A1029
Title: Bayesian additive regression trees model for high-dimensional potential confounders
Authors: Chanmin Kim - SungKyunKwan University (Korea, South) [presenting]
Abstract: A solution to the increasing challenge faced in the analysis of observational studies is offered, which involves identifying the covariates required to establish the assumption of ignorable treatment assignment for causal effect estimation. The proposal adopts a Bayesian nonparametric approach that addresses this challenge in three ways. First, it prioritizes the inclusion of adjustment variables based on established confounder selection principles. Second, it enables estimating causal effects by accounting for complex relationships among confounders, exposures, and outcomes. Finally, it provides causal estimates that consider the uncertainty in the confounding nature. The method involves using multiple Bayesian Additive Regression Tree models that share a prior distribution, accumulating posterior selection probability to covariates associated with both the exposure and the outcome of interest. Several simulation studies demonstrate that the proposed method performs well relative to other similar methods in various scenarios. The approach is applied to examine the causal effect of SO2 emissions from coal-fired power plants on ambient air pollution concentrations, providing strong evidence of the causal relationship between SO2 emissions and ambient particulate pollution over adjacent years.