2006. Environmental Justice Pilot Project: Pesticide Air Monitoring in Parlier, Second Progress Report. Attachment IV. Sacramento, CA:California Environmental Protection Agency, CDPR.

CDPR. 2009. Pesticide Air Monitoring in Parlier, CA (Final Report). Sacramento, CA:California Environmental Protection Agency, CDPR.

Costa LG, Cole TB, Vitalone A, Furlong CE. 2005. Measurement of paraoxonase 1 (PON1) status as a potential biomarker for organophosphate toxicity. Clin Chim Acta 352:37–47.

Fenske RA, Lu C, Barr D, Needham L. 2002. Children’s exposure to chlorpyrifos and parathion in an agricultural community in central Washington State. Environ Health Perspect 110:549–553.

Fenske RA, Lu C, Curl CL, Shirai JH, Kissel JC. 2005. Biological monitoring to characterize organophosphorus pesticide exposure among children and workers: an analysis of recent studies in Washington State. Environ Health Perspect 113:1651–1657, doi: 10.1289/ehp.8022.

Fenske RA, Yost M, Galvin K, Tchong M, Negrete M, Palmendez P, et al. 2009. Organophosphorus Pesticide Air Monitoring Project. Final Report. University of Washington; Washington State Department of Health Pesticide Program.​s/4300/Pesticide-UW-OP-Report.pdf [accessed 20 May 2011].

Gomez LE. 2009. Use and Benefits of Chlorpyrifos in U.S. Agriculture. Dow Indianapolis, IN:AgroSciences LLC.​ifosUseBenefits_18May2009.pdf [accessed 31 May 2011].

Hourani GF, Underhill DW. 1988. Long term passive sampling of environmental airborne contaminants. In: Advances in Air Sampling. American Conference of Governmental Industrial Hygienists (ACGIH) Symposium, 16–18 February 1987, Pacific Grove, California, Chelsea, MI:Lewis Publishers.

Howard PH. 1991. Handbook of Environmental Fate and Exposure Data for Organic Chemicals. Vol. III: Pesticides. Chelsea, MI:Lewis Publishers, 5–13.

Keenan JJ, Ross JH, Sell V, Vega HM, Krieger RI. 2010. Deposition and spatial distribution of insecticides following fogger, perimeter sprays, spot sprays, and crack and crevice applications for treatment and control of indoor pests. Regul Toxicol Pharmacol 58:189–195.

Laumbach R, Meng Q, Kipen H. 2015. What can individuals do to reduce personal health risks from air pollution? J Thorac Dis 7:96–107.

Lewis RG. 2005. Residential post-application pesticide exposure monitoring. In: Occupational and Residential Exposure Assessment for Pesticides. Franklin CA, Worgan JP, eds. Sussex, UK:John Wiley and Sons, 71–128.

Loewenherz C, Fenske RA, Simcox NJ, Bellamy G, Kalman D. 1997. Biological monitoring of organophosphorus pesticide exposure among children of agricultural workers. Environ Health Perspect 105:1344–1353.

Lu C, Fenske RA. 1998. Air and surface chlorpyrifos residues following residential broadcast and aerosol pesticide applications. Environ Sci Technol 32:1386–1390.

Lu C, Fenske RA, Simcox NJ, Kalman D. 2000. Pesticide exposure of children in an agricultural community: evidence of household proximity to farmland and take home exposure pathways. Environ Res 84:290–302.

Meylan WM, Howard PH. 2005. Estimating octanol-air partition coefficients with octanol-water partition coefficients and Henry’s law constants. Chemosphere 61:640–644.

NCBI (National Center for Biotechnology Information). 2015a. PubChem Database; CID=2268, Azinphos-methyl.​d/2268 [accessed 15 December 2015].

NCBI. 2015b. PubChem Database; CID=2730, Chlorpyrifos.​d/chlorpyrifos [accessed 15 December 2015].

NIOSH (National Institute for Occupational Safety and Health). 1994. Method 5600. Organophosphorous pesticides. In: NIOSH Manual of Analytical Methods. 4th ed. Cincinnati, OH:Centers for Disease Control and Prevention.​dfs/5600-F.pdf [accessed 15 June 2011].

Quandt SA, Hernández-Valero MA, Grzywacz JG, Hovey JD, Gonzales M, Arcury T. 2006. Workplace, household, and personal predictors of pesticide exposure for farmworkers. Environ Health Perspect 114(6):943–952.

Roberts JF, Karr CJ, Council on Environmental Health. 2012. Technical Report: Pesticide exposure in children. Pediatrics 130:1765–1788.

Schuster JK, Gioia R, Harner T, Lee SC, Breivik K, Jones KC. 2012. Assessment of sorbent impregnated PUF disks (SIPs) for long-term sampling of legacy POPs. J Environ Monit 14:71–78.

Shih DM, Gu L, Xia YR, Li WF, Castellani LW, Furlong CE, et al. 1998. Mice lacking serum paraoxonase are susceptible to organophosphate toxicity and atherosclerosis. Nature 394:284–287.

Shoeib M, Harner T. 2002. Characterization and comparison of three passive air samplers for persistent organic pollutants. Environ Sci Technol 36:4142–4151.

Simcox NJ, Fenske RA, Wolz SA, Lee IC, Kalman DA. 1995. Pesticides in household dust and soil: exposure pathways for children of agricultural families. Environ Health Perspect 103:1126–1134.

Thompson B, Coronado GD, Vigoren EM, Griffith WC, Fenske RA, Kissel JC, et al. 2008. Para Ninos Saludables: a community intervention trial to reduce organophosphorus pesticide exposure in children of farmworkers. Environ Health Perspect 116:687–694, doi: 10.1289/ehp.10882.

Tuduri L, Harner T, Hung H. 2006. Polyurethane foam (PUF) disks passive air samplers: wind effect on sampling rates. Environ Pollut 144:377–383.

U.S. Census Bureau Geography Division. 2010. MAF/Tiger Database.​a/tiger-line.html [accessed 20 May 2011].

U.S. EPA (U.S. Environmental Protection Agency). 1999. Method TO-10A: determination of pesticides and polychlorinated biphenyls in ambient air using low volume polyurethane foam (PUF) sampling followed by gas chromatographic/multi-detector detection (GC/MD). In: Compendium of Methods of the Determination of Toxic Organic Compounds in Ambient Air. 2nd ed. EPA/625/R-96/010b. Cincinnati, OH:U.S. EPA.​ent/airtox/to-10ar.pdf [accessed 1 June 2011].

USDA (U.S. Department of Agriculture). 2008. Washington’s Agricultural Fruit Chemical Usage, 2007. National Agricultural Statistics Service. Apples. Washington Field Office, Oympia, WA. [accessed 11 June 2011].

USDA. 2009. 2007 Census of Agriculture. United States Summary and State Data: Washington. Vol 1: Part 51. AC-07-A-51. [accessed 11 June 2011].

USDA-NASS (U.S. Department of Agriculture, National Agricultural Statistics Service. 2012. National Agricultural Statistics Service Cropland Data Layer. 2005–2011. Washington, DC:USDA-NASS. [accessed 1 June 2012].

Wauchope RD, Buttler TM, Hornsby AG, Augustijn-Beckers PWM, Burt JP. 1992. SCS/ARS/CES pesticide properties database for environmental decision-making. Rev Environ Contam Toxicol 123:1–155.

Whyatt RM, Garfinkel R, Hoepner LA, Holmes D, Borjas M, Williams MK, et al. 2007. Within- and between-home variability in indoor-air insecticide levels during pregnancy among an inner-city cohort from New York City. Environ Health Perspect 115:383–389, doi: 10.1289/ehp.9546.

Zimmerman DW, Zumbo BD. 1993. Relative power of the Wilcoxon test, the Friedman test, and repeated-measures ANOVA on ranks. J Exp Educ 62:75–86.

Estimating State-Specific Contributions to PM2.5‐ and O3-Related Health Burden from Residential Combustion and Electricity Generating Unit Emissions in the United States

Author Affiliations open
1Department of Environmental Health, Boston University School of Public Health, Boston, Massachusetts, USA; 2Institute for the Environment, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, USA; 3U.S. Environmental Protection Agency, Durham, North Carolina, USA; 4Department of Biostatistics, Boston University School of Public Health, Boston, Massachusetts, USA

PDF icon PDF Version (1.3 MB)

  • Background: Residential combustion (RC) and electricity generating unit (EGU) emissions adversely impact air quality and human health by increasing ambient concentrations of fine particulate matter (PM2.5) and ozone (O3). Studies to date have not isolated contributing emissions by state of origin (source-state), which is necessary for policy makers to determine efficient strategies to decrease health impacts.

    Objectives: In this study, we aimed to estimate health impacts (premature mortalities) attributable to PM2.5 and O3 from RC and EGU emissions by precursor species, source sector, and source-state in the continental United States for 2005.

    Methods: We used the Community Multiscale Air Quality model employing the decoupled direct method to quantify changes in air quality and epidemiological evidence to determine concentration–response functions to calculate associated health impacts.

    Results: We estimated 21,000 premature mortalities per year from EGU emissions, driven by sulfur dioxide emissions forming PM2.5. More than half of EGU health impacts are attributable to emissions from eight states with significant coal combustion and large downwind populations. We estimate 10,000 premature mortalities per year from RC emissions, driven by primary PM2.5 emissions. States with large populations and significant residential wood combustion dominate RC health impacts. Annual mortality risk per thousand tons of precursor emissions (health damage functions) varied significantly across source-states for both source sectors and all precursor pollutants.

    Conclusions: Our findings reinforce the importance of pollutant-specific, location-specific, and source-specific models of health impacts in design of health-risk minimizing emissions control policies.

  • Citation: Penn SL, Arunachalam S, Woody M, Heiger-Bernays W, Tripodis Y, Levy JI. 2017. Estimating state-specific contributions to PM2.5‐ and O3-related health burden from residential combustion and electricity generating unit emissions in the United States. Environ Health Perspect 125:324–332;

    Address correspondence to S.L. Penn, Boston University School of Public Health, Department of Environmental Health, 715 Albany St. 4W, Boston, MA 02118 USA. Telephone: (617) 638-5881. E-mail:

    We thank B. Harvey from Boston University (BU) for his assistance with image segmentation algorithm processing and M. Woo from BU for her assistance in final manuscript preparation.

    This research was supported by the North American Insulation Manufacturers Association (NAIMA).

    NAIMA suggested the research topic and was provided the opportunity to give comments on the manuscript, but the authors had full editorial control of the content and the findings should not be attributed to NAIMA or its member companies.

    All authors declare they have no actual or potential competing financial interests.

    Received: 20 January 2016
    Revised: 25 May 2016
    Accepted: 23 July 2016
    Published: 2 September 2016

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (2.2 MB)

    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact Our staff will work with you to assess and meet your accessibility needs within 3 working days.

