About

Corwin M Zigler is a Professor of Biostatistics at the Brown University School of Public Health and serves as the Director of Biostatistics. His research focuses on quantitative methodology for evaluating the health impacts of environmental and climate-related exposures. His statistical research primarily involves causal inference, with particular emphasis on spatially-indexed data, spatial confounding, interference networks, physical process modeling, and Bayesian methodology. His work integrates statistical methodology, epidemiology, large-scale computation, and atmospheric science to better understand how environmental and climate policies impact human health. Dr. Zigler completed his Ph.D. in Biostatistics at the University of California, Los Angeles in 2010. Prior to his current position at Brown, he served as faculty in the Department of Statistics and Data Sciences at the University of Texas at Austin and in the Department of Biostatistics at the Harvard T.H. Chan School of Public Health. His research has been funded by the National Institutes of Health, the Health Effects Institute, and the U.S. Environmental Protection Agency. He has also served on multiple specialty panels for the U.S. EPA Clean Air Scientific Advisory Committee. His professional recognitions include being elected a Fellow of the American Statistical Association in 2023, receiving the Health Policy Statistics Section Mid-Career Achievement Award in 2019, the Rothman Epidemiology Prize in 2019, and an Honorable Mention for the Mitchell Prize from the International Society for Bayesian Analysis.

Research topics

• Medicine
• Computer Science
• Internal medicine
• Pediatrics
• Intensive care medicine
• Information Retrieval
• Artificial Intelligence
• Environmental health
• Pathology
• Environmental planning
• Mathematics
• Environmental chemistry
• Chemistry
• Data science
• Immunology
• Virology
• Econometrics
• Environmental science
• Emergency medicine
• Statistics

Selected publications

• A prospective study of periconceptional perceived stress and rate of miscarriage

  Human Reproduction · 2026-02-10

  articleOpen access

  STUDY QUESTION: To what extent is perceived stress during preconception and pregnancy associated with miscarriage incidence? SUMMARY ANSWER: Perceived stress during early pregnancy, but not preconception, was associated with higher miscarriage incidence. WHAT IS KNOWN ALREADY: Some studies have found that higher stress levels are associated with miscarriage risk. However, many of these studies were retrospective, focused on occupational stress only, and/or suffered from under-ascertainment of miscarriage. STUDY DESIGN, SIZE, DURATION: Pregnancy Study Online (PRESTO) is an ongoing prospective preconception cohort study that recruited participants during 2013-2025. Eligible participants were females aged 21-45 years, who resided in the USA or Canada and were trying to conceive without fertility treatments. Eligible partners were males aged ≥21 years. PARTICIPANTS/MATERIALS, SETTING, METHODS: We collected data on perceived stress using the 10-item version of the Perceived Stress Scale (PSS-10) during preconception (every 8 weeks) and early pregnancy for female participants and during preconception only for male participants. We identified pregnancies and miscarriages on bimonthly follow-up questionnaires during preconception and additional questionnaires during early and late pregnancy and postpartum. We fit Cox proportional hazards regression models to estimate hazard ratios (HR) and 95% CIs for the effect of preconception PSS-10 scores (n = 11 189 female and 2656 male participants) and early pregnancy PSS-10 scores (n = 8319 female participants) on miscarriage incidence, adjusting for potential confounders. MAIN RESULTS AND THE ROLE OF CHANCE: About 20% of the pregnancies ended in miscarriage, with the loss occurring at a median of six gestational weeks. Preconception PSS-10 scores in the female or male partner were not appreciably associated with miscarriage incidence. Female PSS-10 scores during gestational weeks 5-8 were strongly associated with higher miscarriage incidence: adjusted HRs for PSS-10 scores of 10-14, 15-19, 20-24, and ≥25 vs <10 in gestational weeks 5-8 were 1.38 (95% CI: 1.07, 1.77), 1.17 (95% CI: 0.89, 1.52), 1.35 (95% CI: 1.00, 1.83), and 2.05 (95% CI: 1.40, 2.99), respectively. In week-specific analyses, an association existed during weeks 4-8 and peaked at week 7. LIMITATIONS, REASONS FOR CAUTION: Our results may be susceptible to reverse causation, unmeasured confounding by nausea and vomiting in pregnancy, and exposure misclassification. WIDER IMPLICATIONS OF THE FINDINGS: Interventions aimed at decreasing stress during early pregnancy may be effective at reducing miscarriage incidence, but confirmation of our results in randomized studies is warranted. STUDY FUNDING/COMPETING INTEREST(S): This work was supported by the Eunice Kennedy Shriver National Institute of Child Health and Human Development (R01-HD086742, R01-HD105863). Lauren Wise has received in-kind donations for primary data collection in PRESTO from ChartNeo.com. The other authors have no conflicts to report. TRIAL REGISTRATION NUMBER: N/A.

