Simulation Study Evaluating Methods for Estimating the Health Effects of High-Dimensional Exposures (306587)Jennifer Bobb, Kaiser Permanente Washington Health Research Institute
*Qianqian Chen, University of Washington
Keywords: Environmental Statistics, High-Dimensional Data Analysis, Statistical Learning, Bayesian Analysis
Estimating the health effects of multi-pollutant mixtures is of increasing interest in environmental epidemiology. Recently, a new approach called Bayesian kernel machine regression (BKMR) has been developed to estimate the multivariable exposure-response function in a flexible and parsimonious way via a kernel function and allows for variable selection to be conducted on the exposures. However, to date BKMR has only been applied and developed in settings with a relatively modest number of exposures (<20). In this project, we conduct simulation studies to evaluate the performance of BKMR when there are greater number of exposures. We generate exposure data using a realistic correlation structure from an existing exposome study, and present different scenarios varying in their ratios of number of exposures that are truly associated with the outcome to the number of the observations in the data. Additionally, we may also compare high-dimensional BKMR to other approaches such as LASSO and Bayesian Additive Regression Trees (BART). The performance of the methods is evaluated in terms of the ability to estimate the exposure-response function into correctly identified exposures.