    PDF icon Supplemental Table of Contents PDF (145 KB)


Elevated concentrations of ambient ozone (O3) and fine particulate matter ≤ 2.5 μm in aerodynamic diameter (PM2.5) contribute to adverse health outcomes in exposed populations (Jerrett et al. 2009; Krewski et al. 2009). Epidemiological literature has described relationships between population exposure to these air pollutants and chronic and acute health effects, including premature mortality (Brook et al. 2002; Bell et al. 2004; Ito et al. 2005; Levy et al. 2005) and multiple morbidities (Zanobetti et al. 2009; Ji et al. 2011; Mustafic et al. 2012; Levy et al. 2012).

A number of emitting source sectors that are spatially distributed across the United States contribute to total ambient concentrations of these pollutants, including electricity generating units (EGUs), which burn fossil fuels like coal and natural gas to produce electricity, and residential combustion (RC) sources, including oil and natural gas-burning furnaces or wood-burning stoves to heat homes. Among the most significant contributors to air pollution-related health impacts are emissions related to EGUs, which are elevated stack point sources, and area sources, which are ground-level, widely distributed sources and include RC. In 2005, Fann et al. (2013) estimated EGUs contribute 38,000 premature deaths per year across the United States, highest among source sectors, with area sources contributing another 27,000 premature deaths per year. Similarly, a recent study estimated that EGUs contributed 53,900 premature deaths from PM2.5 and O3 across the United States in 2005, while commercial and residential combustion together contributed 42,150 deaths from PM2.5 and O3 (Caiazzo et al. 2013). Another recent study estimated PM2.5-related health risks of 41,660 premature deaths from EGUs and 35,790 premature deaths from commercial and residential combustion (Dedoussi and Barrett 2014).

While these comparisons provide valuable insight about high-priority source sectors, they do not include information on impacts of specific emitted pollutants from individual states and source types for both PM2.5 and O3. Caiazzo et al. (2013) estimated total premature mortalities by state from a receptor perspective rather than a source perspective (i.e., the premature mortalities for populations living in California rather than the premature mortalities attributable to sources in California) and do not differentiate by emitted pollutant, providing less useful information from a control strategy perspective. Dedoussi and Barrett (2014) estimate total premature mortalities by source-state for PM2.5 using a different modeling approach [adjoint modeling using GEOS (Goddard Earth Observing System)–Chem chemical transport model ( with slightly coarser resolution] and lacking insight about O3-related impacts or impacts by source-state and precursor pollutant. GEOS-Chem chemical transport model with KPP chemical solver and RPMARES aerosol equilibrium model were used. Many federal policies targeting EGUs, including the U.S. Environmental Protection Agency’s (EPA) Clean Power Plan (U.S. EPA 2015a) and Cross-State Air Pollution Rule (U.S. EPA 2016) have mechanisms for differential actions by states, and it is important to understand how alternative combinations of emissions reductions could influence public health. RC may be influenced by a policy like the Clean Power Plan, which considers energy efficiency as one mechanism to achieve emissions reductions, and may be directly targeted as part of State Implementation Plans (SIPs) or other state policy measures. Quantification of source-specific and pollutant-specific health risks by source-state provides a tool for policy makers to create efficient emission control strategies.

Premature mortalities from different source sectors can be estimated by combining source-specific air quality changes with population characteristics and epidemiologically derived concentration–response functions (Fann and Risley 2013; Hubbell et al. 2009; Tagaris et al. 2009). In addition to determining total premature mortalities, health damage functions (estimated as premature mortalities per unit emissions) can be calculated to provide insight about sources and locations in which emissions reductions are more or less efficient from a public health perspective. Heterogeneity in health damage functions is associated with ambient atmospheric chemistry and meteorology, source and chemical profiles of emitted pollutant precursors, and the geographic distribution of exposed populations (Fann et al. 2009). EGU and RC sources provide interesting contrasts: EGUs are individual point sources that vary in location, stack height, age, and efficiency, while RC is a ground-level area source spread over a wider area and directly tied to population patterns. Both are spatially distributed across the United States, with between-sector and within-sector differences including proximity to populations, height of emissions origin, and atmospheric chemistry and meteorology in each location and downwind. Analysis of RC and EGUs specifically allows us to consider two sectors that would be influenced by policies such as the Clean Power Plan that target EGUs but could have ancillary effects on RC (e.g., through residential energy efficiency).

The Community Multiscale Air Quality (CMAQ) model, a peer-reviewed atmospheric chemistry and transport model capable of modeling gas-phase, aerosol, and aqueous chemistry including the formation of O3 and PM2.5 from emitted precursors, can predict changes in ambient air quality associated with these two source sectors, among others (Byun and Ching 1999; Byun and Schere 2006). Utilized with the decoupled direct method (DDM), which decouples sensitivity equations from model equations to allow for stability and accuracy of values and computational efficiency, CMAQ-DDM has the power to determine individual source contributions by analyzing the sensitivity of ambient concentrations of PM2.5 and O3 to specific precursor emissions in the presence of different atmospheric and meteorological conditions (Dunker 1984; Dunker et al. 2002; Koo et al. 2007). CMAQ-DDM has been used in previous studies to quantify exposure to pollutants from source-tagged precursors (Bergin et al. 2008; Odman et al. 2002; Itahashi et al. 2012), and has been used to assess health impacts due to climate change in the United States (Tagaris et al. 2010).

In this study, we quantified premature mortalities from EGUs and RC for each emitted pollutant and source-state individually across the continental United States. We used CMAQ version 4.7.1 (Byun and Ching 1999; Byun and Schere 2006) instrumented with DDM-3D (Dunker 1984; Napelenok et al. 2006) to determine estimated changes in ambient pollutant concentrations of PM2.5 and O3 based on EGU and RC emissions, using these air quality changes to determine predicted total premature mortalities and health damage functions by source-state and sector. This approach will allow state and federal policy makers to determine which sources to target to decrease public health burdens and which policies will be most efficient in achieving improvements. Comparisons of health damage functions by source sector and source-state will allow further assessment of differential attributes of RC and EGU emissions.


Study Design

Key model components are presented in Figure 1. Briefly, to determine changes in ambient air quality associated with EGUs and RC, we used CMAQ (Byun and Ching 1999; Byun and Schere 2006) instrumented with DDM in three dimensions (Dunker 1984; Napelenok et al. 2006). This model isolated PM2.5– and O3-specific contributions from state-wide EGU and RC precursors to assess the sensitivity of ambient pollutant concentrations to these precursors. Resultant ambient pollutant concentrations were then linked with population and mortality rate data from the Centers for Disease Control and Prevention (CDC 2015). Concentration-response functions associating ambient pollutant concentrations with health effects were derived from the epidemiological literature. We estimated total premature mortalities for each source sector by source-state for each precursor pollutant-ambient concentration relationship, including primary elemental carbon (PEC), primary organic carbon (POC), and primary sulfate (PSO4) as primary PM2.5 precursors; nitrogen oxides (NOx), sulfur dioxide (SO2), and volatile organic compounds (VOCs) as secondary PM2.5 precursors; and NOx and VOCs as O3 precursors, detailed in Table S1. We also estimated health damage functions, or premature mortality risk per 1,000 tons of precursor emissions. Emissions details can be found in Figures S1 and S2 and Tables S2 and S3.

Figure 1. Conceptual diagram.

Figure 1. Health damage function model inputs and outputs.

View larger image (TIF File)

CMAQ-DDM Design and Modeling

Due to the computationally intensive nature of CMAQ-DDM, it was not practical to construct separate runs for each source sector and source-state. To maximize efficiency, we incorporated one to three states into a single DDM run for each of RC and EGUs, and we developed algorithms to separate the concentration impacts from each state (described in Section 2.3 and in “Image Segmentation Algorithm” in the Supplemental Material). To design these runs, we overlaid concentration surface results from a pilot analysis of SO2 tracer emissions from multiple source-states and grouped states to minimize errors in source-state attribution with the smallest number of runs. For EGUs, a subset of states cut across electricity dispatch regions, so we subdivided those states into two areas to facilitate future connection with energy efficiency or renewable energy projects. In total, 65 model runs were conducted (described in Table S4), including 25 groups of states for modeling RC and 40 groups of states (and partial states) for modeling EGUs.

Details of the CMAQ-DDM modeling are provided in “Community Multiscale Air Quality (CMAQ) model” in the Supplemental Material. RC sources were modeled as low-level area sources including all residential fuel types, aggregated to county level for apportionment to grid cells by state. EGUs were modeled by power plant and aggregated to grid cells by state. Cells of 36 km × 36 km covering the continental United States were used to grid state-specific emissions from each source sector. Because modeling the full year was computationally intensive, we selected 2 months (January and July) to provide bi-seasonal representation, using all-source emissions and meteorology from 2005. To provide initial background conditions, a spin-up period of 11 days prior to each month was simulated. Whole-month sensitivity values from January and July were averaged to represent annual estimated contributions of statewide RC and EGU sources to ambient PM2.5 and O3 concentrations. Values are reported as 24-hr averages for PM2.5 constituents and 8-hr maximum values for O3 for consistency with current regulatory policies. These values were used in total health impact and health damage function calculations.

Separation of State-Specific Concentration Surfaces

To separate contributions of individual source-state’s contributions to ambient concentrations from one another within a DDM run, we applied image separation techniques using MATLAB 8.1.0, R2013a (MathWorks, Natick, MA). We developed a region-growing algorithm to determine regions of concentrations attributable to each source-state for each emitted precursor and associated ambient pollutant relationship within each model run and season. This algorithm allowed for both positive and negative sensitivities to be included within regions, and ensured that within a run, a smaller state’s region could capture the extent of its health impacts. Quality assurance (QA) analyses were performed, including analysis of total health impact and health damage function distributions for resultant health values, as well as visual inspection of concentration surfaces. For runs that did not meet QA criteria, we re-ran CMAQ-DDM for individual states in isolation. This process allowed determination of emissions impacts from individual source-states within a CMAQ-run group. “Image Segmentation Algorithm” in the accompanying Supplemental Material contains more information regarding the image segmentation algorithm.

Total Health Impact Calculation and Health Damage Function Modeling

Calculation of total premature mortalities by source-state and source sector is analogous to the calculation of health damage functions, with the exception of normalization by precursor emissions. Changes in air quality associated with state-wide emissions were linked with premature deaths using a standard health impact modeling equation, calculated separately for each precursor and associated ambient pollutant pair for each source sector. The equation is as follows:

Delta y equals sum under i equals 1 over N,sum under j equals 1 over M,open left-parenthesis y sub 0 ij, open left-parenthesis e to-the-power-of beta times delta x sub ij,, minus 1 right-parenthesis, Pop sub ij, right-parenthesis

where i is row number and j is column number, N is total number of rows and M is total number of columns in the CMAQ grid. Δy is change in mortality across the continental United States, y0 is baseline mortality incidence rate in grid cell at location ij, β is concentration–response function as derived from the epidemiological literature, Δx is change in air quality for a given precursor in grid cell ij, and Pop is the population of interest in grid cell ij. To associate premature mortalities with PM2.5 concentrations, we applied a central estimate concentration-response function of a 1% increase in mortality associated with every 1-μg/m3 increase in annual average PM2.5 concentration (Roman et al. 2008). To associate premature mortalities with O3 concentrations, we applied a central estimate concentration–response function of 0.4% increase in daily mortality per 10-ppb increase in daily 8-hr maximum O3 concentrations, based on major multi-city and meta-analysis studies that evaluated health impacts across the year (Ji et al. 2011; Bell et al. 2004, 2005; Ito et al. 2005; Levy et al. 2005; Schwartz 2005). To estimate county-wide population and baseline mortality rates for adults ≥ 25 years old in 2005, values from 2001 to 2010 were obtained from CDC WONDER (CDC 2015) and averaged for stability of values. County-wide values were projected as Lambert conformal conic in ArcMap (version 10.1; ESRI, Redlands, CA, USA) and intersected with grid cells, assuming uniform density of population and mortality rate within counties.

