Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

What Can Epidemiological Studies Tell Us about the Impact of Chemical Mixtures on Human Health?

What Can Epidemiological Studies Tell Us about the Impact of Chemical Mixtures on Human Health? Perspectives Brief Communication A Section 508– conformant HTML version of this ar ticle is available at http://dx.doi.org/10.1289/ehp.1510569. What Can Epidemiological Studies Tell Us about the Impact of Chemical Mixtures on Human Health? http://dx.doi.org/10.1289/ehp.1510569 study (Savitz 2014), or the inability for the biomarker to r epresent Summary : Humans are exposed to a large number of environmental exposure during the etiologically relevant time period. chemicals: Some of these may be toxic, and many others have unknown Second, epidemiologists who study mixtures must consider or poorly characterized health effects. There is intense interest in deter- pragmatic factors when measuring a large number of environmental mining the impact of exposure to environmental chemical mixtures on chemicals. Financial cost is perhaps the most important limiting factor human health. As the study of mixtures continues to evolve in the field when using biomarker-based approaches to study chemical mixtures of environmental epidemiology, it is imperative that we understand the because the inclusion of more components in targeted analytical chem- methodologic challenges of this research and the types of questions we istry methods increases the cost, often at the sake of sample size. In can address using epidemiological data. In this article, we summarize addition to cost, the volume of biospecimens (e.g., blood, urine, and some of the unique challenges in exposure assessment, statistical meth- ods, and methodology that epidemiologists face in addressing chemical plasma) required for these assays and the collection of samples from mixtures. We propose three broad questions that epidemiological stud- special populations (e.g., neonates or toddlers) must be considered. ies can address: a) What are the potential health impacts of individual The streetlight effect, a type of observational bias, has limited the chemical agents? b) What is the interaction among agents? And c) what number of chemicals studied because epidemiologists have typically are the health effects of cumulative exposure to multiple agents? As the measured only a few chemicals, choosing from those known to be field of mixtures research grows, we can use these three questions as a of concern or those for which measurement methods currently exist. basis for defining our research questions and for developing methods However, advances in analytic chemistry methods (e.g., nontargeted that will help us better understand the effect of chemical exposures on analysis) allow epidemiologists to broaden their scope and identify new human disease and well-being. or replacement chemicals introduced into commerce and industry. Some statistical challenges. The risk of false-positive results is a Introduction concern when analyzing a large number of exposures. Several statistical Biomonitoring studies confirm that humans are exposed to a large methods, including the Bonferroni correction, are used to reduce type number of environmental chemicals across the life span, often simul- I error rates in studies with a large number of hypotheses (Glickman taneously (CDC 2015; Woodruff et al. 2011). Although there is et al. 2014). The Bonferroni approach is an appealing method when growing concern that exposure to chemical mixtures during critical dealing with hundreds or thousands of potential hypotheses in studies periods of human development could increase the risk of adverse of mixtures; however, over-reliance on significance testing in obser - health effects including allergic diseases, cancer, neurodevelopmental vational studies where exposures are not randomized and are often disorders, reproductive disorders, and respiratory diseases, researchers correlated with one another can be problematic (Poole 2001; Rothman primarily study chemicals as if exposure occurs individually. This one- 1986, 1990; Savitz 1993). Although hypothesis testing is still used as chemical-at-a-time approach has left us with insufficient knowledge a method of inference, epidemiologists must also assess the validity, about the human health effects of exposure to chemical mixtures. magnitude, and precision of observed associations rather than just the Quantifying the risk of disease from environmental chemical mixtures statistical significance of associations. could help identify modifiable exposures that may be amenable to Type II errors can be equally problematic in studies of chemical public health interventions. mixtures. The statistical power to precisely estimate subtle effects As interest in chemical mixtures evolves, there is a need for greater between chemicals and human health may be limited by sample size, involvement of epidemiologists in this area of research (Carlin et al. the accuracy of exposure assessment methods (e.g., nondifferential 2013). We describe some of the unique challenges to studying envi- exposure misclassification), or multicollinearity issues due to correla - ronmental chemical mixtures in human populations and propose three tions among chemicals in the mixture (i.e., inflated variance estimates) broad questions related to chemical mixtures that epidemiology can (Cox et al. 2015; Braun et al. 2014). address. We believe this information will help investigators select the Confounding due to correlated exposures. While confounding best epidemiological and statistical methods for studying chemical due to socioeconomic factors associated with both the exposure and mixtures in human populations and consider the limitations of these outcome is almost always considered as a potential source of bias in methods in their studies. environmental epidemiology studies, confounding due to correlated copollutants can also exist. For example, in studies of persistent pollut Challenges to Studying Chemical Mixtures ants like polychlorinated biphenyls (PCBs), dioxins, and organochlo- Measuring environmental chemical exposure. Measuring human rine pesticides, exposure biomarkers are often correlated with each exposure to a large number of chemicals is a daunting task. First, other and may also be correlated with health outcomes (Longnecker the study of chemical mixtures requires accurate measurement of the et al. 2000). Such confounding, depending on the magnitude of corre individual components of the mixture. Sensitive and specific expo - lation between the pollutants, can make identifying the effect of an sure biomarkers are one method to assess chemical exposures. These individual chemical difficult, if not impossible. Thus, it is essential to biomarkers have revolutionized the study of chemical mixtures by understand the patterns of environmental exposures in human popula- allowing investigators to directly measure individual