Tyler J VanderWeele

Harvard University, Cambridge, Massachusetts, United States

Are you Tyler J VanderWeele?

Claim your profile

Publications (197)861.3 Total impact

  • Source
    Peng Ding · Tyler J. VanderWeele ·
    [Show abstract] [Hide abstract]
    ABSTRACT: If an effect measure is more homogeneous than others, then its value is more likely to be stable across different subgroups or subpopulations. Therefore, it is of great importance to find a more homogeneous effect measure that allows for transportability of research results. For a binary outcome, applied researchers often claim that the risk difference is more heterogeneous than the risk ratio or odds ratio, because they find, based on evidence from surveys of meta-analyses, that the null hypotheses of homogeneity are rejected more often for the risk difference than for the risk ratio and odds ratio. However, the evidence for these claims are far from satisfactory, because of different statistical powers of the homogeneity tests under different effect scales. For binary treatment, covariate and outcome, we theoretically quantify the homogeneity of different effect scales. Because the four outcome probabilities lie in a three dimensional space of the four dimensional space when homogeneity holds for any effect scale, we compute the volumes of these three dimensional spaces to compare the relative homogeneity of the risk difference, risk ratio, and odds ratio. We demonstrate that the homogeneity space for the risk difference has the smallest volume, and the homogeneity space for the odds ratio has the largest volume, providing further evidence for the previous claim that the risk difference is more heterogeneous than the risk ratio and odds ratio.
  • L. Valeri · J. T. Chen · X. Garcia-Albeniz · N. Krieger · T. J. VanderWeele · B. A. Coull ·

