Methods for Confounding Adjustment and High-Dimensional Environmental Exposures

Abstract

Environmental exposures have complex multivariate relationships with one another and with geographic, anthropogenic, social, and physiological factors. This dissertation comprises methods for addressing the confounding and high-dimensional challenges of environmental exposures in cohort studies. We consider three dierent settings for improving statistical inference about associations between exposures and health eects using these multivariate relationships. First we present a method for clustering multi-pollutant observations in the context of an air pollution epidemiology cohort, where exposure must be predicted at subject locations. We then present a method for shrinkage estimation, with particular focus on small sample benet in the presence of many confounders. Third, we present methods for adjusting for unmeasured spatial confounding in analyses with environmental exposures. We apply each method to analyses of cardiovascular outcomes in a cohort study.