chemical concen - tions, as well as the correlation between individual agents, to determine trations in a variety of biospecimens (Needham et al. 2008). While if copollutant confounding may be present and whether public health chemical exposure biomarkers have many strengths, caution should interventions designed to reduce chemical exposures should target the be exercised because of the potential limitations related to misclas- entire mixture or components of it. Identifying important mixtures. The pattern of human exposure sification of exposures with high within-person variability (e.g., many short half-life chemicals such as bisphenol A), reverse causality due to to environmental chemicals is complex and multifactorial. Many pharmacokinetic factors (e.g., excretion) related to the outcome under pollutants are correlated with each other and some combinations A 6 volume 124 | number 1 | January 2016 • Environmental Health Perspectives Brief Communication of exposures are more likely than others. Because there is a need to statistical models and then decide which are the most important identify patterns of exposure that are most likely to be relevant to (Patel et al. 2012). This approach can be extended by accounting human health, some pollutant combinations may be of less relevance for the correlated nature of copollutants and adjusting for potential if there are no individuals with a given pattern of exposure. u Th s, in confounding bias using hierarchical or Bayesian methods (Braun et al. ranking the importance of these patterns, epidemiologists will need to 2014), as well as variable selection techniques such as weighted quan- consider the variability and prevalence of the exposure in the source tile sum (WQS) regression, elastic net, or least absolute shrinkage and population, the potential potency of the individual chemical compo- selection operator (LASSO) (Czarnota et al. 2015; Lenters et al. 2015). nents, and the ability to effectively reduce or mitigate the impact of Because of the correlated nature of many environmental pollutants, it exposure if adverse health effects are identified. is important to adjust for copollutant confounding using appropriate Lack of standard methods to evaluate environmental mixtures. methods when trying to identify single exposures within a mixture that A variety of statistical methods are available to address questions are most important to human health. Failure to do so could result in related to chemical mixtures (Billionnet et al. 2012; Sun et al. 2013), attributing one exposure to an adverse health outcome, when it might but there is no consensus on standard methods for studying envi- be due to another correlated copollutant. ronmental mixtures in epidemiological studies. Although we do not What are the interactions between chemicals within a mixture? advocate for a formulaic approach, we believe it would be helpful to The second question epidemiological studies can address is whether have a better understanding of the types of mixtures-related ques- two or more environmental chemical exposures have a greater than tions that epidemiologists can address so that appropriate methods additive (i.e., synergistic) or subadditive (i.e., antagonistic) association and statistical tools can be selected to adequately address research and with the health outcome of interest. For example, if we examine the public health needs. risk of disease in relation to two binary exposures, then the standard epidemiological approach to interaction determines if the risk of Types of Questions Epidemiology Can Address disease among those exposed to both agents simultaneously is greater In this section, we describe three broad questions related to chemical than the additive risk among those exposed to each agent individually. mixtures that epidemiological studies could address; in Table 1, we list Two points are important to consider with interactions: First, even examples of how these questions have been addressed using different in the absence of a greater than additive interaction between two or approaches, as well as the challenges to implementing them. more chemicals, joint exposure to these chemicals could have a cumu- What are the health effects of individual chemicals within a lative effect (Howdeshell et al. 2015). Second, it is critical to note mixture? The first question epidemiology can address is the associa - that toxicologists and epidemiologists define interaction differently . tion between individual chemical exposures in a mixture and human For instance, simple concentration-additive effects that are observed health outcomes. Because of the large number of environmental in toxicology experiments would be considered synergistic or antago- agents that humans are exposed to, there is a need to identify expo- nistic using epidemiological definitions when dose–response curves sures that are most strongly associated with adverse health outcomes are nonlinear (Howard and Webster 2013). including individual exposures or groups of highly correlated and Statistically examining interactions between chemicals would related exposures with a common source (e.g., Aroclors of PCB). help identify synergies or antagonisms between exposures or deter- The results of these studies would help guide public health efforts mine if one or more exposure modifies the effect of other expo- by allowing us to intervene on those agents that are most likely to be sures. This could be approached agnostically using variable selection associated with human health. procedures (e.g., LASSO or elastic net) or Bayesian kernel machine There are several methods to quantify the association between indi - regression (Bobb et al. 2015; Sun et al. 2013). Alternatively, a candi- vidual chemical exposures and human health outcomes. An approach date approach could examine interactions between chemicals that taken by many researchers is to quantify the association between each act on common biological pathways related to the health outcome chemical exposure and the health outcome of interest in separate of interest. Two primary determinants of our ability to identify Table 1. Description and examples of questions related to chemical mixtures and human health that epidemiological studies can address. Question Examples and Methods Challenges What are the health effects of • Quantified the association between prenatal exposure to 52 • Some approaches may not adequately address copollutant individual chemicals within a endocrine-disrupting chemicals and children’s autistic behaviors confounding. mixture? using semi-Bayesian shrinkage methods (Braun et al. 2014). • Multiple comparisons. • Used elastic net to examine the association between 16 prenatal • Disentangling the effect of highly correlated copollutants. exposures and birth weight (Lenters et al. 2015). • Examined the association between 188 environmental factors and serum lipid levels using an environment-wide association study (Patel et al. 2012). What