Total premature mortalities were calculated by emitted precursor and associated ambient pollutant pair for each source-state for both EGUs and RC, assuming January and July each represent 6 months. These 6-month values were summed to obtain annual health impact estimates. Health damage function values were calculated by normalizing total premature mortalities by total amount of emitted precursor for January and July, each representative of what the health damage function would be if these individual month conditions were present for an entire year. Annual health damage function estimates were calculated by averaging January and July health damage functions, interpreted as the mortality risk associated with uniform emissions across the year. Ozone estimates were calculated for both January and July given epidemiological evidence based on year-round exposures.

Comparison of RC and EGU Source Sectors

Descriptive statistics were calculated for total premature mortalities and health damage functions for EGU and RC by precursor and source-state. We examined between-state variation in total premature mortalities and health damage functions by source sector and precursor pollutant, as well as between-pollutant and within-state variation. To facilitate interpretation, we calculated the percentage of source-state mortalities found within that state (i.e., percentage of deaths from California RC emissions that occur in California), examined emissions inventories and mapped source locations.


Total Health Impacts

Total number of premature mortalities per year for each precursor were modeled for each state for both RC and EGUs (Figure 2; see also Tables S5 and S6). RC contributes 10,000 additional deaths per year, and EGUs contribute 21,000 additional deaths per year from both PM2.5 and O3.

Figure 2. (A) Map of the continental United States indicating estimated numbers of premature deaths in all states associated with the RC emissions from each individual state. (B) Map of the continental United States indicating estimated numbers of premature deaths in all states associated with the EGU emissions from each individual state. (C) Map of the continental United States indicating the estimated percentage of premature deaths in each state from RC emissions in the same state. (D) Map of the continental United States indicating the estimated percentage of premature deaths in each state from EGU emissions in the same state.

Figure 2. (A) Total premature deaths associated with source-state RC emissions (e.g., California RC emissions caused 980 premature deaths across all states). (B) Total premature deaths associated with source-state EGU emissions (e.g., EGU emissions from Ohio caused 2,300 premature deaths across all states) (U.S. Census Bureau 2016). (C) Percentage of source-state premature deaths from RC emissions occurring in the source-state (e.g., 93% of the 980 premature deaths from California RC emissions occurred in California). (D) Percentage of source-state premature deaths from EGU emissions occurring in the source-state (e.g., 21% of the 2,300 premature deaths from Ohio EGU emissions occurred in Ohio).

View larger image (TIF File)

States contributing the most deaths related to RC are those with combustion-type home heating near or upwind of highly populated areas, including Ohio, California, Maryland, and New York (Figure 2a). RC emissions are tied to population, so highly populated areas will have both greater emissions and greater exposed populations. Primary PM2.5 precursors contribute 74% of premature mortalities for RC, driven by POC, and the vast majority of primary PM2.5 emissions are associated with wood burning (see Figure S2). The percentage of RC-related premature mortalities found within the source-state varies widely across states (Figure 2C), with values exceeding 75% in geographically large states without substantial downwind populations (e.g., Washington, California, Florida) and values below 10% in smaller states with large downwind populations (e.g., Washington District of Columbia, Delaware, Vermont).

States with the greatest total mortalities from EGUs are those with the greatest coal-fired power plant emissions upwind of highly populated areas, including Ohio, Indiana, and Pennsylvania (Figure 2B). For EGUs, SO2 contributes most to premature mortality burden, with 77% of premature mortalities related to secondary PM2.5 or O3 attributable to SO2 and NOx. The vast majority of SO2 and NOx emissions from EGUs are related to coal combustion (see Figure S1). As anticipated, given the dominance of secondarily-formed pollutants, the percentage of premature mortalities found within the source-state for EGUs is less than that for RC (22% vs. 38% overall). In contrast to RC, only 3 states have more than half of their EGU-related health impacts within the source-state (California, Florida, and Washington), with only 12 states having > 25% of their EGU-related health impacts within the source-state (Figure 2D).

Ratios of RC-related deaths to EGU-related deaths vary greatly across source-states. Deaths from RC exceed those from EGUs for source-states in the Northeast and West Coast where population density is high, EGU coal combustion is limited, and wood or oil is used in some homes for heating. In contrast, deaths from EGUs exceed those from RC in source-states with appreciable EGU coal combustion and significant usage of electricity for home heating. Excluding the five lowest emitting states for primary PM2.5 from RC and SO2 from EGUs, where health damage functions may be biased due to limited emissions (explained further in Section 3.2), ratios of EGU-related deaths to RC-related deaths vary from 0.05 to 20 across source-states.

There is significant seasonal variation in total premature mortalities by source sector and precursor-pollutant pair. RC-related deaths are dominated by cold weather emissions, as deaths are 20 times greater for January (representing cold months) versus July (representing warm months). RC emissions are greatest for January in the Northwest, Midwest and Northeast, driven by climate, population density, and fuel types (see Figure S3). Conversely, EGU-related deaths are 5 times greater for July than for January, given the substantial contribution from SO2 emissions and enhanced secondary particle formation from SO2 in warmer seasons. EGU emissions of SO2 are most prominent in the Midwest and Mid-Atlantic regions (see Figure S4). The impact of NOx on O3 has an inverse relationship with deaths in January due to O3 titration in cold weather and a positive relationship with deaths in July, as high temperatures are needed for O3 formation and high ambient NOx can contribute to VOC-limited regimes.

Health Damage Functions

Health damage functions for RC and EGUs were modeled for each precursor and season for each source-state. Figure 3 shows the distribution of health damage functions for RC and EGUs by precursor for January and July. Health damage functions for primary PM2.5 precursors are greatest on average for January EGU emissions, while distributions of RC and EGU July health damage functions for primary PM2.5 precursors are similar to one another. States with very low emissions provide abnormally inflated health damage functions, which have been excluded from Figure 3 (but shown in Tables S7–S10).

Figure 3. Eight box-and-whiskers plots summarizing the distribution of estimated mortality risk per 1,000 tons of precursor emissions for PM2.5 related to PEC, POC, PSO4, NOx, SO2, VOCs; and for ozone emissions related to NOx and VOC, respectively (y-axes) based on January RC, July RC, January EGU, and July EGU, respectively (x-axes).

Figure 3. Box plots of health damage functions for RC and EGUs for January and July by precursor-pollutant pair. (A) Health damage functions as mortality risk per 1,000 tons precursor emissions for PM2.5 related to PEC; (B) PM2.5 related to POC; (C) PM2.5 related to PSO4; (D) PM2.5 related to NOx; (E) PM2.5 related to SO2; (F) PM2.5 related to VOC; (G) O3 related to NOx; (H) O3 related to VOC. Note: y-axes display different ranges for each panel. Boxplots show 5%, first quartile, median, third quartile, and 95% values for each precursor and associated pollutant damage function.

View larger image (TIF File)

Across both source sectors, health damage function values are much smaller for secondary pollutants compared with primary pollutants. SO2-PM2.5 damage functions display more seasonality than NOx-PM2.5, with heightened impacts per unit emissions in July. NOx-O3 health damage functions are generally negative for RC in January but positive in July. NOx-O3 health damage functions for EGUs display smaller negative values in January and less variability overall. VOC-O3 health damage functions are significantly higher for RC than for EGUs in both seasons.

While Figure 3 is able to show the range of health damage functions for both source sectors, it does not describe their relationship on a state-by-state basis, which is important in understanding relative magnitudes of pollutant impacts from different sources. The relationship between health damage functions for RC and EGUs varies greatly across states (see Figures S5 and S6). Many states with low RC primary PM2.5 health damage functions also have low EGU primary PM2.5 damage functions, especially for July emissions. States where RC and EGU primary PM2.5 health damage functions differ greatly from one another (e.g., South Dakota, Montana, Maine, Oklahoma) tend to be large low-population states where EGUs are located in areas geographically removed from the locations of RC combustion (see Figures S3 and S4). In comparison with primary PM2.5, the association between EGU and RC health damage functions is similar for NOx but not SO2 and VOCs. O3-related health damage functions for EGUs are smaller in magnitude than those for RC, with an inverse association between RC and EGU values.


We estimated the premature mortality burden of ambient PM2.5 and O3 concentrations attributable to RC and EGU emissions by source-state and precursor pollutant in the continental United States using CMAQ-DDM and health damage function modeling based on 2005 air quality and population estimates. Health impacts of these source sectors have not previously been compared directly, nor has the literature provided insight about dominant pollutants and source-states. We quantify 10,000 additional premature deaths per year due to RC emissions and 21,000 additional premature deaths per year due to EGU emissions, with RC health impacts dominated by PEC and POC emissions and EGU health impacts dominated by SO2 and NOx emissions (forming PM2.5 and O3).

Comparing Total Health Impacts with Other Studies

While comparisons with previous studies are challenging given underlying model differences, examination of similarities and differences in estimates can provide insights about our findings. Total mortalities associated with EGUs have been previously calculated for the continental United States for 2005 from PM2.5 and O3 (Caiazzo et al. 2013; Fann et al. 2013). Fann et al. (2013) found EGUs were responsible for 38,000 premature deaths in 2005 versus the 21,000 in our study. For RC, while Fann et al. (2013) do not report a value directly, their sectoral values imply approximately 8,000 deaths per year from residential wood combustion. The vast majority of our 10,000 attributable premature deaths are likely related to wood combustion given its dominance in primary PM2.5 emissions. In addition, EPA recently published a regulatory impact analysis for residential wood heaters and utilized data from Fann et al. (2013) to determine 0.07 deaths per ton of primary PM2.5 emissions (U.S. EPA 2015b), identical to our national average value. Caiazzo et al. (2013) estimated EGUs caused 52,000 premature deaths from PM2.5 and 1,700 premature deaths from O3, and commercial and residential combustion combined contributed 41,800 deaths from PM2.5 and 350 deaths from O3 in 2005. While we found O3 contributed 2,000 premature deaths from EGUs and 320 premature deaths from RC, values in line with Caiazzo et al. (2013) estimates, our estimates for PM2.5-related premature deaths are a factor of 2–3 lower for EGUs and a factor of 4 lower for RC, albeit with commercial combustion included in Caiazzo et al. (2013). All three studies analyzed health impacts for 2005 conditions using the National Emissions Inventory, yet magnitude differences are expected given utilization of different atmospheric dispersion models [CMAQ-DDM, version 4.7.1 in our study; CMAQ, version 4.7.1 brute force in Caiazzo et al. (2013), Comprehensive Air Quality Model with Extensions (CAMX), version 5.30 using SMAT/MATS for Fann et al. (2013)] and different concentration–response functions.