    Cancer Epidemiology Biomarkers & Prevention 10/2015; DOI:10.1158/1055-9965.EPI-15-0456 · 4.13 Impact Factor
  • Source
    M Arfan Ikram · Tyler J VanderWeele ·
    [Show abstract] [Hide abstract]
    ABSTRACT: Understanding of causal pathways in epidemiology involves the concepts of direct and indirect effects. Recently, causal mediation analysis has been formalized to quantify these direct and indirect effects in the presence of exposure-mediator interaction and even allows for four-way decomposition of the total effect: controlled direct effect, reference interaction, mediated interaction, pure indirect effect. Whereas the other three effects can be intuitively conceptualized, mediated interaction is often considered a nuisance in statistical analysis. In this paper, we focus on mediated interaction and contrast it against pure mediation. We also propose a clinical and biological interpretation of mediated interaction using three hypothetical examples. With these examples we aim to make researchers aware that mediated interaction can actually provide important clinical and biological information.
    European Journal of Epidemiology 10/2015; DOI:10.1007/s10654-015-0087-5 · 5.34 Impact Factor
  • M-A C Bind · T J Vanderweele · B A Coull · J D Schwartz ·
    [Show abstract] [Hide abstract]
    ABSTRACT: Mediation analysis is a valuable approach to examine pathways in epidemiological research. Prospective cohort studies are often conducted to study biological mechanisms and often collect longitudinal measurements on each participant. Mediation formulae for longitudinal data have been developed. Here, we formalize the natural direct and indirect effects using a causal framework with potential outcomes that allows for an interaction between the exposure and the mediator. To allow different types of longitudinal measures of the mediator and outcome, we assume two generalized mixed-effects models for both the mediator and the outcome. The model for the mediator has subject-specific random intercepts and random exposure slopes for each cluster, and the outcome model has random intercepts and random slopes for the exposure, the mediator, and their interaction. We also expand our approach to settings with multiple mediators and derive the mediated effects, jointly through all mediators. Our method requires the absence of time-varying confounding with respect to the exposure and the mediator. This assumption is achieved in settings with exogenous exposure and mediator, especially when exposure and mediator are not affected by variables measured at earlier time points. We apply the methodology to data from the Normative Aging Study and estimate the direct and indirect effects, via DNA methylation, of air pollution, and temperature on intercellular adhesion molecule 1 (ICAM-1) protein levels. Our results suggest that air pollution and temperature have a direct effect on ICAM-1 protein levels (i.e. not through a change in ICAM-1 DNA methylation) and that temperature has an indirect effect via a change in ICAM-1 DNA methylation. © The Author 2015. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.
    Biostatistics 08/2015; DOI:10.1093/biostatistics/kxv029 · 2.65 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Background: Low birth weight and unhealthy lifestyles in adulthood have been independently associated with an elevated risk of hypertension. However, no study has examined the joint effects of these factors on incidence of hypertension. Methods: We followed 52,114 women from the Nurses' Health Study II without hypercholesterolemia, diabetes, cardiovascular disease, cancer, prehypertension, and hypertension at baseline (1991-2011). Women born preterm, of a multiple pregnancy, or who were missing birth weight data were excluded. Unhealthy adulthood lifestyle was defined by compiling status scores of body mass index, physical activity, alcohol consumption, the Dietary Approaches to Stop Hypertension diet, and the use of non-narcotic analgesics. Results: We documented 12,588 incident cases of hypertension during 20 years of follow-up. The risk of hypertension associated with a combination of low birth weight at term and unhealthy lifestyle factors (RR, 1.95; 95 % CI, 1.83-2.07) was more than the addition of the risk associated with each individual factor, indicating a significant interaction on an additive scale (P interaction <0.001). The proportions of the association attributable to lower term birth weight alone, unhealthy lifestyle alone, and their joint effect were 23.9 % (95 % CI, 16.6-31.2), 63.7 % (95 % CI, 60.4-66.9), and 12.5 % (95 % CI, 9.87-15.0), respectively. The population-attributable-risk for the combined adulthood unhealthy lifestyle and low birth weight at term was 66.3 % (95 % CI, 56.9-74.0). Conclusion: The majority of cases of hypertension could be prevented by the adoption of a healthier lifestyle, though some cases may depend on simultaneous improvement of both prenatal and postnatal factors.
    BMC Medicine 07/2015; 13:175. DOI:10.1186/s12916-015-0409-1 · 7.25 Impact Factor
  • Charlie Poole · Ian Shrier · Tyler J VanderWeele ·
    [Show abstract] [Hide abstract]
    ABSTRACT: There are claims in the literature that the risk difference is a more heterogeneous measure than the odds ratio or risk ratio. These claims are based on surveys of meta-analyses showing that tests reject the null hypothesis of homogeneity more often for the risk difference than for the ratio measures. Discussions of this point have neglected the fact that homogeneity tests can have different levels of statistical power (i.e., different probabilities of rejecting the null when it is false) across different scales. We give hypothetical examples in which there is arguably equal heterogeneity across risk difference and odds ratio measures but in which the risk difference homogeneity test rejects more often, and therefore has higher power, than the odds ratio homogeneity test. These examples suggest that current empirical evidence for the claim that the risk difference is more heterogeneous is not at present satisfactory. Further research could consider other approaches to empirical comparisons of the heterogeneity of the three measures.
    Epidemiology (Cambridge, Mass.) 07/2015; 26(5). DOI:10.1097/EDE.0000000000000354 · 6.20 Impact Factor
  • Source
    Peng Ding · Tyler VanderWeele ·
    [Show abstract] [Hide abstract]
    ABSTRACT: Unmeasured confounding may undermine the validity of causal inference with observational studies. Sensitivity analysis provides an attractive way to partially circumvent this issue by assessing the potential influence of unmeasured confounding on the causal conclusions. However, previous sensitivity analysis approaches often make strong and untestable assumptions such as having a confounder that is binary, or having no interaction between the effects of the exposure and the confounder on the outcome, or having only one confounder. Without imposing any assumptions on the confounder or confounders, we derive a bounding factor and a sharp inequality such that the sensitivity analysis parameters must satisfy the inequality if an unmeasured confounder is to explain away the observed effect estimate or reduce it to a particular level. Our approach is easy to implement and involves only two sensitivity parameters. Surprisingly, our bounding factor, which makes no simplifying assumptions, is no more conservative than a number of previous sensitivity analysis techniques that do make assumptions. Our new bounding factor implies not only the traditional Cornfield conditions that both the relative risk of the exposure on the confounder and that of the confounder on the outcome must satisfy, but also a high threshold that the maximum of these relative risks must satisfy. Furthermore, this new bounding factor can be viewed as a measure of the strength of confounding between the exposure and the outcome induced by a confounder.
  • [Show abstract] [Hide abstract]
    ABSTRACT: Objectives To prospectively assess the joint association of birth weight and established lifestyle risk factors in adulthood with incident type 2 diabetes and to quantitatively decompose the attributing effects to birth weight only, to adulthood lifestyle only, and to their interaction.Design Prospective cohort study.Setting Health Professionals Follow-up Study (1986-2010), Nurses’ Health Study (1980-2010), and Nurses’ Health Study II (1991-2011).Participants 149 794 men and women without diabetes, cardiovascular disease, or cancer at baseline.Main outcome measure Incident cases of type 2 diabetes, identified through self report and validated by a supplementary questionnaire. Unhealthy lifestyle was defined on the basis of body mass index, smoking, physical activity, alcohol consumption, and the alternate healthy eating index.Results During 20-30 years of follow-up, 11 709 new cases of type 2 diabetes were documented. The multivariate adjusted relative risk of type 2 diabetes was 1.45 (95% confidence interval 1.32 to 1.59) per kg lower birth weight and 2.10 (1.71 to 2.58) per unhealthy lifestyle factor. The relative risk of type 2 diabetes associated with a combination of per kg lower birth weight and per unhealthy lifestyle factor was 2.86 (2.26 to 3.63), which was more than the addition of the risk associated with each individual factor, indicating a significant interaction on an additive scale (P for interaction<0.001). The attributable proportions of joint effect were 22% (95% confidence interval 18.3% to 26.4%) to lower birth weight alone, 59% (57.1% to 61.5%) to unhealthy lifestyle alone, and 18% (13.9% to 21.3%) to their interaction.Conclusion Most cases of type 2 diabetes could be prevented by the adoption of a healthier lifestyle, but simultaneous improvement of both prenatal and postnatal factors could further prevent additional cases.
    BMJ: British medical journal 07/2015; 351:h3672. DOI:10.1136/bmj.h3672 · 16.30 Impact Factor
  • John W Jackson · Tyler J VanderWeele · Deborah Blacker · Sebastian Schneeweiss ·
    [Show abstract] [Hide abstract]
    ABSTRACT: Observational studies of older adults showed higher mortality for first-generation antipsychotics than their second--generation counterparts, which led to US Food and Drug Administration warnings, but the actual mechanisms involved remain unclear. A cohort of 9,060 initiators of first-generation antipsychotics and 17,137 of second-generation antipsychotics enrolled in New Jersey and Pennsylvania Medicare were followed for 180 days. Medical events were assessed using diagnostic and procedure codes on inpatient billing claims. For the individual and joint set of medical events (mediators), we estimated the total, direct, and indirect effects of antipsychotic type (first versus second generation) on mortality on the risk ratio scale and the proportion mediated on the risk difference scale, obtaining 95% confidence intervals through bootstrapping. We performed bias analyses for false-negative mediator misclassification in claims data, with sensitivity ranging from 0.25 to 0.75. There were 3,199 deaths (outcomes), 862 cardiovascular events, 675 infectious events, and 491 hip fractures (potential mediators). Mortality was higher for first- than second-generation antipsychotic initiators (adjusted risk ratio: 1.14; 95% confidence interval: 1.06, 1.22). In naïve analyses, that ignored potential misclassification, less than 4% of this difference was explained by any particular medical event. In bias analyses, the proportion mediated ranged from 6% to 16% for stroke, 3% to 9% for ventricular arrhythmia, 3% to 11% for myocardial infarction, 0% venous thromboembolism, 3% to 9% for pneumonia, 0% to 1% for other bacterial infection, and 1% to 3% for hip fracture. Acute cardiovascular events and pneumonia may explain part of the mortality difference between first- and second-generation antipsychotic initiators in this analysis.
    Epidemiology (Cambridge, Mass.) 05/2015; 26(5). DOI:10.1097/EDE.0000000000000321 · 6.20 Impact Factor
  • Zhichao Jiang · Tyler J VanderWeele ·
    [Show abstract] [Hide abstract]
    ABSTRACT: Assessment of indirect effects is useful for epidemiologists interested in understanding the mechanisms of exposure-outcome relationships. A traditional way of estimating indirect effects is to use the "difference method," which is based on regression analysis in which one adds a possible mediator to the regression model and examines whether the coefficient for the exposure changes. The difference method has been criticized for lacking a causal interpretation when it is used with logistic regression. In this article, we use the counterfactual framework to define the natural indirect effect (NIE) and assess the relationship between the NIE and the difference method. We show that under appropriate assumptions, the difference method consistently estimates the NIE for continuous outcomes and is always conservative for binary outcomes. Thus, the difference method can be used to provide evidence for the presence of mediation but not for the absence of mediation. © The Author 2015. Published by Oxford University Press on behalf of the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.
    American journal of epidemiology 05/2015; 182(2). DOI:10.1093/aje/kwv059 · 5.23 Impact Factor
  • Zhichao Jiang · Tyler J VanderWeele ·