are the interactions • Determined if the neurotoxic effects of lead were greater • Difference in toxicologic and epidemiologic definitions of between chemicals within a among children with higher manganese exposure using product interaction (Howard and Webster 2013). mixture? interaction terms (Claus Henn et al. 2012). • Multiple comparisons. • Identified and examined interactions between multiple metal • Imprecise effect estimates and reduced statistical power for biomarkers and child mental development using Bayesian kernel detecting interactions. machine regression (Bobb et al. 2015). What is the health effect of • Examined the relationship between child anthropometry and • Verifying the assumption of no interaction between individual cumulative chemical exposure? exposure to dioxins using a toxic equivalency summary measure components. (Burns et al. 2011). • Estimating cumulative exposure metrics for specific health • Estimated the association between different chemical classes and outcomes. non-Hodgkin lymphoma using weighted quantile sum regression • Availability of information to create biologically weighted (Czarnota et al. 2015). summary measures. • Used principal components analysis to examine the association • Interpretation of results from more complex statistical methods. between phthalate exposures and child anthropometry (Maresca et al. 2015). Environmental Health Perspectives • volume 124 | number 1 | January 2016 A 7 Brief Communication Conclusions interactions will be sample size and the pattern of correlation between exposures. With a fixed sample size, it may be difficult to identify By defining the types of research questions related to chemical mixtures interactions between chemicals because the number of observations that epidemiological studies can address, we hope to identify the gaps will diminish as smaller and smaller strata are examined for each in our knowledge and develop or apply appropriate statistical methods additional chemical-by-chemical interaction considered. In addition, that accurately quantify the impact of chemical mixtures on human when two or more exposures are highly correlated, there may be an health. In this article, we have chosen to focus on environmental chemi- insuc ffi ient number of participants with exposure to only one of the cals, but the three questions we describe naturally extend to other agents, thus limiting our ability to examine the impact of only one environmental exposures (e.g., air pollution and infectious agents), as exposure. Indeed, when exposures are highly correlated, their indi- well as the broader exposome (e.g., stress and nutrition) (Wild 2005). vidual or interactive effects are of less interest because public health By examining chemical mixtures, instead of one chemical at a time, we interventions aimed at reducing one exposure would likely reduce the may identify risk factors for diseases with environmental origins and other exposures. develop more targeted public health interventions. What is the health effect of cumulative chemical exposure? A third 1 2 3 Joseph M. Braun, Chris Gennings, Russ Hauser, and question estimates the association between cumulative chemical expo- Thomas F. Webster sure and human health. Here we are trying to quantify the summary Department of Epidemiology, Brown University, Providence, Rhode Island, effect of a class or multiple classes of exposure. Unlike the question of 2 USA; Department of Preventive Medicine, Icahn School of Medicine at Mount interaction, we assume that joint exposure to the chemicals does not Sinai, New York, New York, USA; Departments of Environmental Health and Epidemiology, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, have a greater than additive effect on the outcome (in the toxicolog - USA; Department of Environmental Health, Boston University School of Public ical sense) and that we can meaningfully condense the different expo - Health, Boston, Massachusetts, USA sures into a single summary metric. This may be most appropriate Address correspondence to J.M. Braun, Department of Epidemiology, Brown and insightful when the individual components of the mixture act University School of Public Health, 121 Main St., Providence, RI 02912 USA. via common biological pathways (e.g., phthalates or dioxins), when E-mail: Joseph_Braun_1@Brown.edu the exposure to individual agents is below some threshold of concern Motivation for the ideas in this article arose during the planning process [e.g., no-observed-adverse-effect level (NOAEL) or lowest-observed- for the National Institute of Environmental Health Sciences (NIEHS) work- shop, “Statistical Approaches for Assessing Health Effects of Environmental adverse-effect level (LOAEL)], and when there are individuals whose Chemical Mixtures in Epidemiology Studies.” aggregate exposure is over this threshold. We wish to thank the following scientists for facilitating the NIEHS work- Summaries of cumulative exposure can include simple summa- shop: L.S. Birnbaum, D.J. Carlin, G.W. Collman, C.H. Dilworth, K.A. Gray, tions of the concentration of individual exposures or by weighting J.J. Heindel, B.R. Joubert, C.V. Rider, K.W. Taylor, C.L. Thompson, W. Suk, them according to their biological potency [e.g., toxic equivalency and R. Woychik. We also thank D.A. Savitz, Brown University, Providence, factors (TEFs) for dioxin-like compounds] (Burns et al. 2011; Safe RI, for his helpful feedback on an earlier draft of this article. This work was supported by the following NIEHS grants: R00 ES020346, 1998). Although simple summary measures such as total serum R01 ES024381, P42 ES007381, P30 ES023515, R01 ES009718, and R01 PCB concentrations can be used, they often reflect the individual ES022955. component with the highest concentration in the mixture (Axelrad J.M.B. was financially compensated for conducting a re-analysis of a study of et al. 2009). Thus, these summary measures may not accurately child lead exposure for the plaintiffs in a public nuisance case related to childhood capture the cumulative effect of the mixture if the lower concen - lead poisoning that is not directly related to the present study. tration components are more potent than the higher concentra- The authors declare they have no actual or potential competing financial interests. tion ones. As an alternative, more complex weighting approaches Refe Rences can be used when making certain assumptions about the under- lying biology of the dose–response relationship (e.g., concentration Axelrad DA, Goodman S, Woodruff TJ. 2009. PCB body burdens in U.S. women of childbearing addition). One limitation to this approach is that epidemiologists age 2001–2002: an evaluation of alternate summary metrics of NHANES data. Environ Res will often require toxicological data that quantifies the biological 109(4):368–378. Billionnet C, Sherrill D, Annesi-Maesano I. 2012. Estimating the health effects of exposure to activity of individual components of the mixture (e.g., TEFs for multi-pollutant mixture. Ann Epidemiol 22(2):126–141. dioxin-like compounds). Furthermore, different health end points Bobb JF, Valeri L, Claus Henn B, Christiani DC, Wright RO, Mazumdar M, et al. 2015. Bayesian (e.g., cancer vs. neuro development) may need different summary kernel machine regression for estimating the health effects of multi-pollutant mixtures. Biostatistics 16(3):493–508. measures or weights to accurately describe the cumulative exposure Braun JM, Kalkbrenner AE, Just AC, Yolton K, Calafat AM, Sjödin A, et  al. 2014. Gestational to the mixture. exposure to endocrine-disrupting chemicals and reciprocal social, repetitive, and stereo- There are several additional strategies that can be used to estimate typic behaviors in 4- and 5-year-old children: the HOME study. Environ Health Perspect the cumulative health effects of a mixture. One could quantify the 122(5):513–520; doi: 10.1289/ehp.1307261. Burns JS, Williams PL, Sergeyev O, Korrick S, Lee MM, Revich B, et al. 2011. Serum dioxins and total biological activity in individual biospecimens through integra- polychlorinated biphenyls are associated with growth among Russian boys. Pediatrics tive assays (e.g., total estrogenicity) and use it as a measure of exposure 127(1):e59–e68. (Howard and Webster 2013; Vilahur et al. 2013). These measures Carlin DJ, Rider CV, Woychik R, Birnbaum LS. 2013. Unraveling the health effects of environmental mixtures: an NIEHS priority. Environ Health Perspect 121(1):A6–A8; doi: 10.1289/ehp.1206182. have the advantage of capturing both additive and interactive effects. CDC (Centers for Disease Control and Prevention). 2015. Fourth National Report on Human Statistically driven approaches, such as principal components anal- Exposure to Environmental Chemicals. Available: http://www.cdc.gov/biomonitoring/pdf/ ysis, can identify latent factors that explain the correlation between FourthReport_UpdatedTables_Feb2015.pdf [accessed 25 April 2015]. Claus Henn B, Schnaas L, Ettinger AS, Schwartz J, Lamadrid-Figueroa H, Hernandez-Avila M, mixture components. These factors can be used as an exposure vari - et al. 2012. Associations of early childhood manganese and lead co-exposure with neuro- able in statistical models (Maresca et al. 2015). Although principal development. Environ Health Perspect 120(1):126–131; doi: 10.1289/ehp.1003300. components methods are advantageous for studying some exposures, Cox KJ, Porucznik CA, Anderson DJ, Brozek EM, Szczotka KM, Bailey NM, et al. 2015. Exposure classification and temporal variability in urinary bisphenol-A concentrations among couples particularly those with common sources (e.g., air pollution), the in Utah: the HOPE Study. Environ Health Perspect; http://dx.doi.org/10.1289/ehp.1509752. derived factors are difficult to interpret because they are on a dimen - Czarnota J, Gennings C, Colt JS, De Roos AJ, Cerhan JR, Severson RK, et al. 2015. Analysis of sionless scale that is not specific to any one chemical exposure, and environmental chemical mixtures and non-Hodgkin lymphoma risk in the NCI-SEER NHL study. Environ Health Perspect 123(10):965–970; doi: 10.1289/ehp.1408630. they may be unique to the population being studied, thus limiting Glickman ME, Rao SR, Schultz MR. 2014. False discovery rate control is a recommended alter- their generalizability. Other methods, including empirically estimated native to Bonferroni-type adjustments in health studies. J Clin Epidemiol 67(8):850–857. weights, may be used to create weighted sums of standardized concen- Howard GJ, Webster TF. 2013. Contrasting theories of interaction in epidemiology and trations (Czarnota et al. 2015). toxicology. Environ Health Perspect 121(1):1–6; doi: 10.1289/ehp.1205889. A 8 volume 124 | number 1 | January 2016 • Environmental Health Perspectives Brief Communication Howdeshell KL, Rider CV, Wilson VS, Furr J, Lambright CR, Gray LE, Jr. 2015. Dose addition Rothman KJ. 1986. Significance questing. Ann Intern Med 105(3):445–447. models based on biologically relevant reductions in fetal testosterone accurately predict Rothman KJ. 1990. No adjustments are needed for multiple comparisons. Epidemiology postnatal reproductive tract alterations by a phthalate mixture in rats. Toxicol Sci; http:// 1(1):43–46. dx.doi.org/10.1093/toxsci/kfv196. Safe SH. 1998. Hazard and risk assessment of chemical mixtures using the toxic equivalency Lenters V, Portengen L, Rignell-Hydbom A, Jonsson BA, Lindh CH, Piersma AH, et  al. 2015. factor approach. Environ Health Perspect 106(Suppl 4):1051–1058. Prenatal phthalate, perfluoroalkyl acid, and organochlorine exposures and term birth Savitz DA. 1993. Is statistical significance testing useful in interpreting data? Reprod Toxicol weight in three birth cohorts: multi-pollutant models based on elastic net regression. 7(2):95–100. Environ Health Perspect; http://dx.doi.org/10.1289/ehp.1408933. Savitz DA. 2014. Invited commentary: interpreting associations between exposure biomarkers Longnecker MP, Ryan JJ, Gladen BC, Schecter AJ. 2000. Correlations among human plasma and pregnancy outcome. Am J Epidemiol 179(5):545–547. levels of dioxin-like compounds and polychlorinated biphenyls (PCBs) and implications for Sun Z, Tao Y, Li S, Ferguson KK, Meeker JD, Park SK, et  al. 2013. Statistical strategies for epidemiologic studies. Arch Environ Health 55(3):195–200. constructing health risk models with multiple pollutants and their interactions: possible Maresca MM, Hoepner LA, Hassoun A, Oberfield SE, Mooney SJ, Calafat AM, et  al. 2015. choices and comparisons. Environ Health 12:85; doi: 10.1186/1476-069X-12-85. Prenatal exposure to phthalates and childhood body size in an urban cohort. Environ Health Vilahur N, Molina-Molina JM, Bustamante M, Murcia M, Arrebola JP, Ballester F, et  al. 2013. Perspect; http://dx.doi.org/10.1289/ehp.1408750. Male specific association between xenoestrogen levels in placenta and birthweight. Needham LL, Calafat AM, Barr DB. 2008. Assessing developmental toxicant exposures via bio- Environ Int 51(1):174–181. monitoring. Basic Clin Pharmacol Toxicol 102(2):100–108. Wild CP. 2005. Complementing the genome with an “exposome”: the outstanding challenge Patel CJ, Cullen MR, Ioannidis JP, Butte AJ. 2012. Systematic evaluation of environmental fac- of environmental exposure measurement in molecular epidemiology. Cancer Epidemiol tors: persistent pollutants and nutrients correlated with serum lipid levels. Int J Epidemiol Biomarkers Prev 14(8):1847–1850. 41(3):828–843. Woodruff TJ, Zota AR, Schwartz JM. 2011. Environmental chemicals in pregnant women in the Poole C. 2001. Low P-values or narrow confidence intervals: which are more durable? United States: NHANES 2003–2004. Environ Health Perspect 119(6):878–885; doi: 10.1289/ Epidemiology 12(3):291–294. ehp.1002727. Environmental Health Perspectives • volume 124 | number 1 | January 2016 A 9 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Environmental Health Perspectives Pubmed Central