Total Health Impact Analysis

Total health impacts from RC are driven by POC emissions across the United States. The number of deaths caused by each source-state is related to population, which influences both the extent of residential emissions and size of the exposed population, the need for home heating, and the degree to which wood, oil, and gas are used. As such, states causing the most deaths from RC have large populations within the state and immediately downwind and experience cold weather. In contrast, while downwind population plays a role for EGU-related premature mortalities, SO2 emissions patterns from EGUs differ greatly from POC emissions patterns from RC, and regional-scale atmospheric chemistry and transport plays a more significant role. States with the greatest EGU health impacts have the greatest coal-fired power plant emissions and atmospheric conditions amenable to secondary PM2.5 formation, specifically sulfate aerosol that is abundant in the eastern United States (Bell et al. 2007) during summer months. Our analyses of geographic patterns of health impacts reinforces the greater spatial extent of impact for secondarily formed pollutants from EGUs versus primarily emitted pollutants from RC.

Health Damage Function Analysis

Health damage functions do not follow the same patterns as total health impacts. Considering between-state differences, states with high health damage functions for primary PM2.5 emissions are similar for RC and EGUs, largely in the Northeast and Mid-Atlantic regions. The highest health damage functions for secondary PM2.5 precursors are in those same regions, with higher population states having higher health damage functions for RC than for EGUs. Western states, which tend to have lower populations with other low population states surrounding them, have the lowest health damage functions for primary PM2.5 precursors, but not secondary PM2.5 precursors, as they may be in areas that favor secondary particulate formation. O3-related health damage functions follow different patterns, with a tight association between values for EGUs and RC for both NOx and VOCs.


Despite this study’s use of a sophisticated air quality model and epidemiologically derived concentration–response functions to estimate total premature mortalities and health damage functions associated with RC and EGU emissions, there are a number of limitations, some of which are related to computational limitations. To determine sensitivity of ambient pollutant concentrations to precursor emissions from a source it is advantageous to model each source individually for an entire year. Due to computational constraints we chose not to model each state’s emissions individually and instead created CMAQ-DDM runs for sets of two and three states whose concentration surfaces would be sufficiently far from one another such that they could be separated and attributed to their source-state. Our separation algorithm deliberately omitted a small fraction of total premature deaths to ensure sufficient separation of concentration surfaces and attribution to the appropriate source-state. This omission was less than 10% for each run, providing a modest downward bias in total premature deaths, but potentially greater biases for individual states included in multi-state runs. Similarly, we had to limit modeling to 2 months—January and July—chosen to be representative of opposing meteorological and atmospheric conditions. Choosing only 2 months requires us to assume that each of January and July reasonably represents half of the year, and that the average of these 2 months reasonably represents annual patterns. This approach has been used in previous studies and has been shown to represent seasonal and annual conditions appropriately, and our modeling of baseline concentrations showed only modest differences in comparison with full annual runs (< 5% on a domain average basis for both PM2.5 and O3, represented in Figure S5), but will have greater uncertainty than annual runs in predicted concentrations.

Outlier health damage function values appear in states with very low emissions. For example, Idaho emits 0.02 tons per year of primary PM2.5 from EGUs, far less than other states. These small emissions lead to very low modeled health impacts (0.05 deaths) over the course of a year, so the influence on total premature mortalities across the United States is miniscule, but the premature mortalities per ton emitted are much higher than anticipated. There may be an issue with utilizing CMAQ-DDM in discerning sensitivity of ambient concentrations to these miniscule emissions values, which is only pointed out in assessing the health damage function as normalized by these small emissions. This indicates there may be a lower limit on emissions when applying CMAQ-DDM in this manner.

Calculation of total premature mortalities and health damage functions relies upon accurate population and baseline mortality values, which were obtained as county-wide values and spatially joined to CMAQ’s 36 km × 36 km grid cells assuming uniform population characteristics. As population density is not uniform across a county, this assumption may have led to misattributed premature mortalities and health damage functions in specific grid cells. Because of the large spatial domains over which health impacts occur, these uncertainties are likely modest, although sources in dense urban areas with relatively small downwind populations could exhibit greater errors, especially for primary pollutants where the spatial domain of impact is smaller. Concentration-response functions contain uncertainty not presented within our analysis, but all values would scale linearly and conclusions about variability would be unaffected.

A considerable strength of our modeling platform is that precursor-specific findings along with characterization of background concentrations could allow for sensitivity analyses on these assumptions in future analyses. Although our analysis includes a number of uncertainties including those from use of the National Emissions Inventory, meteorological fields used, and CMAQ atmospheric model, we have not constructed distributions around our output values or formally propagated uncertainty. This is in part because of the complexity in quantifying CMAQ-DDM uncertainty for individual sources, and because of our focus on relative comparisons within this manuscript, but remains a limitation in interpreting and applying our results.


In this study, we generated a novel set of estimates of both health impacts and health damage functions for RC and EGUs for the continental United States. We attribute premature deaths to emissions by source-state and precursor pollutant, which has not been done previously. These estimates can be used to address strategic emissions control policies on a state-by-state basis. Health damage functions can be used to determine which targeted emissions reductions will have the largest health benefits, an important part of creating efficient control strategies and designing SIPs that optimize health. Our use of CMAQ-DDM coupled with a complex image segmentation technique to isolate impacts of individual states can be extended to other source sectors, and source-based health damage functions can allow for understanding of how emissions impact health in a manner that can be helpful for state and federal policy makers.


Bell ML, Dominici F, Ebisu K, Zeger SL, Sarnet JM. 2007. Spatial and temporal variation in PM2.5 chemical composition in the United States for health effects studies. Environ Health Perspect 115(7):989–995.

Bell ML, Dominici F, Samet JM. 2005. A meta-analysis of time series studies of ozone and mortality with comparison to the National Morbidity, Mortality, and Air Pollution Study. Epidemiology 16(4):436–445.

Bell ML, McDermott A, Zeger SL, Samet JM, Dominici F. 2004. Ozone and short-term mortality in 95 US urban communities, 1987–2000. JAMA 292(19):2372–2378.

Bergin MS, Russell AG, Odman MT, Cohan DS, Chameides WL. 2008. Single-source impact analysis using three-dimensional air quality models. J Air Waste Manag Assoc 58(10):1351–1359.

Brook RD, Brook JR, Urch B, Vincent R, Rajagopalan S, Silverman F. 2002. Inhalation of fine particulate air pollution and ozone causes acute arterial vasoconstriction in healthy adults. Circulation 105(13):1534–1536.

Byun DW, Ching JKS, eds. 1999. Science Algorithms of the EPA Models-3 Community Multiscale Air Quality (CMAQ) Modeling System. EPA/600/R-99/030. Research Triangle Park, NC:US Environmental Protection Agency.

Byun DW, Schere KL. 2006. Review of the governing equations, computational algorithms, and other components of the Models-3 Community Multiscale Air Quality (CMAQ) modeling system. Appl Mech Rev 59(2):51–77, doi: 10.1115/1.2128636.

Caiazzo F, Ashok A, Waitz IA, Yim SHL, Barrett SRH. 2013. Air pollution and early deaths in the United States. Part I: quantifying the impact of major sectors in 2005, Atmos Environ 79:198–208.

CDC (Centers for Disease Control and Prevention). 2015. CDC WONDER. Underlying Cause of Death 1999–2015. Hyattsville, MD:National Center for Health Statistics.

Dedoussi IC, Barrett SRH. 2014. Air pollution and early deaths in the United States. Part II: attribution of PM2.5 exposure to emissions species, time, location and sector. Atmos Environ 99:610–617.

Dunker AM. 1984. The decoupled direct method for calculating sensitivity coefficients in chemical kinetics. J Chem Phys 81(5):2385, doi: 10.1063/1.447938.

Dunker AM, Yarwood G, Ortmann JP, Wilson GM. 2002. The decoupled direct method for sensitivity analysis in a three-dimensional air quality model – implementation, accuracy, and efficiency. Environ Sci Technol 36:2965–2976.

Fann N, Fulcher CM, Baker K. 2013. The recent and future health burden of air pollution apportioned across U.S. sectors. Environ Sci Technol 47(8):3580–3589.

Fann N, Fulcher CM, Hubbell BJ. 2009. The influence of location, source, and emission type in estimates of the human health benefits of reducing a ton of air pollution. Air Qual Atmos Health 2(3):169–176.

Fann N, Risley D. 2013. The public health context for PM2.5 and ozone air quality trends. Air Qual Atmos Health 6(1):1–11.

Hubbell B, Fann N, Levy JI. 2009. Methodological considerations in developing local-scale health impact assessments: balancing national, regional, and local data. Air Qual Atmos Health 2(2):99–110.

Itahashi S, Uno I, Kim S. 2012. Source contributions of sulfate aerosol over East Asia estimated by CMAQ-DDM. Environ Sci Technol 46(12):6733–6741.

Ito K, De Leon SF, Lippmann M. 2005. Associations between ozone and daily mortality: analysis and meta-analysis. Epidemiology 16(4):446–457.

Jerrett M, Burnett RT, Pope CA III, Ito K, Thurston G, Krewski D, et al. 2009. Long-term ozone exposure and mortality. N Engl J Med 360(11):1085–1095.

Ji M, Cohan DS, Bell ML. 2011. Meta-analysis of the association between short-term exposure to ambient ozone and respiratory hospital admissions. Environmental Research Letters Environ Res Lett 6(2):024006, doi: 10.1088/1748-9326/6/2/024006.

Koo BD, Dunker AM, Yarwood G. 2007. Implementing the decoupled direct method for sensitivity analysis in a particulate matter air quality model. Environ Sci Technol 41:2847–2854.

Krewski D, Jerrett M, Burnett RT, Ma R, Hughes E, Shi Y, et al. 2009. Extended follow-up and spatial analysis of the American Cancer Society study linking particulate air pollution and mortality. Res Rep Health Eff Inst 140:5–114.

Levy JI, Chemerynski SM, Sarnat JA. 2005. Ozone exposure and mortality. Epidemiology 16(4):458–468.

Levy JI, Woody M, Baek BH, Shankar U, Arunachalam S. 2012. Current and future particulate-matter-related mortality risks in the United States from aviation emissions during landing and takeoff. Risk Anal 32(2):237–249.

Mustafic H, Jabre P, Caussin C, Murad MH, Escolano S, Tafflet M, et al. 2012. Main air pollutants and myocardial infarction: a systematic review and meta-analysis. JAMA 307(7):713–721.

Napelenok SL, Cohan DS, Hu Y, Russell AG. 2006. Decoupled direct 3D sensitivity analysis for particulate matter (DDM-3D/PM). Atmos Environ 40(32):6112–6121.