  PublisherOA PDFDOI

• Subtypes and onset of hypertensive disorders of pregnancy and cardiovascular disease within 5 years after delivery

  Frontiers in Cardiovascular Medicine · 2026-02-19

  articleOpen access

  Background: Hypertensive disorders of pregnancy (HDP) are established predictors of long-term cardiovascular disease (CVD), but the short-term postpartum CVD risk by HDP subtype and onset remains unclear. Materials and methods: We linked electronic health records with vital statistics for 755,606 singleton deliveries in Florida (2012-2017) to examine how different subtypes and onset of HDP are associated with CVD within 5 years postpartum. We classified HDP into six subtypes-chronic hypertension, gestational hypertension, mild preeclampsia, severe preeclampsia, eclampsia, and superimposed preeclampsia - and defined onset as early (<34 weeks) or late (≥34 weeks). Seven CVD outcomes (heart failure, ischemic heart disease, cerebrovascular disease/stroke, arrhythmia/cardiac arrest, cardiomyopathy, peripheral vascular disease, and new-onset chronic hypertension) within five years postpartum were identified. Cox proportional hazards models estimated hazard ratios (HRs) and 95% confidence intervals (CIs) after adjusting for sociodemographic and clinical covariates. Results: Compared with normotensive pregnancies, superimposed preeclampsia carried the highest risks for stroke (HR 3.39; 95% CI 2.93-3.92), arrhythmia (2.62; 2.25-3.07), and peripheral vascular disease (3.09; 2.67-3.57). Eclampsia showed the strongest associations with heart failure (5.23; 3.70-7.39), ischemic heart disease (3.61; 2.65-4.92), and cardiomyopathy (5.25; 3.37-8.18). Severe preeclampsia was most strongly associated with new hypertension (3.27; 3.10-3.46). Early-onset eclampsia and superimposed preeclampsia showed higher CVD risks than their late-onset counterparts, whereas late-onset gestational hypertension and mild preeclampsia were more strongly associated with new hypertension and cardiomyopathy, respectively. Conclusions: HDP subtypes and onset timing impart distinct CVD risk profiles within five years postpartum.

  PublisherOA PDFDOI

• Evidence-based water fluoridation policy

  Science Advances · 2025-11-19 · 4 citations

  articleOpen access1st authorCorresponding

  A national-scale fluoridation study addresses policy-relevant exposure levels and provides evidence that adverse neurodevelopmental effects do not result from municipal fluoridation.

  PublisherDOI

• Deployment to Military Bases With Open Burn Pits and Mental Health Conditions and Injury Mortality Among US Veterans

  Journal of Occupational and Environmental Medicine · 2025-07-15 · 3 citations

  articleSenior author

  OBJECTIVE: This study aimed to examine associations between deployment to US military bases with open burn pits and mental health conditions and injury-related mortality among veterans. METHODS: We analyzed a cohort of 439,919 US Army and Air Force Veterans deployed to Operation Enduring Freedom or Operation Iraqi Freedom (2001-2011). Deployment records were linked with Veterans Health Administration data. Exposure was defined as cumulative days deployed to bases with burn pits. RESULTS: Deployment duration to burn pit-exposed bases are associated with increased risk of postdeployment diagnoses of posttraumatic stress disorder, intracranial damage and injuries, and unintentional injury-related mortality. CONCLUSIONS: Cumulative exposure to open burn pits was associated with elevated risk of long-term psychiatric and injury-related outcomes among veterans. These findings highlight the need for continued monitoring and support for service members exposed to environmental hazards during military deployment.