    American journal of epidemiology 05/2015; 182(2). DOI:10.1093/aje/kwv058 · 5.23 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Health care providers' lack of education on spiritual care is a significant barrier to the integration of spiritual care into health care services. The study objective was to describe the training program, Clinical Pastoral Education for Healthcare Providers (CPE-HP) and evaluate its impact on providers' spiritual care skills. Fifty CPE-HP participants completed self-report surveys at baseline and posttraining measuring frequency of and confidence in providing religious/spiritual (R/S) care. Four domains were assessed: (1) ability and (2) frequency of R/S care provision; (3) comfort using religious language; and (4) confidence in providing R/S care. At baseline, participants rated their ability to provide R/S care and comfort with religious language as "fair." In the previous two weeks, they reported approximately two R/S patient conversations, initiated R/S conversations less than twice, and prayed with patients less than once. Posttraining participants' reported ability to provide spiritual care increased by 33% (p<0.001). Their comfort using religious language improved by 29% (p<0.001), and frequency of R/S care increased 75% (p<0.001). Participants reported having 61% more (p<0.001) R/S conversations and more frequent prayer with patients (95% increase; p<0.001). Confidence in providing spiritual care improved by 36% overall, by 20% (p<0.001) with religiously concordant patients, and by 43% (p<0.001) with religiously discordant patients. This study suggests that CPE-HP is an effective approach for training health care providers in spiritual care. Dissemination of this training may improve integration of spiritual care into health care, thereby strengthening comprehensive patient-centered care.
    Journal of palliative medicine 04/2015; 18(5). DOI:10.1089/jpm.2014.0306 · 1.91 Impact Factor
  • L. Valeri · T.J. Vanderweele ·