What Can Epidemiological Studies Tell Us about the Impact of Chemical Mixtures on Human Health?

Environmental Health Perspectives , Volume 124 (1) – Jan 1, 2016

Loading next page...
 
/lp/pubmed-central/what-can-epidemiological-studies-tell-us-about-the-impact-of-chemical-GOCCWnN0dS

References

References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.

Publisher
Pubmed Central
ISSN
0091-6765
eISSN
1552-9924
DOI
10.1289/ehp.1510569
Publisher site
See Article on Publisher Site

Abstract

Perspectives Brief Communication A Section 508– conformant HTML version of this ar ticle is available at http://dx.doi.org/10.1289/ehp.1510569. What Can Epidemiological Studies Tell Us about the Impact of Chemical Mixtures on Human Health? http://dx.doi.org/10.1289/ehp.1510569 study (Savitz 2014), or the inability for the biomarker to r epresent Summary : Humans are exposed to a large number of environmental exposure during the etiologically relevant time period. chemicals: Some of these may be toxic, and many others have unknown Second, epidemiologists who study mixtures must consider or poorly characterized health effects. There is intense interest in deter- pragmatic factors when measuring a large number of environmental mining the impact of exposure to environmental chemical mixtures on chemicals. Financial cost is perhaps the most important limiting factor human health. As the study of mixtures continues to evolve in the field when using biomarker-based approaches to study chemical mixtures of environmental epidemiology, it is imperative that we understand the because the inclusion of more components in targeted analytical chem- methodologic challenges of this research and the types of questions we istry methods increases the cost, often at the sake of sample size. In can address using epidemiological data. In this article, we summarize addition to cost, the volume of biospecimens (e.g., blood, urine, and some of the unique challenges in exposure assessment, statistical meth- ods, and methodology that epidemiologists face in addressing chemical plasma) required for these assays and the collection of samples from mixtures. We propose three broad questions that epidemiological stud- special populations (e.g., neonates or toddlers) must be considered. ies can address: a) What are the potential health impacts of individual The streetlight effect, a type of observational bias, has limited the chemical agents? b) What is the interaction among agents? And c) what number of chemicals studied because epidemiologists have typically are the health effects of cumulative exposure to multiple agents? As the measured only a few chemicals, choosing from those known to be field of mixtures research grows, we can use these three questions as a of concern or those for which measurement methods currently exist. basis for defining our research questions and for developing methods However, advances in analytic chemistry methods (e.g., nontargeted that will help us better understand the effect of chemical exposures on analysis) allow epidemiologists to broaden their scope and identify new human disease and well-being. or replacement chemicals introduced into commerce and industry. Some statistical challenges. The risk of false-positive results is a Introduction concern when analyzing a large number of exposures. Several statistical Biomonitoring studies confirm that humans are exposed to a large methods, including the Bonferroni correction, are used to reduce type number of environmental chemicals across the life span, often simul- I error rates in studies with a large number of hypotheses (Glickman taneously (CDC 2015; Woodruff et al. 2011). Although there is et al. 2014). The Bonferroni approach is an appealing method when growing concern that exposure to chemical mixtures during critical dealing with hundreds or thousands of potential hypotheses in studies periods of human development could increase the risk of adverse of mixtures; however, over-reliance on significance testing in obser - health effects including allergic diseases, cancer, neurodevelopmental vational studies where exposures are not randomized and are often disorders, reproductive disorders, and respiratory diseases, researchers correlated with one another can be problematic (Poole 2001; Rothman primarily study chemicals as if exposure occurs individually. This one- 1986, 1990; Savitz 1993). Although hypothesis testing is still used as chemical-at-a-time approach has left us with insufficient knowledge a method of inference, epidemiologists must also assess the validity, about the human health effects of exposure to chemical mixtures. magnitude, and precision of observed associations rather than just the Quantifying the risk of disease from environmental chemical mixtures statistical significance of associations. could help identify modifiable exposures that may be amenable to Type II errors can be equally problematic in studies of chemical public health interventions. mixtures. The statistical power to precisely estimate subtle effects As interest in chemical mixtures evolves, there is a need for greater between chemicals and human health may be limited by sample size, involvement of epidemiologists in this area of research (Carlin et al. the accuracy of exposure assessment methods (e.g., nondifferential 2013). We describe some of the unique challenges to studying envi- exposure misclassification), or multicollinearity issues due to correla - ronmental chemical mixtures in human populations and propose three tions among chemicals in the mixture (i.e., inflated variance estimates) broad questions related to chemical mixtures that epidemiology can (Cox et al. 2015; Braun et al. 2014). address. We believe this information will help investigators select the Confounding due to correlated exposures. While confounding best epidemiological and statistical methods for studying chemical due to socioeconomic factors associated with both the exposure and mixtures in human populations and consider the limitations of these outcome is almost always considered as a potential source of bias in methods in their studies. environmental epidemiology studies, confounding due to correlated copollutants can also exist. For example, in studies of persistent pollut Challenges to Studying Chemical Mixtures ants like polychlorinated biphenyls (PCBs), dioxins, and organochlo- Measuring environmental chemical exposure. Measuring human rine pesticides, exposure biomarkers are often correlated with each exposure to a large number of chemicals is a daunting task. First, other and may also be correlated with health outcomes (Longnecker the study of chemical mixtures requires accurate measurement of the et al. 2000). Such confounding, depending on the magnitude of corre individual components of the mixture. Sensitive and specific expo - lation between the pollutants, can make identifying the effect of an sure biomarkers are one method to assess chemical exposures. These individual chemical difficult, if not impossible. Thus, it is essential