Odman MT, Boylan JW, Wilkinson JG, Russell AG, Mueller SF, Imhoff RE, et al. 2002. SAMI Air Quality Modeling Final Report. Asheville, NC:Southern Appalachian Mountains Initiative.​ts/Reports/final_report.pdf [accessed 5 March 2015].

Roman HA, Walker KD, Walsh TL, Conner L, Richmond HM, Hubbell BJ, et al. 2008. Expert judgment assessment of the mortality impact of changes in ambient fine particulate matter in the U.S. Environ Sci Technol 42(7):2268–2274.

Schwartz J. 2005. How sensitive is the association between ozone and daily deaths to control for temperature? Am J Respir Crit Care Med 171(6):627–631.

Tagaris E, Liao KJ, Delucia AJ, Deck L, Amar P, Russell AG. 2009. Potential impact of climate change on air pollution-related human health effects. Environ Sci Technol 43(13):4979–4988.

Tagaris E, Liao KJ, DeLucia AJ, Deck L, Amar P, Russell AG. 2010. Sensitivity of air pollution-induced premature mortality to precursor emission under the influence of climate change. Int J Environ Res Public Health 7(5):2222–2237, doi: 10.3390/ijerph7052222.

U.S. Census Bureau. 2016. TIGER/Line® Shapefiles.​a/tiger-line.html [accessed 12 November 2016].

U.S. Environmental Protection Agency (U.S. EPA). 2015a. Carbon pollution emission guidelines for existing stationary sources: electric utility generating units.​es/2015-08/documents/cpp-final-rule.pdf [accessed 31 December 2016].

U.S. EPA. 2015b. Regulatory impact analysis (RIA) for residential wood heaters NSPS revision: Final report.​ood-heaters_ria_final-nsps-revision_2015​-02.pdf [accessed 15 August 2015].

U.S. EPA. 2016. Final cross state air pollution rule update.​ssstate-air-pollution-rule-update [accessed 31 December 2016].

Zanobetti A, Franklin M, Koutrakis P, Schwartz J. 2009. Fine particulate air pollution and its components in association with cause-specific emergency admissions. Environmental Health Environ Health 8:58, doi: 10.1186/1476-069X-8-58.

Association of Ambient Air Pollution with Depressive and Anxiety Symptoms in Older Adults: Results from the NSHAP Study

Author Affiliations open
Department of Health Sciences, Northeastern University, Boston, Massachusetts, USA

PDF icon PDF Version (186 KB)

  • Background: Ambient fine particulate matter (PM2.5) is among the most prevalent sources of environmentally induced inflammation and oxidative stress, both of which are implicated in the pathogenesis of most mental disorders. Evidence, however, concerning the impact of PM2.5 on mental health is just emerging.

    Objective: We examined the association between PM2.5 and current level of depressive and anxiety symptoms using a nationally representative probability sample (n = 4,008) of older, community-dwelling individuals living across the United States (the National Social Life, Health and Aging project).

    Methods: Mental health was evaluated using validated, standardized questionnaires and clinically relevant cases were identified using well-established cutoffs; daily PM2.5 estimates were obtained using spatiotemporal models. We used generalized linear mixed models, adjusting for potential confounders, and explored effect modification.

    Results: An increase in PM2.5 was significantly associated with anxiety symptoms, with the largest increase for 180-days moving average (OR = 1.61; 95% CI: 1.35, 1.92) after adjusting for socioeconomic measures (SES); PM2.5 was positively associated with depressive symptoms, and significantly for 30-day moving average (OR = 1.16; 95% CI: 1.05, 1.29) upon SES adjustment. The observed associations were enhanced among individuals who had low SES and history of comorbidity. When considering mental health as chronic conditions, PM2.5 was significantly associated with incident depressive symptoms for all exposure windows examined, but with incident anxiety symptoms only for shorter exposure windows, which may be due to a drop in power resulting from the decreased between-subject variability in chronic PM2.5 exposure.

    Conclusion: PM2.5 was associated with depressive and anxiety symptoms, with associations the strongest among individuals with lower SES or among those with certain health-related characteristics.

  • Citation: Pun VC, Manjourides J, Suh H. 2017. Association of ambient air pollution with depressive and anxiety symptoms in older adults: results from the NSHAP study. Environ Health Perspect 125:342–348;

    Address correspondence to H. Suh, Department of Civil and Environmental Engineering, Tufts University, Medford, MA 02153 USA. Telephone: (617) 627-2941. Email:

    We acknowledge M.-A. Kioumourtzoglou from Harvard University for her advice on regression modeling, and J. Yanosky from Pennsylvania State University for providing daily PM2.5 grid data.

    This work was supported by National Institute of Environmental Health Sciences, National Institutes of Health (NIH) (grant 1R01ES022657-01A1), with health and other covariate data obtained through NIH grants R01-AG021487, R37-AG030481, R01-AG033903, and R01-ES019168.

    The authors declare they have no actual or potential competing financial interests.

    Received: 7 December 2015
    Revised: 1 June 2016
    Accepted: 19 July 2016
    Published: 12 August 2016

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (637 KB)

    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact Our staff will work with you to assess and meet your accessibility needs within 3 working days.

    PDF icon Supplemental Table of Contents PDF (92 KB)


Mental health disorders accounted for over 140 million disability-adjusted life years worldwide in 2010 (Whiteford et al. 2013) and is the third most costly non-fatal condition in the United States, totaling $60 billion annually (Soni 2011). Adult mental disorder rates are substantial, with 18% experiencing anxiety disorder and 9.8% major depressive, dysthymic, and bipolar disorder in the past year (NIH/NIMH 2015). One of the hypothesized biological pathways is that mental disorders occur through oxidative stress and neuro-inflammation pathways (Ng et al. 2008; Vogelzangs et al. 2013). Compared to other organs, the brain is vulnerable to oxidative stress damage because of its high-energy use, low-endogenous scavenger levels, high-metabolic demands, and high-cellular lipid and protein content (Halliwell 2006; Mattson 2001). It is also susceptible to secondary and self-perpetuating damage from oxidative cellular injury via activated neuro-inflammatory responses or other pathways (Halliwell 2006; Ng et al. 2008). While genetic profiles, brain damage, substance abuse, socioeconomic status, and life situations have been important risk factors of mental disorders, recent evidence has indicated a role of physical environmental factors in the pathogenesis of mental disorders.

Airborne particulate matter (PM) pollution is a major source of environmentally induced inflammation and oxidative stress (Block and Calderón-Garcidueñas 2009). Ambient PM exposure has been consistently linked to adverse cardiovascular and respiratory effects, with oxidative stress and systemic inflammation considered the primary pathways through which air pollution damages health (Brook et al. 2004, 2010). While epidemiologic studies in the 1980s suggested associations between air pollution and mood (Bullinger 1989), depressive symptoms (Jacobs et al. 1984), and psychiatric emergencies (Rotton and Frey 1984), only recently have studies examined the possible effects of PM on mental illness and stress, with conflicting results (Lim et al. 2012; Marques and Lima 2011; Mehta et al. 2015; Power et al. 2015; Szyszkowicz 2007; Szyszkowicz et al. 2009; Wang et al. 2014). Toxicological studies, however, have shown neuropathological effects (e.g., increased levels of pro-inflammatory cytokines, degenerated dopaminergic neurons) and neurobehavioral responses (e.g., depression-like behaviors) upon PM exposure (Calderón-Garcidueñas et al. 2003; Campbell et al. 2005; Davis et al. 2013; Fonken et al. 2011; Veronesi et al. 2005). In this study, we used data from the National Social Life, Health and Aging Project (NSHAP) to examine the association of exposure to PM with aerodynamic diameter of ≤ 2.5 μm (PM2.5) with current depressive and anxiety symptom severity.



NSHAP is a longitudinal, nationally representative study of community-dwelling individuals (57–85 years) without known cognitive impairment living across the United States, with oversampling of African-Americans, Hispanics, men, and individuals between 75–84 years old (Shega et al. 2014; Waite et al. 2014a, 2014b). Numerous social, psychological, functional, and physiological health measures were collected for each participant in two data collection waves. Wave 1 was conducted from July 2005 to March 2006, with in-home interviews, biospecimen collection, and respondent completed questionnaires performed for 3,005 individuals. The same data were obtained in Wave 2 (August 2010 to May 2011) for 3,377 participants, including 2,261 Wave 1 respondents, 161 Wave 1 eligible, but non-interviewed respondents, and 955 spouses or cohabitating romantic partners (see Figure S1). Individuals from Wave 1 who did not participate in Wave 2 included those who were either deceased, moved away, or whose health (e.g., stroke) was too poor to participate in Wave 2. Participants and nonrespondents did not differ with regard to air pollution levels and cognitive scores. The overall weighted response rate was 75.5% and 76.9% for Waves 1 and 2, respectively (O’Muircheartaigh et al. 2009; Smith et al. 2009). The protocol was approved by the Institutional Review Boards of Northeastern University, the University of Chicago, and NORC at the University of Chicago. All participants provided written informed consent.

Mental Health Measures

Current level of depressive symptomatology was assessed using an 11-item form of the Center for Epidemiological Studies–Depression (CESD-11) Scale questionnaire (Kohout et al. 1993). The CESD-11 is a shorter version of the well-validated 20-item CESD (CESD-20) and is a self-reported screening tool that has been shown to capture the same dimensions as CESD-20 with similar precision. Participants were asked to indicate their response to 11 statements (see Table S1). Each statement asked participants to rate the frequency of their feelings during the previous week as rarely or none of the time (0), some of the time (1), occasionally (2), and most of the time (3), corresponding to a 4-point Likert scale. Positively phrased statements were reverse coded before summation (range: 0 to 33), with higher summed scores indicating more severe depressive symptoms. The Cronbach’s alpha for internal consistency was 0.80 for the entire NSHAP sample. A score of ≥ 9 on the CESD-11 was used to identify individuals with moderate-to-severe depressive symptoms based on previous studies (Kohout et al. 1993; Torres 2012).

The Hospital Anxiety and Depression Scale (HADS) has been used successfully as a self-rating instrument to measure current state of anxiety, with well-established reliability and validity in population-based studies (Mykletun et al. 2001). NSHAP participants were asked to complete a 7-item anxiety subscale of HADS (HADS-A) to indicate the frequency of feelings of anxious mood, thoughts, and restlessness over the past week on a 4-point Likert scale (see Table S1) One positively phrased statement was reverse coded. Individual statement scores were then summed to obtain the total HADS-A score (range: 0 to 21), with higher scores indicating increasing levels of anxiety. The Cronbach’s alpha of the HADS-A was 0.76. A HADS-A cutoff score of 8 gives the optimal sensitivity and specificity (approximately 0.80) to categorize individuals as having an anxiety disorder or not (Bjelland et al. 2002). Thus, we defined participants with a cutoff score of ≥ 8 as cases with moderate-to-severe anxiety symptoms.