  PublisherDOI

• The impact of burn pit waste segregation practices on respiratory and cardiovascular health risks among US military veterans deployed to Iraq and Afghanistan

  Environmental Health · 2025-07-01 · 2 citations

  articleOpen access

  BACKGROUND: Open-air burning was a prevalent waste management method at many U.S. military bases during the wars in Afghanistan and Iraq. Past studies of the health impacts of burn pit exposure have relied on exposure assessments that did not account for waste segregation practices introduced in the later years of the wars, such as removing hazardous and medical waste before open burning and the use of incinerators. OBJECTIVE: We developed a refined exposure assessment that accounts for waste management practices on military bases and evaluated the impact of waste segregation and incineration on cardiovascular and respiratory health outcomes among veterans deployed during these conflicts. METHODS: The study cohort consisted of 459,381 Army and Air Force veterans who were deployed between 2005 and 2011 and received health care through the Veterans Health Administration (VHA) after deployment. The 109 most populated military bases in Afghanistan and Iraq were classified into four waste disposal categories by year: unsegregated, segregated, incineration, and no burning or incineration. Individual exposure was defined as the total number of days spent at bases based on the Department of Defense deployment histories. Health outcomes were determined through VHA healthcare records, from the end of deployment through the end of follow-up in 2020. Logistic regression was performed to investigate the association between deployment to bases with varying waste management practices and the risk of respiratory and cardiovascular diseases. RESULTS: Deployment to bases using burn pits with unsegregated waste was associated with elevated risks of hypertension and asthma, whereas deployment to bases that segregated waste or used incinerators was not. Prolonged deployment (highest duration tertile of > 240 days) to bases with unsegregated waste burning was associated with a 16% higher risk of hypertension (aOR 1.16, 95% CI 1.13-1.19) compared to those never stationed at such bases. There was a clear deployment duration-response association for hypertension, but this was not observed for asthma. CONCLUSIONS: The observed increased risk of hypertension and asthma among military veterans deployed to bases that used open burning of unsegregated waste - but not among those deployed to bases that segregated waste or used incinerators - highlights the importance of considering waste management methods in future studies examining the health effects of burn pit exposures among military veterans.

  PublisherOA PDFDOI

• Hill's considerations are not causal criteria

  Journal of Clinical Epidemiology · 2025-11-22 · 1 citations

  articleOpen access1st authorCorresponding

  Hill's list of considerations for assessing causality, proposed 60 years ago, became a landmark in the interpretation of epidemiologic evidence. However, it has been and continues to be misused as a list of causal criteria to be scored and summed, despite causal inference being unattainable through the application of this or any other algorithm. Recognizing the distinction between statistical associations and causal effects was a key contribution of Hill. While he identified several clues for distinguishing between causal and noncausal associations, causal inference in epidemiology has become much more explicit and effective. Rather than relying on Hill's indirect hints of potential bias by considering strength of association or dose-response gradients, newer methods such as quantitative bias analysis directly assess confounding and other candidate biases that compete with causal explanations, leading to more informed inferences. Similarly, the interpretation of consistency depends on variation in methods across studies; triangulation may be used to search for informative inconsistencies, strengthening causal inference. Most importantly, a causal connection is not a categorical property bestowed upon an association based on Hill's considerations or any other checklist. Causal inference is an inherently indirect process, with the inference gradually crystallizing by withstanding challenges from competing theories in which other explanations, including random error or biases, are found not to account for the measured association.

  PublisherDOI

• To What End? Clarifying the Purpose and Value of Birth Cohorts

  Paediatric and Perinatal Epidemiology · 2025-01-10

  article1st authorCorresponding

  The author declares no conflicts of interest.

