Ander Wilson1, Corwin M Zigler2, Chirag J Patel3, Francesca Dominici2. 1. Department of Statistics, Colorado State University, Fort Collins, Colorado, U.S.A. 2. Department of Biostatistics, Harvard T. H. Chan School of Public Health, Boston, Massachusetts, U.S.A. 3. Department of Biomedical Informatics, Harvard Medical School, Boston, Massachusetts, U.S.A.
Abstract
In environmental and nutritional epidemiology and in many other fields, there is increasing interest in estimating the effect of simultaneous exposure to several agents (e.g., multiple nutrients, pesticides, or air pollutants) on a health outcome. We consider estimating the effect of a multivariate exposure that includes several continuous agents and their interactions-on an outcome, when the true confounding variables are an unknown subset of a potentially large (relative to sample size) set of measured covariates. Our approach is rooted in the ideas of Bayesian model averaging: the exposure effect is estimated as a weighted average of the estimated exposure effects obtained under several linear regression models that include different sets of the potential confounders. We introduce a data-driven prior that assigns to the likely confounders a higher probability of being included into the regression model. We show that our approach can also be formulated as a penalized likelihood formulation with an interpretable tuning parameter. Through a simulation study, we demonstrate that the proposed approach identifies parsimonious models that are fully adjusted for observed confounding and estimates the multivariate exposure effect with smaller mean squared error compared to several alternatives. We apply the method to an Environmental Wide Association Study using National Heath and Nutrition Examination Survey to estimate the effect of mixtures of nutrients and pesticides on lipid levels.
In environmentn class="Chemical">al anpan>d nutritionpan>pan> class="Chemical">al epidemiology and in many other fields, there is increasing interest in estimating the effect of simultaneous exposure to several agents (e.g., multiple nutrients, pesticides, or air pollutants) on a health outcome. We consider estimating the effect of a multivariate exposure that includes several continuous agents and their interactions-on an outcome, when the true confounding variables are an unknown subset of a potentially large (relative to sample size) set of measured covariates. Our approach is rooted in the ideas of Bayesian model averaging: the exposure effect is estimated as a weighted average of the estimated exposure effects obtained under several linear regression models that include different sets of the potential confounders. We introduce a data-driven prior that assigns to the likely confounders a higher probability of being included into the regression model. We show that our approach can also be formulated as a penalized likelihood formulation with an interpretable tuning parameter. Through a simulation study, we demonstrate that the proposed approach identifies parsimonious models that are fully adjusted for observed confounding and estimates the multivariate exposure effect with smaller mean squared error compared to severalalternatives. We apply the method to an Environmental Wide Association Study using National Heath and Nutrition Examination Survey to estimate the effect of mixtures of nutrients and pesticides on lipid levels.
Authors: M Alan Brookhart; Sebastian Schneeweiss; Kenneth J Rothman; Robert J Glynn; Jerry Avorn; Til Stürmer Journal: Am J Epidemiol Date: 2006-04-19 Impact factor: 4.897
Authors: Chirag J Patel; David H Rehkopf; John T Leppert; Walter M Bortz; Mark R Cullen; Glenn M Chertow; John Pa Ioannidis Journal: Int J Epidemiol Date: 2013-12-16 Impact factor: 7.196
Authors: Emily Goren; Chong Wang; Zhulin He; Amy M Sheflin; Dawn Chiniquy; Jessica E Prenni; Susannah Tringe; Daniel P Schachtman; Peng Liu Journal: BMC Bioinformatics Date: 2021-07-06 Impact factor: 3.169