A Bayesian Causal Mediation Analysis for Multilevel Genome-Wide Epigenetic Data with High-Dimensional Mediators

Date of Award

4-1-2023

Degree Name

Doctor of Philosophy

Department

Statistics

First Advisor

Duy Ngo, Ph.D.

Second Advisor

Joshua Naranjo, Ph.D.

Third Advisor

Hyun Bin Kang, Ph.D.

Fourth Advisor

Bilinda Straight, Ph.D.

Keywords

Bayesian variable selection, epigenetic study, high-dimensional mediation analysis, multilevel data modeling

Abstract

Causal mediation analysis has increasingly become a popular practice in various clinical trials and epidemiological applications to evaluate whether an intermediate variable is on the pathway between the exposure of interest and a response. Previous mediation analyses in the literature mainly focused on settings with a single or low– dimensional mediators and single–level data. In this article, we propose a Bayesian causal mediation analysis method that can handle our multilevel intergenerational epigenetic mechanisms study (IEMS) with high–dimensional mediators. Specifically, we develop a Bayesian hierarchical model for data with such complexity, and then employ the Bayesian spike–and–slab priors on the exposure–mediator–outcome effect pathway to identify active mediators involved in mediation. We derive the natural indirect and direct effects based on our hierarchical model and provide statistical inference based on Markov chain Monte Carlo (MCMC) methods. Simulation study demonstrates that our proposed Bayesian method outperforms a univariate mediation method in various scenarios. We further illustrate the utility of our method to IEMS to assess the causal mechanisms between maternal exposure to climate extremes and offspring’s growth outcomes through DNA methylation.

Access Setting

Dissertation-Abstract Only

Restricted to Campus until

4-1-2033

This document is currently not available here.

Share

COinS