  PublisherDOI

• LIFE-Moms: effects of multicomponent lifestyle randomized control trial on physical activity during pregnancy in women with overweight and obesity

  International Journal of Behavioral Nutrition and Physical Activity · 2025-09-30 · 2 citations

  articleOpen access

  BACKGROUND: This report details the effect of LIFE-Mom's multicomponent lifestyle interventions on physical activity (PA) and inactivity time across pregnancy (2nd and 3rd trimesters) and their effect on gestational weight gain (GWG) and maternal/neonatal outcomes, a pre-specified secondary analysis. METHODS: were randomized to lifestyle interventions with dietary and PA counseling or standard care. PA and inactivity time measured by accelerometry and metabolic and inflammatory biomarkers measured in fasting blood are reported in 522 pregnant people at baseline and end of pregnancy. Generalized linear models with and without covariates were used to evaluate group differences (intervention vs. control) and, separately, time differences (total sample with both groups combined). RESULTS: Although there were statistically significant differences in vigorous activity between the intervention and control group (p = .024), there were no clinically meaningful differences in PA. In the combined sample, moderate to vigorous PA (MVPA) significantly decreased across pregnancy (mean ± SD: 72.9 ± 29.1 min/day vs 63.9 ± 28.1 min/day; p < 0.0001), and inactivity time increased [617.5 min/day (573.5, 659.6) vs 630.4 min/day (56.7, 679.9); p < 0.0001]. Increased inactivity time was associated with a less favorable maternal milieu (biomarker Z-scores) for pro-inflammatory (0.2 ± 0.1; p = 0.003) and cardiometabolic markers (0.1 ± 0.07; p = 0.030). CONCLUSIONS: Physical activity declined over the course of pregnancy, though the intervention group experienced a smaller reduction in activity levels. Our results linked increased inactivity time to maternal metabolic dysregulation and inflammation. Further research is needed to determine if intensive interventions reducing inactivity can improve maternal health and weight outcomes in pregnant people with overweight and obesity. TRIAL REGISTRATION: NCT01545934, NCT01616147, NCT01771133, NCT01631747, NCT01768793, NCT01610752, and NCT01812694.

  PublisherOA PDFDOI

• Healthcare utilization during acute medically attended episodes of respiratory syncytial virus-related lower respiratory tract infection among infants in the United States

  PLoS ONE · 2025-02-10 · 1 citations

  articleOpen accessCorresponding

  BACKGROUND: Respiratory syncytial virus (RSV) is the leading cause of infant hospitalization in the United States. Understanding healthcare utilization associated with medically attended (MA) RSV lower respiratory tract infection (LRTI) might inform research priorities aimed at reducing RSV-associated pediatric morbidity. We described healthcare utilization during acute MA RSV LRTI episodes within a geographically diverse cohort of infants in the United States. METHODS: We created retrospective cohorts of infants born in the United States from July 1, 2016 through February 29, 2020 in each of three de-identified insurance claims datasets: Merative MarketScan Commercial Claims and Encounters, Multi-State MarketScan Medicaid, and Optum's de-identified Clinformatics ® Data Mart. We identified infants' first MA RSV LRTI diagnosis during their first RSV season and followed them for 7 subsequent days to record outpatient, emergency department, and inpatient hospital utilization. We calculated the number of outpatient visits, emergency department visits, and inpatient hospital stays occurring during this acute episode and estimated the proportion of episodes involving ≥ 2 visits to a given healthcare setting. RESULTS: In the CCAE database, we identified 25,409 acute MA RSV LRTI episodes under the specific RSV definition and 69,068 under the sensitive definition. In the MDCD database, these totals were 67,357 and 170,744, while in the CDM database, they were 12,402 and 31,363, respectively. Across data sources, 34%-69% of infants' first acute MA RSV LRTI episodes involve 2 or more visits to a healthcare setting within 7 days. The percentage of episodes involving at least 2 visits ranged from 34-62% among healthy term infants, 38-65% for Palivizumab-eligible infants, and 38-69% for infants with other comorbidities. CONCLUSIONS: Within a week of their first MA RSV LRTI diagnosis, infants frequently experience at least 2 visits to one or more healthcare settings, regardless of their comorbidity profile. The percentage of MA RSV LRTI episodes involving at least 2 visits to a healthcare setting may vary by insurance claims database, even between commercial payers.