to biomarkers have revolutionized the study of chemical mixtures by understand the patterns of environmental exposures in human popula- allowing investigators to directly measure individual chemical concen - tions, as well as the correlation between individual agents, to determine trations in a variety of biospecimens (Needham et al. 2008). While if copollutant confounding may be present and whether public health chemical exposure biomarkers have many strengths, caution should interventions designed to reduce chemical exposures should target the be exercised because of the potential limitations related to misclas- entire mixture or components of it. Identifying important mixtures. The pattern of human exposure sification of exposures with high within-person variability (e.g., many short half-life chemicals such as bisphenol A), reverse causality due to to environmental chemicals is complex and multifactorial. Many pharmacokinetic factors (e.g., excretion) related to the outcome under pollutants are correlated with each other and some combinations A 6 volume 124 | number 1 | January 2016 • Environmental Health Perspectives Brief Communication of exposures are more likely than others. Because there is a need to statistical models and then decide which are the most important identify patterns of exposure that are most likely to be relevant to (Patel et al. 2012). This approach can be extended by accounting human health, some pollutant combinations may be of less relevance for the correlated nature of copollutants and adjusting for potential if there are no individuals with a given pattern of exposure. u Th s, in confounding bias using hierarchical or Bayesian methods (Braun et al. ranking the importance of these patterns, epidemiologists will need to 2014), as well as variable selection techniques such as weighted quan- consider the variability and prevalence of the exposure in the source tile sum (WQS) regression, elastic net, or least absolute shrinkage and population, the potential potency of the individual chemical compo- selection operator (LASSO) (Czarnota et al. 2015; Lenters et al. 2015). nents, and the ability to effectively reduce or mitigate the impact of Because of the correlated nature of many environmental pollutants, it exposure if adverse health effects are identified. is important to adjust for copollutant confounding using appropriate Lack of standard methods to evaluate environmental mixtures. methods when trying to identify single exposures within a mixture that A variety of statistical methods are available to address questions are most important to human health. Failure to do so could result in related to chemical mixtures (Billionnet et al. 2012; Sun et al. 2013), attributing one exposure to an adverse health outcome, when it might but there is no consensus on standard methods for studying envi- be due to another correlated copollutant. ronmental mixtures in epidemiological studies. Although we do not What are the interactions between chemicals within a mixture? advocate for a formulaic approach, we believe it would be helpful to The second question epidemiological studies can address is whether have a better understanding of the types of mixtures-related ques- two or more environmental chemical exposures have a greater than tions that epidemiologists can address so that appropriate methods additive (i.e., synergistic) or subadditive (i.e., antagonistic) association and statistical tools can be selected to adequately address research and with the health outcome of interest. For example, if we examine the public health needs. risk of disease in relation to two binary exposures, then the standard epidemiological approach to interaction determines if the risk of Types of Questions Epidemiology Can Address disease among those exposed to both agents simultaneously is greater In this section, we describe three broad questions related to chemical than the additive risk among those exposed to each agent individually. mixtures that epidemiological studies could address; in Table 1, we list Two points are important to consider with interactions: First, even examples of how these questions have been addressed using different in the absence of a greater than additive interaction between two or approaches, as well as the challenges to implementing them. more chemicals, joint exposure to these chemicals could have a cumu- What are the health effects of individual chemicals within a lative effect (Howdeshell et al. 2015). Second, it is critical to note mixture? The first question epidemiology can address is the associa - that toxicologists and epidemiologists define interaction differently . tion between individual chemical exposures in a mixture and human For instance, simple concentration-additive effects that are observed health outcomes. Because of the large number of environmental in toxicology experiments would be considered synergistic or antago- agents that humans are exposed to, there is a need to identify expo- nistic using epidemiological definitions when dose–response curves sures that are most strongly associated with adverse health outcomes are nonlinear (Howard and Webster 2013). including individual exposures or groups of highly correlated and Statistically examining interactions between chemicals would related exposures with a common source (e.g., Aroclors of PCB). help identify synergies or antagonisms between exposures or deter- The results of these studies would help guide public health efforts mine if one or more exposure modifies the effect of other expo- by allowing us to intervene on those agents that are most likely to be sures. This could be approached agnostically using variable selection associated with human health. procedures (e.g., LASSO or elastic net) or Bayesian kernel machine There are several methods to quantify the association between indi - regression (Bobb et al. 2015; Sun et al. 2013). Alternatively, a candi- vidual chemical exposures and human health outcomes. An approach date approach could examine interactions between chemicals that taken by many researchers is to quantify the association between each act on common biological pathways related to the health outcome chemical exposure and the health outcome of interest in separate of interest. Two primary determinants of our ability to identify Table 1. Description and examples of questions related to chemical mixtures and human health that epidemiological studies can address. Question Examples and Methods Challenges What are the health effects of • Quantified the association between prenatal exposure to 52 • Some approaches may not adequately address copollutant individual chemicals within a endocrine-disrupting chemicals and children’s autistic behaviors confounding. mixture? using semi-Bayesian shrinkage methods (Braun et al. 2014). • Multiple comparisons. • Used elastic net to examine the association between 16 prenatal • Disentangling the effect of highly correlated copollutants. exposures and birth weight (Lenters et al. 2015). • Examined the