Howard Chang: Estimating Complex Exposure-Response Functions with Bayesian Additive Regression Trees

Estimating Complex Exposure-Response Functions with Bayesian Additive Regression Trees: Examples from Environmental Epidemiology

Epidemiologic studies are often interested in estimating health effects of multiple risk factors and examining multiple effect modifiers. Bayesian additive regression trees (BART) is a nonparametric approach that has gained popularity due to its computational efficiency, probabilistic framework and prediction performance.  These advantages have encouraged the use of BART as a data-driven tool to characterize complex exposure-response surfaces. In this presentation, we will present our recent work in applying BART to model joint effects of environmental exposures and explore effect heterogeneity. We will also highlight some challenges associated with interpreting risk estimates obtained from a “black box” prediction algorithm.

BIO: Howard Chang is a Professor in the Department of Biostatistics and Bioinformatics at Emory University Rollins School of Public Health. He conducts environmental health studies by analyzing large health databases (e.g., birth certificates, electronic health records and disease surveillance systems). He also develops statistical methods motivated by analytical challenges from these large population-based studies. His research interests include Bayesian methods, climate and health, and spatial epidemiology. He received his PhD in Biostatistics from Johns Hopkins University (2009) and is an elected ASA fellow (2024).