  PublisherDOI

• Consequential (and inconsequential) environmental epidemiology

  Environmental Epidemiology · 2025-10-22

  articleOpen access1st authorCorresponding

  Epidemiologists have effectively exploited advances in assessing exposure and health outcomes to pursue potential causal links between the environment and disease. The rapidly decreasing cost of assaying biospecimens for multiple chemicals and the proliferation of biobanks from established surveys (e.g., the National Health and Nutrition Examination Surveys, https://www.cdc.gov/nchs/hus/sources-definitions/nhanes.htm), repositories (e.g., UK Biobank, https://www.ukbiobank.ac.uk/), and study cohorts (e.g., the MIREC Biobank, https://www.mirec-canada.ca/en/about) provide a vast array of research opportunities. Analogously, the assessment of health endpoints based on clinical biomarkers is often inexpensive for common measures such as hormones, lipids, or markers of inflammation. Although the designs vary somewhat, the prototypic study examines one or more exposure biomarkers (generally in the low range) in relation to one or more clinical biomarkers (typically in the subclinical range). Although in principle, more evidence can only be beneficial, the rapid growth in the availability of comprehensive datasets coupled with lower barriers to analyzing those data (including through AI-assisted analyses and manuscript preparation) is leading to a deluge of new publications.1 Many of these studies identify some positive associations, often with reasonable statistical power, even in relatively small populations. However, given (a) the tenuous connection between environmental biomarkers and exposure sources, (b) the indirect relevance of subclinical health indicators to disease, and (c) abundant opportunities for both false positives and selective publication of positive results, it is not clear that the high volume of such analyses offers progress toward identification of important etiologic relationships or the improvement of public health. Disconnect between biomarkers and environmental exposures Ideally, exposure biomarkers provide an integrated measure of internal dose of an environmental toxicant, circumventing the need to assess environmental sources or query individual behaviors. Given advances in technology, the menu of candidate exposure biomarkers is extensive, and for many exposures, nearly everyone has detectable levels. But for many chemicals, the connection between environmental sources and biomarker levels is ill-defined. For example, some forms of PFAS are present in virtually everyone, but except for individuals exposed to specific sources from contaminated drinking water, a unique dietary source, or through the work environment, the sources of exposure are unknown. In populations with typical background exposures, differences in exposure biomarkers among study participants tend to be quite small and studies end up contrasting those with very low exposure to those with extremely low exposure. Measurement error alone generates interindividual differences, but differences due to random measurement error alone would be expected to underestimate any underlying causal effects. Metabolic variation in uptake, metabolism, and excretion among individuals with essentially the same exogenous sources can generate variation in exposure biomarkers. To the extent that metabolic variation drives variation in exposure biomarkers, spurious associations with health outcomes are likely to be found,2,3 unrelated to exogenous exposure. Since the goal of etiologic research is to identify opportunities to reduce exposure and improve public health,4 biomarkers are problematic since they are not typically amenable to intervention; only the exogenous determinants of exposure can be changed. For the study to suggest actions to reduce exposure, it needs to address a specific exposure source, such as the concentration of a toxicant in air or water, or a behavioral influence on exposure, such as diet or use of a consumer product. In simple terms, if the goal of research is to identify causal effects of environmental exposures5 rather than to merely describe associations involving environmental agents, then studies based on biomarkers without a clear exogenous source are minimally helpful. Significance of health outcomes Many studies of the potential health effects of environmental toxicants consider subclinical health outcomes, variation within the normal range, given their reliance on relatively small convenience samples. Commonly used health endpoints include clinical biomarkers (e.g., lipids, hormones, and micronutrients), anthropometric measures (e.g., birthweight and body mass index), physiological measures (e.g., blood pressure), and scales assessing symptoms, behaviors, or capability (e.g., neurobehavioral tests). By considering continuous measures, even small studies may have sufficient statistical power to detect associations, but the magnitude of difference in the health outcome across individuals with relatively low exposures is often clinically inconsequential. The standard defense for such studies of variation in the normal range includes two claims: (1) a small change in a marker of health applied across a large population may have important public health consequences, even if the magnitude is not important for any individual, and (2) even if the effect is modest, some individuals will be pushed over a threshold such that their health is meaningfully worse as a result. If the small change is causal, and a sizeable population is affected, there would be public health relevance so long as there is clear, independent evidence of the presumed downstream consequences. For example, while studies of environmental toxicants may not be capable of demonstrating that a small increment in blood pressure results in elevated risk of stroke, since other studies have already established the relationship between blood pressure and stroke, even a small increase in average blood pressure is meaningful on a population level. In contrast, a small shift in thyroid hormones or birthweight may not have significant health consequences. The second argument concerns exposure causing some individuals to cross a clinically consequential threshold. There are very few health phenomena that have real thresholds dividing subclinical alterations from clinically consequential changes, even though arbitrary cutpoints are frequently used to facilitate clinical diagnoses or risk stratification. Statistically, shifting the entire distribution of an outcome will change the proportion of the population falling above any given threshold, but that does not mean that even those individuals who cross the threshold are harmed. Going from a birthweight of 2505 to 2495 g would lead to a shift from “normal birthweight” to “low birthweight” but the threshold itself is arbitrary, and the 10 g shift is without health consequences. A similar argument can be made for defining obesity based on body mass index, diabetes or prediabetes based on categorization of levels of hemoglobin A1c, hypercholesterolemia, or hypertension. The main consequence of crossing the threshold is qualifying for clinical intervention. Pathways to false positive findings Identifying small causal effects is challenging, requiring researchers to effectively distinguish between “no effect” and “a very small effect.” Many health outcomes have strong sociodemographic determinants and many (but by no means all) environmental toxicants are higher in socially disadvantaged groups. In this scenario, confounding by social factors is