Exposure Assessment

Daily PM2.5 estimates on a 6-km grid covering the conterminous United States were obtained from a set of five spatio-temporal generalized additive mixed models (GAMMs) of daily PM2.5 mass levels in the conterminous United States, fit separately to 1999–2001, 2002–2004, 2005–2007, 2008–2009, and 2010–2011. These models were based on the spatiotemporal GAMM of monthly PM2.5 mass from 1999–2007 documented in Yanosky et al. (2014). PM2.5 data were obtained primarily from the U.S. Environmental Protection Agency (EPA) Air Quality System database and Interagency Monitoring of Protected Visual Environments (IMPROVE) network (IMPROVE 2013; U.S. EPA 2016). The model included three meteorological covariates (i.e., wind speed, temperature, and total precipitation) that influence pollutant dispersion as well as several geospatial covariates (i.e., smoothed county population density from the 2000 U.S. Census, point-source PM2.5 emissions density within 7.5 km, proportion of urban land use within 1 km, elevation, and annual average PM2.5 for 2002 from the U.S. EPA’s Community Multiscale Air Quality model). Finally, the daily PM2.5 model includes traffic-related PM levels, represented as the output of a Gaussian line-source dispersion modeling approach. The line-source model uses ADMS-Roads software (version 4.0; Cambridge Environmental Research Consultants Ltd.) and associated spatially smoothed traffic intensity and daily meteorological inputs to describe small-scale spatial gradients in primary PM concentrations near roadways. The daily PM2.5 model has undergone validation during development using cross-validation techniques (see Yanosky et al. 2014), and had a cross-validation R2 of 0.76. NSHAP participants were matched to the grid (n = 894 in total) closest to their residential addresses. Two participants were excluded from the study as their residential addresses were outside the conterminous United States. The mean distance between each grid centroid–residential address pair was 2.23 km, with a range of 0.05–4.21 km.

Statistical Analysis

Given the longitudinal study design and multiple participants per household, we used the generalized linear mixed models PROC GLIMMIX procedure (version 9.3; SAS Institute Inc.) to study the association of PM2.5 and each mental health condition, modeled as binary outcome based on a CESD-11 score ≥ 9 and HADS-A ≥ 8 for moderate-to-severe depressive and anxiety symptoms, respectively, and to account for random effects of repeated measurements for participants and households. We fit penalized spline models to evaluate deviations from linearity, with the linear model preferable for each outcome based on Akaike information criterion. We examined associations for PM2.5 exposure windows averaged from previous 7 days, to up to 4 years prior to the interview date of NSHAP participants to study the impact of semi-acute and chronic PM2.5 exposure for mental disorders, respectively.

In the basic models, we adjusted for age, sex, race, year, season and day of week of questionnaire completion, region of residence (West, Midwest, South, Northeast), and whether participants lived within a metropolitan statistical area (MSA). Multivariable models were also constructed to control for confounding by socioeconomic measures (SES) as assessed using individual-specific education attainment and family income, and census-level median household income and percent of population with income below poverty level. To further evaluate potential confounding, additional wave-specific covariates were selected a priori based on their previous associations with mental illness or air pollution: individual-specific obesity status [i.e., body mass index (BMI) ≥ 30)], current smoking status, physical activity, alcohol consumption (drinks per day), UCLA Loneliness scale (range: 0–9), current use of antidepressant medication, and history of diabetes, hypertension, stroke, heart failure, emphysema, chronic obstructive pulmonary disease (COPD) or asthma (see Table S2). Two covariates (i.e., BMI and family income) had 10% and 29% missing data, respectively; their missing values were imputed by simple mean substitution. Missing data of other covariates (< 5%) were not imputed. Both base and SES-adjusted analyses were restricted to a subset of data for which values for all covariates were not missing [i.e., 6,199 nonmissing out of 6,382 total observations (97.1%) for covariates]. Additional covariates were added individually in separate basic models to avoid multicollinearity and reduce potential bias on the estimates if covariates were not shown to be confounders (Xing and Xing 2010). Since certain covariates (e.g., sex) could be possible effect modifiers, their modification of PM2.5-mental health findings was examined through interaction terms, using the PROC GLIMMIX procedure, which provides added options to compute customized odds ratios and the corresponding confidence intervals (CIs) automatically for each level of the interaction term.

We conducted several sensitivity analyses. First, we considered mental health measures as continuous rather than binary measures. Second, we restricted the longitudinal analysis to individuals who participated in both waves, to those living in MSAs only, those who did not move between waves or did not currently take antidepressant medication, respectively. We also reanalyzed the models using multiple imputation technique. Third, we constructed the model using PM2.5 concentrations measured at the nearest U.S. EPA ambient monitors within 60 km of the residential address. Lastly, we considered our depression and anxiety outcomes to be chronic relapsing disorders, by restricting our analyses to Wave 2 participants who did not have moderate-to-severe depressive (CESD-11 < 9) or anxiety (HADS-A < 8) symptoms in Wave 1. In doing so, we acknowledge that if mental disorders are chronic conditions, PM2.5 exposures for Wave 2 could not be associated with mental disorders that occurred at Wave 1 or earlier. If that is the case, inclusion of individuals reporting mental disorders in Wave 1 in longitudinal analyses would bias the effect estimates towards the null. Since information on the history of mental illness was not available in the study, we conducted logistic regression analysis examining the association between PM2.5 exposure and incident moderate-to-severe depressive and anxiety symptoms in Wave 2. Results are expressed as the odds ratio (OR) per 5 μg/m3 increment in PM2.5 exposure; all effect estimates and their corresponding confidence intervals were obtained through the ODDSRATIO (DIFF = ALL) option in the GLIMMIX procedure.


A total of 4,008 community-dwelling participants were available for analysis. Overall, participants were on average 69 and 72 years old in Wave 1 and 2 respectively, and nearly half were men (Table 1). Most participants were white, exercised ≥ 1 time per week, and had a high school education or greater. Approximately three-fifths of the participants reported a history of high blood pressure or hypertension; one-fifth diabetes, one-sixth emphysema, COPD, or asthma; and 10% or less stroke or heart failure, respectively. Participants reported slightly higher current use of antidepressant medications and lower UCLA Loneliness score in Wave 2 than in Wave 1. The prevalence of current moderate-to-severe depressive symptoms decreased from 24% in Wave 1 to 21% in Wave 2, while that of moderate-to-severe anxiety symptoms increased in Wave 2 (14%) compared with Wave 1 (21%). Four (< 1%) and 744 participants (12%) did not complete the depression or anxiety assessments, respectively, with missingness not related to air pollution exposures. Intra-wave correlation for CESD-11 score was 0.55, and that for HADS-A score was 0.37. The mean annual concentration (± SD) of PM2.5 was 11.1 (± 3.0) μg/m3 and 8.8 (± 2.2) μg/m3 in Wave 1 and 2, respectively (Table 1; see also Table S3). Refer to Table S4 for descriptive characteristics stratified by data collection wave and pollution category.

Table 1. Select View Table (HTML Version) for a 508-conformant version

Table 1. Characteristics of NSHAP study participants by wave.

View Table (HTML Version)
View larger image (TIF File)

The associations of ambient PM2.5 in the previous 7, 30, 180, and 365 days and 4 years prior with each measure of mental health are presented in Table 2. In basic models, a 5 μg/m3 increase in PM2.5 was significantly and positively associated with moderate-to-severe anxiety symptoms for all exposure windows, with the largest increase in odds for 180-days PM2.5 exposure (OR = 1.55; 95% CI: 1.31, 1.85). On the other hand, exposure to PM2.5 averaged over the previous 7 days and 30 days was significantly associated with 1.09 (95% CI: 1.01, 1.17) and 1.20 times (95% CI: 1.08, 1.33) the odds of moderate-to-severe depressive symptoms, respectively. Elevations in odds, though statistically insignificant, were also seen for longer moving averages. Analysis of an extended range of exposure windows shows that the effect estimates of depressive and anxiety symptoms increase gradually and are the largest at 60-days and 180-days PM2.5 exposure, respectively (see Figure S2). Pattern of associations from multivariable models, which further adjusted for SES, were generally consistent to those from basic models (Table 2). Findings were similar in sensitivity analyses a) considering mental health measures as linear continuous variables, b) controlling for additional covariates, c) using an alternative imputation technique, d) using different PM2.5 exposure measures from nearby ambient monitors, and e) restricting to individuals who participated in both waves, lived in MSAs, did not move between waves, or did not currently take antidepressant medication (see Tables S5–S9).

Table 2. Select View Table (HTML Version) for a 508-conformant version

Table 2. ORs (95% CIs) for mental illness per 5 μg/m3 increment in PM2.5 levels over various moving averages.

View Table (HTML Version)
View larger image (TIF File)

Table 3 (see also Table S10) shows evidence of effect modification for the relationship between mental illness and average 30-day PM2.5 level, the exposure window that shows generally significant associations. Individuals who had less than a high school education were at significantly higher odds of PM2.5-associated moderate-to-severe anxiety symptoms (p-interact < 0.001), and suggestive higher odds of moderate-to-severe depressive symptoms. The association of PM2.5 and depressive symptoms was also greater for individuals with low census-level SES (i.e., high percentage of population with income below poverty level) or for those with a history of stroke or respiratory illnesses. Participants who had a history of stroke or heart failure also showed increased odds of moderate-to-severe anxiety symptoms associated with PM2.5, compared to those who had no such history.

Table 3. Select View Table (HTML Version) for a 508-conformant version

Table 3. Effect modification analysis of the association of mental illness with 5 μg/m3 increment in PM2.5 levels over preceding 30-days moving average in multivariable models with interaction terms for the potential modifier.

View Table (HTML Version)
View larger image (TIF File)

When anxiety and depression were considered as chronic relapsing disorders using logistic regression (Table 4), increments of PM2.5 levels in all exposure windows were positively and statistically significantly associated with incident moderate-to-severe depressive symptoms in Wave 2, corresponding to 1.35–1.68 times the odds in multivariable models. Increase in PM2.5 levels averaged over the previous 7-days was also significantly associated with incident moderate-to-severe anxiety symptoms (OR = 1.36; 95% CI: 1.09, 1.68). The increase and statistical significance in odds of incident moderate-to-severe anxiety symptoms gradually reduced with longer PM2.5 exposure windows.

Table 4. Select View Table (HTML Version) for a 508-conformant version

Table 4. Logistic regression analysis [ORs (95% CIs)] of the association between mental disorders and 5 μg/m3 increment in PM2.5 levels over various moving averages—restricting to WAVE 2 participants who did not have moderate-to-severe depressive (CESD-11 < 9) or anxiety (HADS-A < 8) symptoms in Wave 1.

View Table (HTML Version)
View larger image (TIF File)


In our nationally representative sample of U.S. older adults, we observed statistically significantly positive associations with moderate-to-severe anxiety symptoms for all PM2.5 exposure windows (e.g., OR = 1.55; 95% CI: 1.31, 1.85 for PM2.5 averaged over 180 days). We also found increased odds of moderate-to-severe depressive symptoms associated with a 5 μg/m3 increment in PM2.5 exposure, with the largest increase associated with PM2.5 averaged over 30 days (OR = 1.20; 95% CI: 1.08, 1.33). Patterns of associations remain in multivariable models adjusting for SES. The observed associations were enhanced among individuals who were of low SES or had history of certain health-related conditions.