association between 188 environmental factors and serum lipid levels using an environment-wide association study (Patel et al. 2012). What are the interactions • Determined if the neurotoxic effects of lead were greater • Difference in toxicologic and epidemiologic definitions of between chemicals within a among children with higher manganese exposure using product interaction (Howard and Webster 2013). mixture? interaction terms (Claus Henn et al. 2012). • Multiple comparisons. • Identified and examined interactions between multiple metal • Imprecise effect estimates and reduced statistical power for biomarkers and child mental development using Bayesian kernel detecting interactions. machine regression (Bobb et al. 2015). What is the health effect of • Examined the relationship between child anthropometry and • Verifying the assumption of no interaction between individual cumulative chemical exposure? exposure to dioxins using a toxic equivalency summary measure components. (Burns et al. 2011). • Estimating cumulative exposure metrics for specific health • Estimated the association between different chemical classes and outcomes. non-Hodgkin lymphoma using weighted quantile sum regression • Availability of information to create biologically weighted (Czarnota et al. 2015). summary measures. • Used principal components analysis to examine the association • Interpretation of results from more complex statistical methods. between phthalate exposures and child anthropometry (Maresca et al. 2015). Environmental Health Perspectives • volume 124 | number 1 | January 2016 A 7 Brief Communication Conclusions interactions will be sample size and the pattern of correlation between exposures. With a fixed sample size, it may be difficult to identify By defining the types of research questions related to chemical mixtures interactions between chemicals because the number of observations that epidemiological studies can address, we hope to identify the gaps will diminish as smaller and smaller strata are examined for each in our knowledge and develop or apply appropriate statistical methods additional chemical-by-chemical interaction considered. In addition, that accurately quantify the impact of chemical mixtures on human when two or more exposures are highly correlated, there may be an health. In this article, we have chosen to focus on environmental chemi- insuc ffi ient number of participants with exposure to only one of the cals, but the three questions we describe naturally extend to other agents, thus limiting our ability to examine the impact of only one environmental exposures (e.g., air pollution and infectious agents), as exposure. Indeed, when exposures are highly correlated, their indi- well as the broader exposome (e.g., stress and nutrition) (Wild 2005). vidual or interactive effects are of less interest because public health By examining chemical mixtures, instead of one chemical at a time, we interventions aimed at reducing one exposure would likely reduce the may identify risk factors for diseases with environmental origins and other exposures. develop more targeted public health interventions. What is the health effect of cumulative chemical exposure? A third 1 2 3 Joseph M. Braun, Chris Gennings, Russ Hauser, and question estimates the association between cumulative chemical expo- Thomas F. Webster sure and human health. Here we are trying to quantify the summary Department of Epidemiology, Brown University, Providence, Rhode Island, effect of a class or multiple classes of exposure. Unlike the question of 2 USA; Department of Preventive Medicine, Icahn School of Medicine at Mount interaction, we assume that joint exposure to the chemicals does not Sinai, New York, New York, USA; Departments of Environmental Health and Epidemiology, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, have a greater than additive effect on the outcome (in the toxicolog - USA; Department of Environmental Health, Boston University School of Public ical sense) and that we can meaningfully condense the different expo - Health, Boston, Massachusetts, USA sures into a single summary metric. This may be most appropriate Address correspondence to J.M. Braun, Department of Epidemiology, Brown and insightful when the individual components of the mixture act University School of Public Health, 121 Main St., Providence, RI 02912 USA. via common biological pathways (e.g., phthalates or dioxins), when E-mail: Joseph_Braun_1@Brown.edu the exposure to individual agents is below some threshold of concern Motivation for the ideas in this article arose during the planning process [e.g., no-observed-adverse-effect level (NOAEL) or lowest-observed- for the National Institute of Environmental Health Sciences (NIEHS) work- shop, “Statistical Approaches for Assessing Health Effects of Environmental adverse-effect level (LOAEL)], and when there are individuals whose Chemical Mixtures in Epidemiology Studies.” aggregate exposure is over this threshold. We wish to thank the following scientists for facilitating the NIEHS work- Summaries of cumulative exposure can include simple summa- shop: L.S. Birnbaum, D.J. Carlin, G.W. Collman, C.H. Dilworth, K.A. Gray, tions of the concentration of individual exposures or by weighting J.J. Heindel, B.R. Joubert, C.V. Rider, K.W. Taylor, C.L. Thompson, W. Suk, them according to their biological potency [e.g., toxic equivalency and R. Woychik. We also thank D.A. Savitz, Brown University, Providence, factors (TEFs) for dioxin-like compounds] (Burns et al. 2011; Safe RI, for his helpful feedback on an earlier draft of this article. This work was supported by the following NIEHS grants: R00 ES020346, 1998). Although simple summary measures such as total serum R01 ES024381, P42 ES007381, P30 ES023515, R01 ES009718, and R01 PCB concentrations can be used, they often reflect the individual ES022955. component with the highest concentration in the mixture (Axelrad J.M.B. was financially compensated for conducting a re-analysis of a study of et al. 2009). Thus, these summary measures may not accurately child lead exposure for the plaintiffs in a public nuisance case related to childhood capture the cumulative effect of the mixture if the lower concen - lead poisoning that is not directly related to the present study. tration components are more potent than the higher concentra- The authors declare they have no actual or potential competing financial interests. tion ones. As an alternative, more complex weighting approaches Refe Rences can be used when making certain assumptions about the under- lying biology of the dose–response relationship (e.g., concentration Axelrad DA, Goodman S, Woodruff TJ. 2009. PCB body burdens in U.S. women of childbearing addition). One limitation to this approach is that epidemiologists age 2001–2002: an evaluation of alternate summary metrics of NHANES data. Environ Res will often require toxicological data that quantifies the biological 109(4):368–378. Billionnet C, Sherrill D, Annesi-Maesano I. 2012. Estimating the health