present, and even after attempts at statistical adjustment for social determinants, residual confounding is likely to remain. Constructs such as “social disadvantage” are extremely difficult to fully capture, with indicators such as education or income helpful but incomplete. Statistical control of confounding is only effective to the extent that the putative source of confounding is accurately measured. Biomarkers associated with various lifestyle factors and physiologic variation may be confounded in ways that cannot be fully addressed through conventional statistical approaches, calling for more sophisticated designs such as Mendelian randomization or the use of instrumental variables. Finding that an unadjusted association between an environmental agent and disease is markedly attenuated (but not eliminated) by adjustment for an imperfectly measured confounder suggests more complete adjustment would likely result in an even weaker association.6,7 Distinguishing between a small causal effect and an association with residual confounding may simply not be possible. Studies of environmental biomarkers often address multiple chemicals and use a variety of metrics of exposure based on varying cutpoints or combinations of exposures. Health outcomes may include multiple options (e.g., scales of behavior and different hormones) as well as varying cutpoints or combinations of measures. Subgroups based on sex, ethnicity, calendar time, or other covariates may be evaluated, proliferating the candidate pool of associations. The potential for at least some false positives is substantial. However, the solution is not to make formal adjustments for statistical tests8 but instead to examine the coherence of the findings, considering prior research, strength of the association, presence of dose-response gradients, and other informative patterns in the data. Distinguishing between “blips” and “signals” is not purely a statistical issue but a conceptual one that needs to draw on subject matter knowledge, previous research, and patterns in the study data. Use of flexible analytic approaches exacerbates the problem of false positives.9 In environmental epidemiology, splines are now easy to implement and allow the analyst to detect nonlinear or even nonmonotonic relationships. However, segments of the regression line that are positive somewhere in the range of exposure may fit the data well but lead to inexplicable and nonreplicable results. Similarly, when the study of individual exposures does not yield clear associations or readily interpretable results, various mixture models can be used to fit the data and generate positive associations but not point to modifiable causes of disease.10 In studies of chronic disease, there is an opportunity to evaluate multiple lag periods, increasing the opportunity to find positive associations.11 The problem is a combination of overfitting the data and a tendency to overinterpret the results. When the goal is to find positive associations rather than accurately estimate causal effects (including null effects), then positive associations become “interesting” and null or negative associations are “uninformative.” A balanced interpretation of study results means that all results contribute information, including those that fail to support or are counter to the hypothesized association. Investigator bias is most often revealed in the abstract, results text, and in the summary of findings in the discussion; an essentially negative study with some possible exceptions is touted as a positive study based on any deviations from the null in a direction that supports an adverse effect of environmental toxicants. Recommendations for more informative studies Generating actionable knowledge of environmental toxicants and health12 requires identification of promising research opportunities, not just exploiting readily available data. Studies of populations with a well-defined exposure source, even if exposure is measured imperfectly, have notably different strengths (and limitations) than studies of background levels of biomarkers.2 The range of exposures is often higher, the basis for exposure contrasts is clear, sources of potential confounding may be more amenable to control, and the implications for potential interventions are obvious. In that setting, exposure biomarkers may help to validate exposure contrasts even if they are not the primary exposure indicator. Evaluating the health relevance of small, subclinical effects may benefit from considering evidence from other disciplines, including toxicology, clinical research, and mechanistic studies to help determine whether the observed effects are, in fact, pathological in nature. For example, suitable animal models for assessing fetal growth or endocrine disruption may be applicable to judging whether subclinical biological changes observed in epidemiologic studies have relevance to clinically consequential health outcomes. Triangulation of evidence5 may help to determine whether subclinical effects observed in epidemiologic studies have important health consequences. Toxicological or mechanistic research may also help to circumvent the challenge of distinguishing between very small effects and no effect. Identifying more potent forms of exposure, more sensitive health endpoints, or more susceptible populations would produce more informative epidemiologic studies. Well-reasoned hypotheses suggesting where stronger effects should be found under a causal hypothesis are informative regardless of whether the anticipated larger effects occur. Finding that stronger associations are not found where they would be expected under a causal hypothesis provides meaningful evidence against a causal effect. Interpreting results from multiple studies can be much more informative than simply generating a pooled measure of association as in routine meta-analysis or tallying the number of positive studies. Studies with varying features that are expected to bear on validity, such as quality of exposure or health outcome assessment or susceptibility to confounding, allow for informative contrasts across studies.13 If studies with methodologic features that should yield stronger associations under a causal hypothesis do so, a causal effect is supported, and if they fail to do so, the plausibility of a causal effect is diminished. Finally, the negative consequences of exaggerating evidence of health risks warrant consideration, no less than understating evidence of harm and missing an opportunity for beneficial action. Overinterpreting the evidence for causal effects of environmental toxicants and health outcomes and claiming they call for regulatory or behavioral change based on precarious evidence reduces the credibility of environmental epidemiology and weakens lines of research that warrant attention and intervention. In the face of declining public trust in science generally14 and political exploitation and encouragement of that distrust, we all need to be more cautious in distinguishing between actionable evidence and inconsequential exploitation of data. More than ever, there is a need to address pressing environmental health concerns with rigorous science that will ultimately advance public health. Studies in which the link to environmental exposures is tenuous, the health measure is of unclear importance, and the presence of a causal effect is ambiguous provide little if any progress towards beneficial public health action. Conflicts of interest statement The authors declare that they have no conflicts of interest with regard to the content of this report.