  • Tyler J VanderWeele · Eric J Tchetgen Tchetgen ·

    Epidemiology (Cambridge, Mass.) 02/2015; 26(3). DOI:10.1097/EDE.0000000000000263 · 6.20 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: The purpose of this study is to determine how oncology nurses and physicians view their role in providing spiritual care (SC), factors influencing this perception, and how this belief affects SC provision. This is a survey-based, multisite study conducted from October 2008 to January 2009. All oncology physicians and nurses caring for advanced cancer patients at four Boston, MA cancer centers were invited to participate; 339 participated (response rate = 63 %). Nurses were more likely than physicians to report that it is the role of medical practitioners to provide SC, including for doctors (69 vs. 49 %, p < 0.001), nurses (73 vs. 49 %, p < 0.001), and social workers (81 vs. 63 %, p = 0.001). Among nurses, older age was the only variable that was predictive of this belief [adjusted odds ratio (AOR) 1.08; 1.01-1.16, p = 0.02]. For nurses, role perception was not related to actual SC provision to patients. In contrast, physicians' role perceptions were influenced by their intrinsic religiosity (AOR, 1.44; 95 % CI, 1.09-1.89; p = 0.01) and spirituality (AOR, 6.41; 95 % CI, 2.31-17.73, p < 0.001). Furthermore, physicians who perceive themselves as having a role in SC provision reported greater SC provision to their last advanced cancer patients seen in clinic, 69 % compared to 31 %, p < 0.001. Nurses are more likely than physicians to perceive medical practitioners as having a role in SC provision. Physicians' perceptions of their role in SC provision are influenced by their religious/spiritual characteristics and are predictive of actual SC provision to patients. Spiritual care training that includes improved understanding of clinicians' appropriate role in SC provision to severely ill patients may lead to increased SC provision.
    Journal of Pain and Symptom Management 01/2015; 49(2). DOI:10.1007/s00520-015-2611-2 · 2.80 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Epidemiologic data on genetic susceptibility to cardiovascular effects of arsenic exposure from drinking water are limited. We investigated whether the association between well-water arsenic and cardiovascular disease (CVD) differed by 170 single nucleotide polymorphisms (SNPs) in 17 genes related to arsenic metabolism, oxidative stress, inflammation, and endothelial dysfunction. We conducted a prospective case-cohort study nested in the Health Effects of Arsenic Longitudinal Study, with a random subcohort of 1,375 subjects and 447 incident fatal and nonfatal cases of CVD. Well-water arsenic was measured in 2000 at baseline. The CVD cases, 56 of which occurred in the subcohort, included 238 coronary heart disease cases, 165 stroke cases, and 44 deaths due to other CVD identified during follow-up from 2000 to 2012. Of the 170 SNPs tested, multiplicative interactions between well-water arsenic and two SNPs, rs281432 in ICAM1 (Padj = 0.0002) and rs3176867 in VCAM1 (Padj = 0.035), were significant for CVD after adjustment for multiple testing. Compared with those with GC or CC genotype in rs281432 and lower well-water arsenic, the adjusted hazard ratio (aHR) for CVD was 1.82 (95% CI: 1.31, 2.54) for a 1-SD increase in well-water arsenic combined with the GG genotype, which was greater than expected given aHRs of 1.08 and 0.96 for separate effects of arsenic and the genotype alone, respectively. Similarly, the joint aHR for arsenic and the rs3176867 CC genotype was 1.34 (95% CI: 0.95, 1.87), greater than expected given aHRs for their separate effects of 1.02 and 0.84, respectively. Associations between CVD and arsenic exposure may be modified by genetic variants related to endothelial dysfunction.
    Environmental Health Perspectives 01/2015; 123(5). DOI:10.1289/ehp.1307883 · 7.98 Impact Factor
  • Etsuji Suzuki · Tyler J VanderWeele ·
    [Show abstract] [Hide abstract]
    ABSTRACT: Under Bateson’s original conception, the term “epistasis” is used to describe the situation in which the effect of a genetic factor at one locus is masked by a variant at another locus. Epistasis in the sense of masking has been termed “compositional epistasis.” In general, statistical tests for interaction are of limited use in detecting compositional epistasis. Using recently developed epidemiological methods, however, it has been shown that there are relations between empirical data patterns and compositional epistasis. These relations can sometimes be exploited to empirically test for certain forms of compositional epistasis, by using alternative nonstandard tests for interaction. Using the counterfactual framework, we show conditions that can be empirically tested to determine whether there are individuals whose phenotype response patterns manifest epistasis in the sense of masking. Only under some very strong assumptions would tests for standard statistical interactions correspond to compositional epistasis. Even without such strong assumptions, however, one can still test whether there are individuals of phenotype response type representing compositional epistasis. The empirical conditions are quite strong, but the conclusions which tests of these conditions allow may be of interest in a wide range of studies. This chapter highlights that epidemiologic perspectives can be used to shed light on underlying mechanisms at the genetic, molecular, and cellular levels.
    Methods in molecular biology (Clifton, N.J.) 01/2015; 1253:197-216. DOI:10.1007/978-1-4939-2155-3_11 · 1.29 Impact Factor
  • Zhichao Jiang · Tyler J VanderWeele ·