We found that reported depressive and anxiety symptoms at Wave 1 were only weakly correlated with corresponding symptoms at Wave 2. This supports our assumption that each mental health condition is reversible (Bedrosian et al. 2013; NIH/NIMH 2017), and is consistent with the CES-D and HADS-A questionnaires that evaluate current rather than chronic depression and anxiety symptoms. However, when depression and anxiety were considered as chronic conditions using logistic regression, PM2.5 was significantly associated with incident moderate-to-severe depressive symptoms in Wave 2 for all exposure windows examined, with higher effect estimates as compared to when depressive symptoms were assumed to be reversible disorders. In contrast, PM2.5 exposures were significantly associated with incident moderate-to-severe anxiety symptoms only for shorter exposure windows. These findings, which should be interpreted with caution, suggest that shorter-term PM2.5 exposures might be more biologically relevant to incident anxiety symptoms. However, associations of chronic PM2.5 exposure with anxiety should not be ruled out, as the decreased between-subject variability in long-term PM2.5 exposures leads to wider confidence intervals.

This study provides among the first evidence of positive associations between ambient PM2.5 exposures and adverse mental health symptoms, and is the first study to report increased odds of moderate-to-severe depressive symptoms associated with PM2.5 exposure. To date, only three epidemiologic studies have examined the association of long-term exposure to PM with mental health risk (Mehta et al. 2015; Power et al. 2015; Wang et al. 2014). Our observed positive and significant association between PM2.5 and moderate-to-severe anxiety symptoms is consistent with findings from Power et al. (2015) who also reported positive association of PM2.5 with phobic anxiety among a cohort of U.S. nurses, thus lending support for our findings. Yet, our findings of increased odds of depressive symptoms associated with PM2.5 differs from those of a Boston study that found a significant negative association (Wang et al. 2014). The conflicting findings may be attributed to our study’s larger sample size, use of participant-specific exposure measures, greater geographical coverage, and longer exposure windows examined. Other studies have reported positive and statistically significant association between short-term PM exposure (e.g., 1–3 lagged day) and suicidal risk (Bakian et al. 2015; Kim et al. 2010; Szyszkowicz et al. 2010), which may lend support to our findings of increased incident anxiety symptoms with semi-acute PM2.5 exposure windows.

While the biological pathways through which PM2.5 exposures influence mental health remain unknown, PM2.5 exposures may harm mental health through increased neuroinflammation, oxidative stress, cerebrovascular damage and neurodegeneration (Block and Calderón-Garcidueñas 2009; MohanKumar et al. 2008), as evidenced by findings from animal studies that show associations between PM and elevated hippocampal pro-inflammatory cytokine expression (Campbell et al. 2005; Fonken et al. 2011), upregulated expression of innate immunity and oxidative stress pathways (Sama et al. 2007), robust inflammatory and stress protein brain responses (Calderón-Garcidueñas et al. 2003), neuropathological damage in the brains of Apo E-deficient mice (Veronesi et al. 2005), and depression-like responses in mice (Fonken et al. 2011). PM2.5 pollution may also harm mental health by increasing markers of glucocorticoid activity and levels of the stress hormone cortisol (Thomson et al. 2013; Tomei et al. 2003) or through aggravating major respiratory or cardiac medical conditions (Power et al. 2015; Wang et al. 2014). Cardiopulmonary diseases positively associated with PM, such as asthma and heart failure, are also associated with increased prevalence of depression and anxiety disorders (Aben et al. 2003; Aström 1996; Cully et al. 2009; Dossa et al. 2011; Maurer et al. 2008; Scott et al. 2007), possibly mediated through biological (e.g., chronic inflammation) and behavioral (e.g., fear, social isolation) mechanisms (Hsu et al. 2014; Loubinoux et al. 2012; Yohannes and Alexopoulos 2014). Our findings of effect modification of the PM-mental health associations by individuals who had stroke, heart failure, or hypertension provide support for the importance of PM-mediated aggravation of cardiopulmonary conditions and our findings of PM2.5-mediated impacts on adverse mental health symptoms. In addition, our evidence of effect modification by SES suggests that PM exposure may have stronger impacts on depression symptoms among individuals with lower SES. We found that more participants living in neighborhoods with a greater percentage of the population with income below poverty level also had higher ambient PM2.5 pollution level near their residences; previous studies reported greater psychological stress and adverse mental health among people living in census tracts with lower SES and higher unemployment and poverty proportions (Bell and Ebisu 2012; Schwartz et al. 2011). Thus, a combination of greater pollution exposure and susceptibility may best explain how SES modified the association between PM exposure and depression symptoms.

Our study has several limitations. First, CESD and HADS-A are not clinical diagnostic instruments, nor are they designed to assess chronic mental disorder. Also, dichotomizing the continuous scores will likely reduce statistical power (Greenland 1995). However, these questionnaires are widely used screening tools for current level of depressive and anxiety symptom severity in the somatic, psychiatric and general population settings (Bjelland et al. 2002; Radloff 1977); and they provide cutoff scores for probable cases that are of clinical relevance, with high sensitivity, specificity and internal consistency (Dozeman et al. 2011). Second, residual confounding or confounding by unmeasured covariates and/or pollution (e.g., traffic noise pollution) is possible. Nonetheless, adjustment for several known confounding variables, including those related to SES and behaviors, did not eliminate the observed most significant associations and positive trends of PM2.5 with our mental health measures. Third, we assessed PM2.5 exposures using individual-specific exposures based on the nearest grid point values to the residential addresses, with the average distance of 2.23 km. While more precise than nearest monitor values, they do not account for time spent indoors or personal behaviors and thus are imperfect proxies of personal PM2.5 exposures and thus contribute to exposure misclassification. Last, findings from the current study may not be generalizable to younger age groups.

The nationally representative sample of older, community-dwelling Americans was a major strength of our study, since previous research of PM and mental health used convenience samples. We evaluated two affective measures to provide a comprehensive picture of air pollution’s impact on mental health, rather than one mental health measure as in existing studies. Our study was well-powered to detect meaningful associations and adjusted for confounding from individual- and census-level SES measures. Moreover, we showed effect modification of the PM2.5-mental health associations by participant characteristics, providing insight into susceptibility. We also considered multiple PM2.5 exposure windows; consistent with a previous study (Power et al. 2015), we found that intermediate-term PM2.5 exposure (e.g., 30 to 180 days) may be the most biological relevant exposure period to adverse mental symptoms, compared to longer exposure windows. Lastly, our findings were robust to multiple sensitivity analyses.


We reported evidence of positive association between PM2.5 and moderate-to-severe depressive and anxiety symptoms among a representative sample of U.S. older adults. Our findings suggest that people with low SES or with a history of underlying health conditions may be more susceptible to increased odds of mental disorders after PM exposure.


Aben I, Verhey F, Strik J, Lousberg R, Lodder J, Honig A. 2003. A comparative study into the one year cumulative incidence of depression after stroke and myocardial infarction. J Neurol Neurosurg Psychiatry 74:581–585.

Aström M. 1996. Generalized anxiety disorder in stroke patients: a 3-year longitudinal study. Stroke 27:270–275, doi: 10.1161/01.STR.27.2.270.

Bakian AV, Huber RS, Coon H, Gray D, Wilson P, McMahon WM, et al. 2015. Acute air pollution exposure and risk of suicide completion. Am J Epidemiol 181:295–303, doi: 10.1093/aje/kwu341.

Bedrosian TA, Weil ZM, Nelson RJ. 2013. Chronic dim light at night provokes reversible depression-like phenotype: possible role for TNF. Mol Psychiatry 18:930–936, doi: 10.1038/mp.2012.96.

Bell ML, Ebisu K. 2012. Environmental inequality in exposures to airborne particulate matter components in the United States. Environ Health Perspect 120:1699–1704, doi: 10.1289/ehp.1205201.

Bjelland I, Dahl AA, Haug TT, Neckelmann D. 2002. The validity of the Hospital Anxiety and Depression Scale: an updated literature review. J Psychosom Res 52:69–77, doi: 10.1016/S0022-3999(01)00296-3.

Block ML, Calderón-Garcidueñas L. 2009. Air pollution: mechanisms of neuroinflammation and CNS disease. Trends Neurosci 32:506–516, doi: 10.1016/j.tins.2009.05.009.

Brook RD, Franklin B, Cascio W, Hong Y, Howard G, Lipsett M, et al. 2004. Air pollution and cardiovascular disease: a statement for healthcare professionals from the Expert Panel on Population and Prevention Science of the American Heart Association. Circulation 109:2655–2671, doi: 10.1161/01.CIR.0000128587.30041.C8.

Brook RD, Rajagopalan S, Pope CA III, Brook JR, Bhatnagar A, Diez-Roux AV, et al. 2010. Particulate matter air pollution and cardiovascular disease: an update to the scientific statement from the American Heart Association. Circulation 121:2331–2378, doi: 10.1161/CIR.0b013e3181dbece1.

Bullinger M. 1989. Psychological effects of air pollution on healthy residents—a time-series approach. J Environ Psychol 9:103–118.

Calderón-Garcidueñas L, Maronpot RR, Torres-Jardon R, Henríquez-Roldán C, Schoonhoven R, Acuña-Ayala H, et al. 2003. DNA damage in nasal and brain tissues of canines exposed to air pollutants is associated with evidence of chronic brain inflammation and neurodegeneration. Toxicol Pathol 31:524–538, doi: 10.1080/01926230390226645.

Campbell A, Oldham M, Becaria A, Bondy SC, Meacher D, Sioutas C, et al. 2005. Particulate matter in polluted air may increase biomarkers of inflammation in mouse brain. Neurotoxicology 26:133–140, doi: 10.1016/j.neuro.2004.08.003.

Cully JA, Jimenez DE, Ledoux TA, Deswal A. 2009. Recognition and treatment of depression and anxiety symptoms in heart failure. Prim Care Companion J Clin Psychiatry 11:103–109.

Davis DA, Bortolato M, Godar SC, Sander TK, Iwata N, Pakbin P, et al. 2013. Prenatal exposure to urban air nanoparticles in mice causes altered neuronal differentiation and depression-like responses. PLoS One 8:e64128, doi: 10.1371/journal.pone.0064128.

Dossa A, Glickman ME, Berlowitz D. 2011. Association between mental health conditions and rehospitalization, mortality, and functional outcomes in patients with stroke following inpatient rehabilitation. BMC Health Serv Res 11:311, doi: 10.1186/1472-6963-11-311.

Dozeman E, van Schaik DJF, van Marwijk HWJ, Stek ML, van Der Horst HE, Beekman ATF. 2011. The center for epidemiological studies depression scale (CES-D) is an adequate screening instrument for depressive and anxiety disorders in a very old population living in residential homes. Int J Geriatr Psychiatry 26:239–246, doi: 10.1002/gps.2519.