  PublisherDOI

Recent grants

• NIH Grant R01ES019955

  NIH · $1.3M · 2015

• NIH Grant R21CA066201

  NIH · $290k · 1998

• Effect Of Iatrogenic Delivery at 34-38 Weeks' Gestation on Pregnancy Outcome

  NIH · $1.8M · 2014–2019

• NIH Grant R01HD037584

  NIH · $3.1M · 2006

• NIH Grant R01ES007128

  NIH · $1.2M · 1998

Frequent coauthors

• Erika F. Werner

  Tufts University

  205 shared

• Valery A. Danilack

  Yale University

  187 shared

• Katherine E. Hartmann

  University of Pennsylvania

  186 shared

• Heather S. Lipkind

  Weill Cornell Medicine

  138 shared

• Phinnara Has

  Brown University

  115 shared

• John M. Thorp

  114 shared

• Teresa Janević

  Columbia University

  109 shared

• Edmund F. Funai

  Yale University

  106 shared

Education

• Ph.D.

  University of California at Los Angeles

  2010

• M.A.

  Boston University

  2005

• B.A.

  Boston University

  2005

Awards & honors

• Elected Fellow of the American Statistical Association 2023
• Health Policy Statistics Section Mid-Career Achievement Awar…
• Rothman Epidemiology Prize for the best paper published in E…
• International Society for Bayesian Analysis Mitchell Prize (…