B1154
Title: Causal inference with high-dimensional outcome variables
Authors: Ping-Shou Zhong - University of Illinois at Chicago (United States) [presenting]
Nurlan Abdukadyrov - University of Illinois at Chicago (United States)
Wei Biao Wu - University of Chicago (United States)
Xiaohong Joe Zhou - University of Illinois at Chicago (United States)
Abstract: In genomic genetic and neuroimaging studies, biomarker identification often involves detecting the changes in a large number of biomarker candidates caused by the presence of certain diseases. The existing causal inference methods focus almost exclusively on low-dimensional outcome variables. These methods are not applicable to biomarker identification with a large number of biomarker candidates. A causal inference procedure is developed for high-dimensional outcome variables when the dimension of outcome variables is larger than the sample size. The proposed method is doubly robust to the misspecification of propensity score function or outcome regression models. The asymptotic distributions of the proposed statistic are established. The asymptotic distributions change according to different misspecifications of propensity score models or outcome regression models. A bootstrap procedure is developed to estimate the asymptotic variance adaptively. Numerical simulation studies are used to evaluate the finite sample performance of the proposed methods. The procedure is also applied to a diffusion MRI data set to identify regions of interest that may be used as biomarkers for Parkinson's disease.