Fonken LK, Xu X, Weil ZM, Chen G, Sun Q, Rajagopalan S, et al. 2011. Air pollution impairs cognition, provokes depressive-like behaviors and alters hippocampal cytokine expression and morphology. Mol Psychiatry 16:987–995, doi: 10.1038/mp.2011.76.

Greenland S. 1995. Avoiding power loss associated with categorization and ordinal scores in dose-response and trend analysis. Epidemiology 6:450–454.

Halliwell B. 2006. Oxidative stress and neurodegeneration: where are we now? J Neurochem 97:1634–1658, doi: 10.1111/j.1471-4159.2006.03907.x.

Hsu NW, Tsao HM, Chen HC, Chou P. 2014. Anxiety and depression mediate the health-related quality of life differently in patients with cardiovascular disease and stroke-preliminary report of the Yilan study: a population-based community health survey. PLoS One 9:e107609, doi: 10.1371/journal.pone.0107609.

IMPROVE (Interagency Monitoring of Protected Visual Environments). 2013. Interagency Monitoring of Protected Visual Environments Homepage. [accessed 8 January 2015].

Jacobs SV, Evans GW, Catalano R, Dooley D. 1984. Air pollution and depressive symptomatology: exploratory analyses of intervening psychosocial factors. Popul Environ 7:260–272.

Kim C, Jung SH, Kang DR, Kim HC, Moon KT, Hur NW, et al. 2010. Ambient particulate matter as a risk factor for suicide. Am J Psychiatry 167:1100–1107, doi: 10.1176/appi.ajp.2010.09050706.

Kohout FJ, Berkman LF, Evans DA, Cornoni-Huntley J. 1993. Two shorter forms of the CES-D (Center for Epidemiological Studies Depression) depression symptoms index. J Aging Health 5:179–193.

Lim YH, Kim H, Kim JH, Bae S, Park HY, Hong YC. 2012. Air pollution and symptoms of depression in elderly adults. Environ Health Perspect 120:1023–1028, doi: 10.1289/ehp.1104100.

Loubinoux I, Kronenberg G, Endres M, Schumann-Bard P, Freret T, Filipkowski RK, et al. 2012. Post-stroke depression: mechanisms, translation and therapy. J Cell Mol Med 16:1961–1969, doi: 10.1111/j.1582-4934.2012.01555.x.

Marques S, Lima ML. 2011. Living in grey areas: industrial activity and psychological health. J Environ Psychol 31:314–322, doi: 10.1016/j.jenvp.2010.12.002.

Mattson MP. 2001. Mechanisms of neuronal apoptosis and excitotoxicity. In: Pathogenesis of Neurodegenerative Disorders. Totowa, NJ:Humana Press, 1–20.

Maurer J, Rebbapragada V, Borson S, Goldstein R, Kunik ME, Yohannes AM, et al. 2008. Anxiety and depression in COPD. Chest 134(4 suppl):43S–56S, doi: 10.1378/chest.08-0342.

Mehta AJ, Kubzansky LD, Coull BA, Kloog I, Koutrakis P, Sparrow D, et al. 2015. Associations between air pollution and perceived stress: the Veterans Administration Normative Aging Study. Environ Health 14:1–10, doi: 10.1186/1476-069X-14-10.

MohanKumar SMJ, Campbell A, Block M, Veronesi B. 2008. Particulate matter, oxidative stress and neurotoxicity. Neurotoxicology 29:479–488, doi: 10.1016/j.neuro.2007.12.004.

Mykletun A, Stordal E, Dahl AA. 2001. Hospital Anxiety and Depression (HAD) scale: factor structure, item analyses and internal consistency in a large population. Br J Psychiatry 179:540–544, doi: 10.1192/bjp.179.6.540.

NIH/NIMH (National Institutes of Health, National Institute of Mental Health). 2015. Any Mood Disorder Among Adults.​s/prevalence/any-mood-disorder-among-adu​lts.shtml [accessed 4 July 2015].

NIH/NIMH. 2017. Depression Treatment.​epression/index.shtml [accessed 23 January 2017].

Ng F, Berk M, Dean O, Bush AI. 2008. Oxidative stress in psychiatric disorders: evidence base and therapeutic implications. Int J Neuropsychopharmacol 11:851–876, doi: 10.1017/S1461145707008401.

O’Muircheartaigh C, Eckman S, Smith S. 2009. Statistical design and estimation for the National Social Life, Health, and Aging Project. J Gerontol B Psychol Sci Soc Sci 64(suppl 1):i12–i19, doi: 10.1093/geronb/gbp045.

Power MC, Kioumourtzoglou MA, Hart JE, Okereke OI, Laden F, Weisskopf MG. 2015. The relation between past exposure to fine particulate air pollution and prevalent anxiety: observational cohort study. BMJ 350:h1111, doi: 10.1136/bmj.h1111.

Radloff LS. 1977. CES-D scale: a self-report depression scale for research in the general population. Appl Psychol Meas 1:385–401.

Rotton J, Frey J. 1984. Psychological costs of air pollution: atmospheric conditions, seasonal trends, and psychiatric emergencies. Popul Environ 7:3–16.

Sama P, Long TC, Hester S, Tajuba J, Parker J, Chen LC, et al. 2007. The cellular and genomic response of an immortalized microglia cell line (BV2) to concentrated ambient particulate matter. Inhal Toxicol 19:1079–1087, doi: 10.1080/08958370701628721.

Schwartz J, Bellinger D, Glass T. 2011. Exploring potential sources of differential vulnerability and susceptibility in risk from environmental hazards to expand the scope of risk assessment. Am J Public Health 101(suppl 1):S94–S101, doi: 10.2105/AJPH.2011.300272.

Scott KM, Von Korff M, Ormel J, Zhang M, Bruffaerts R, Alonso J, et al. 2007. Mental disorders among adults with asthma: results from the World Mental Health Survey. Gen Hosp Psychiatry 29:123–133, doi: 10.1016/j.genhosppsych.2006.12.006.

Shega JW, Sunkara PD, Kotwal A, Kern DW, Henning SL, McClintock MK, et al. 2014. Measuring cognition: the Chicago Cognitive Function Measure in the National Social Life, Health and Aging Project, Wave 2. J Gerontol B Psychol Sci Soc Sci 69(suppl 2):S166–S176, doi: 10.1093/geronb/gbu106.

Smith S, Jaszczak A, Graber J, Lundeen K, Leitsch S, Wargo E, et al. 2009. Instrument development, study design implementation, and survey conduct for the National Social Life, Health, and Aging Project. J Gerontol B Psychol Sci Soc Sci 64(suppl 1):i20–i29, doi: 10.1093/geronb/gbn013.

Soni A. 2011. Top 10 Most Costly Conditions among Men and Women, 2008: Estimates for the U.S. Civilian Noninstitutionalized Adult Population, Age 18 and Older. Statistical Brief #331. Rockville, MD:Agency for Healthcare Research and Quality,​tions/st331/stat331.shtml [accessed 23 January 2017].

Szyszkowicz M. 2007. Air pollution and emergency department visits for depression in Edmonton, Canada. Int J Occup Med Environ Health 20:241–245, doi: 10.2478/v10001-007-0024-2.

Szyszkowicz M, Rowe BH, Colman I. 2009. Air pollution and daily emergency department visits for depression. Int J Occup Med Environ Health 22:355–362, doi: 10.2478/v10001-009-0031-6.

Szyszkowicz M, Willey JB, Grafstein E, Rowe BH, Colman I. 2010. Air pollution and emergency department visits for suicide attempts in Vancouver, Canada. Environ Health Insights 4:79–86, doi: 10.4137/EHI.S5662.

Thomson EM, Vladisavljevic D, Mohottalage S, Kumarathasan P, Vincent R. 2013. Mapping acute systemic effects of inhaled particulate matter and ozone: multiorgan gene expression and glucocorticoid activity. Toxicol Sci 135:169–181, doi: 10.1093/toxsci/kft137.

Tomei F, Rosati MV, Ciarrocca M, Baccolo TP, Gaballo M, Caciari T, et al. 2003. Plasma cortisol levels and workers exposed to urban pollutants. Ind Health 41:320–326.

Torres E. 2012. Psychometric properties of the Center for Epidemiologic Studies Depression Scale in African-American and Black Caribbean Adults. Issues Ment Health Nurs 33:687–696, doi: 10.3109/01612840.2012.697534.

U.S. Census Bureau. 2000. Census 2000 CD. Washington, DC.: U.S. Government Printing Office.

U.S. EPA (Environmental Protection Agency). 2016. U.S. EPA Air Quality System. [accessed 23 January 2017].

Veronesi B, Makwana O, Pooler M, Chen LC. 2005. Effects of subchronic exposures to concentrated ambient particles. VII. Degeneration of dopaminergic neurons in Apo E–/– mice. Inhal Toxicol 17:235–241, doi: 10.1080/08958370590912888.

Vogelzangs N, Beekman ATF, de Jonge P, Penninx BWJH. 2013. Anxiety disorders and inflammation in a large adult cohort. Transl Psychiatry 3:e249, doi: 10.1038/tp.2013.27.

Waite LJ, Cagney K, Dale W, Huang E, Laumann EO, McClintock M, et al. 2014a. National Social Life, Health, and Aging Project (NSHAP): Wave 2 and Partner Data Collection. ICPSR34921-v1. Ann Arbor, MI:Inter-university Consortium for Political and Social Research.​A/studies/34921 [accessed 23 January 2017].

Waite LJ, Laumann EO, Levinson W, Lindau ST, O’Muircheartaigh CA. 2014b. National Social Life, Health, and Aging Project (NSHAP): Wave 1. ICPSR20541-v6. Ann Arbor, MI:Inter-university Consortium for Political and Social Research.​A/studies/20541 [accessed 23 January 2017].

Wang Y, Eliot MN, Koutrakis P, Gryparis A, Schwartz JD, Coull BA, et al. 2014. Ambient air pollution and depressive symptoms in older adults: results from the MOBILIZE Boston Study. Environ Health Perspect 122:553–558, doi: 10.1289/ehp.1205909.

Whiteford HA, Degenhardt L, Rehm J, Baxter AJ, Ferrari AJ, Erskine HE, et al. 2013. Global burden of disease attributable to mental and substance use disorders: findings from the Global Burden of Disease Study 2010. Lancet 382:1575–1586.

Xing G, Xing C. 2010. Adjusting for covariates in logistic regression models. Genet Epidemiol 34:769–772, doi: 10.1002/gepi.20526.

Yanosky JD, Paciorek CJ, Laden F, Hart JE, Puett RC, Liao D, et al. 2014. Spatio-temporal modeling of particulate air pollution in the conterminous United States using geographic and meteorological predictors. Environ Health 13:63, doi: 10.1186/1476-069X-13-63.

Yohannes AM, Alexopoulos GS. 2014. Depression and anxiety in patients with COPD. Eur Respir Rev 23:345–349, doi: 10.1183/09059180.00007813.