    Epidemiology (Cambridge, Mass.) 01/2015; 26(1):e8-9. DOI:10.1097/EDE.0000000000000204 · 6.20 Impact Factor
  • Zhichao Jiang · Tyler J VanderWeele ·

    American Journal of Epidemiology 12/2014; 181(1). DOI:10.1093/aje/kwu351 · 5.23 Impact Factor
  • Linda Valeri · Xihong Lin · Tyler J VanderWeele ·
    [Show abstract] [Hide abstract]
    ABSTRACT: Mediation analysis is a popular approach to examine the extent to which the effect of an exposure on an outcome is through an intermediate variable (mediator) and the extent to which the effect is direct. When the mediator is mis-measured, the validity of mediation analysis can be severely undermined. In this paper, we first study the bias of classical, non-differential measurement error on a continuous mediator in the estimation of direct and indirect causal effects in generalized linear models when the outcome is either continuous or discrete and exposure-mediator interaction may be present. Our theoretical results as well as a numerical study demonstrate that in the presence of non-linearities, the bias of naive estimators for direct and indirect effects that ignore measurement error can take unintuitive directions. We then develop methods to correct for measurement error. Three correction approaches using method of moments, regression calibration, and SIMEX are compared. We apply the proposed method to the Massachusetts General Hospital lung cancer study to evaluate the effect of genetic variants mediated through smoking on lung cancer risk. Copyright © 2014 John Wiley & Sons, Ltd.
    Statistics in Medicine 12/2014; 33(28). DOI:10.1002/sim.6295 · 1.83 Impact Factor

Publication Stats

3k Citations
861.30 Total Impact Points


  • 2009-2015
    • Harvard University
      Cambridge, Massachusetts, United States
    • Northwestern University
      • Department of Preventive Medicine
      Evanston, IL, United States
    • University of Illinois at Chicago
      Chicago, Illinois, United States
  • 2006-2014
    • Harvard Medical School
      • Department of Medicine
      Boston, Massachusetts, United States
  • 2011
    • Princeton University
      Princeton, New Jersey, United States
    • Eunice Kennedy Shriver National Institute of Child Health and Human Development
      Maryland, United States
    • Boston Medical Center
      Boston, Massachusetts, United States
    • Kinki University
      Ōsaka, Ōsaka, Japan
  • 2010-2011
    • Massachusetts Department of Public Health
      Boston, Massachusetts, United States
  • 2006-2010
    • University of Chicago
      • Department of Health Studies
      Chicago, Illinois, United States