effects of exposure to activity of individual components of the mixture (e.g., TEFs for multi-pollutant mixture. Ann Epidemiol 22(2):126–141. dioxin-like compounds). Furthermore, different health end points Bobb JF, Valeri L, Claus Henn B, Christiani DC, Wright RO, Mazumdar M, et al. 2015. Bayesian (e.g., cancer vs. neuro development) may need different summary kernel machine regression for estimating the health effects of multi-pollutant mixtures. Biostatistics 16(3):493–508. measures or weights to accurately describe the cumulative exposure Braun JM, Kalkbrenner AE, Just AC, Yolton K, Calafat AM, Sjödin A, et  al. 2014. Gestational to the mixture. exposure to endocrine-disrupting chemicals and reciprocal social, repetitive, and stereo- There are several additional strategies that can be used to estimate typic behaviors in 4- and 5-year-old children: the HOME study. Environ Health Perspect the cumulative health effects of a mixture. One could quantify the 122(5):513–520; doi: 10.1289/ehp.1307261. Burns JS, Williams PL, Sergeyev O, Korrick S, Lee MM, Revich B, et al. 2011. Serum dioxins and total biological activity in individual biospecimens through integra- polychlorinated biphenyls are associated with growth among Russian boys. Pediatrics tive assays (e.g., total estrogenicity) and use it as a measure of exposure 127(1):e59–e68. (Howard and Webster 2013; Vilahur et al. 2013). These measures Carlin DJ, Rider CV, Woychik R, Birnbaum LS. 2013. Unraveling the health effects of environmental mixtures: an NIEHS priority. Environ Health Perspect 121(1):A6–A8; doi: 10.1289/ehp.1206182. have the advantage of capturing both additive and interactive effects. CDC (Centers for Disease Control and Prevention). 2015. Fourth National Report on Human Statistically driven approaches, such as principal components anal- Exposure to Environmental Chemicals. Available: http://www.cdc.gov/biomonitoring/pdf/ ysis, can identify latent factors that explain the correlation between FourthReport_UpdatedTables_Feb2015.pdf [accessed 25 April 2015]. Claus Henn B, Schnaas L, Ettinger AS, Schwartz J, Lamadrid-Figueroa H, Hernandez-Avila M, mixture components. These factors can be used as an exposure vari - et al. 2012. Associations of early childhood manganese and lead co-exposure with neuro- able in statistical models (Maresca et al. 2015). Although principal development. Environ Health Perspect 120(1):126–131; doi: 10.1289/ehp.1003300. components methods are advantageous for studying some exposures, Cox KJ, Porucznik CA, Anderson DJ, Brozek EM, Szczotka KM, Bailey NM, et al. 2015. Exposure classification and temporal variability in urinary bisphenol-A concentrations among couples particularly those with common sources (e.g., air pollution), the in Utah: the HOPE Study. Environ Health Perspect; http://dx.doi.org/10.1289/ehp.1509752. derived factors are difficult to interpret because they are on a dimen - Czarnota J, Gennings C, Colt JS, De Roos AJ, Cerhan JR, Severson RK, et al. 2015. Analysis of sionless scale that is not specific to any one chemical exposure, and environmental chemical mixtures and non-Hodgkin lymphoma risk in the NCI-SEER NHL study. Environ Health Perspect 123(10):965–970; doi: 10.1289/ehp.1408630. they may be unique to the population being studied, thus limiting Glickman ME, Rao SR, Schultz MR. 2014. False discovery rate control is a recommended alter- their generalizability. Other methods, including empirically estimated native to Bonferroni-type adjustments in health studies. J Clin Epidemiol 67(8):850–857. weights, may be used to create weighted sums of standardized concen- Howard GJ, Webster TF. 2013. Contrasting theories of interaction in epidemiology and trations (Czarnota et al. 2015). toxicology. Environ Health Perspect 121(1):1–6; doi: 10.1289/ehp.1205889. A 8 volume 124 | number 1 | January 2016 • Environmental Health Perspectives Brief Communication Howdeshell KL, Rider CV, Wilson VS, Furr J, Lambright CR, Gray LE, Jr. 2015. Dose addition Rothman KJ. 1986. Significance questing. Ann Intern Med 105(3):445–447. models based on biologically relevant reductions in fetal testosterone accurately predict Rothman KJ. 1990. No adjustments are needed for multiple comparisons. Epidemiology postnatal reproductive tract alterations by a phthalate mixture in rats. Toxicol Sci; http:// 1(1):43–46. dx.doi.org/10.1093/toxsci/kfv196. Safe SH. 1998. Hazard and risk assessment of chemical mixtures using the toxic equivalency Lenters V, Portengen L, Rignell-Hydbom A, Jonsson BA, Lindh CH, Piersma AH, et  al. 2015. factor approach. Environ Health Perspect 106(Suppl 4):1051–1058. Prenatal phthalate, perfluoroalkyl acid, and organochlorine exposures and term birth Savitz DA. 1993. Is statistical significance testing useful in interpreting data? Reprod Toxicol weight in three birth cohorts: multi-pollutant models based on elastic net regression. 7(2):95–100. Environ Health Perspect; http://dx.doi.org/10.1289/ehp.1408933. Savitz DA. 2014. Invited commentary: interpreting associations between exposure biomarkers Longnecker MP, Ryan JJ, Gladen BC, Schecter AJ. 2000. Correlations among human plasma and pregnancy outcome. Am J Epidemiol 179(5):545–547. levels of dioxin-like compounds and polychlorinated biphenyls (PCBs) and implications for Sun Z, Tao Y, Li S, Ferguson KK, Meeker JD, Park SK, et  al. 2013. Statistical strategies for epidemiologic studies. Arch Environ Health 55(3):195–200. constructing health risk models with multiple pollutants and their interactions: possible Maresca MM, Hoepner LA, Hassoun A, Oberfield SE, Mooney SJ, Calafat AM, et  al. 2015. choices and comparisons. Environ Health 12:85; doi: 10.1186/1476-069X-12-85. Prenatal exposure to phthalates and childhood body size in an urban cohort. Environ Health Vilahur N, Molina-Molina JM, Bustamante M, Murcia M, Arrebola JP, Ballester F, et  al. 2013. Perspect; http://dx.doi.org/10.1289/ehp.1408750. Male specific association between xenoestrogen levels in placenta and birthweight. Needham LL, Calafat AM, Barr DB. 2008. Assessing developmental toxicant exposures via bio- Environ Int 51(1):174–181. monitoring. Basic Clin Pharmacol Toxicol 102(2):100–108. Wild CP. 2005. Complementing the genome with an “exposome”: the outstanding challenge Patel CJ, Cullen MR, Ioannidis JP, Butte AJ. 2012. Systematic evaluation of environmental fac- of environmental exposure measurement in molecular epidemiology. Cancer Epidemiol tors: persistent pollutants and nutrients correlated with serum lipid levels. Int J Epidemiol Biomarkers Prev 14(8):1847–1850. 41(3):828–843. Woodruff TJ, Zota AR, Schwartz JM. 2011. Environmental chemicals in pregnant women in the Poole C. 2001. Low P-values or narrow confidence intervals: which are more durable? United States: NHANES 2003–2004. Environ Health Perspect 119(6):878–885; doi: 10.1289/ Epidemiology 12(3):291–294. ehp.1002727. Environmental Health Perspectives • volume 124 | number 1 | January 2016 A 9

Journal

Environmental Health PerspectivesPubmed Central

Published: Jan 1, 2016

References