A1102
Title: Quantile mediation analysis with convoluted confounding effects via deep neural networks
Authors: Shuoyang Wang - Yale University (United States) [presenting]
Runze Li - The Pennsylvania State University (United States)
Yuan Huang - Yale University (United States)
Abstract: Traditional mediation analysis methods have been limited to dealing with only a few mediators, and they face challenges when the number of mediators is high-dimensional. In practice, these challenges can be compounded by outliers and the complex relationships introduced by confounders. A novel quantile-based partially linear mediation analysis method (QMDNN) that can handle high-dimensional mediators is proposed to address these issues. Deep neural network techniques to model complex nonlinear relationships in confounders are introduced. Unlike most existing works focusing on mediator selection, estimation and inference on mediation effects are emphasized. Theoretical analysis shows that the proposed procedure controls type I error rates for hypothesis testing on mediation effects. When the dimension of mediators is high, the proposed method consistently selects important features in the outcome model. Numerical studies show that the proposed method outperforms existing approaches under a variety of settings, demonstrating the versatility and reliability of QMDNN as a modelling tool for complex data. The application of QMDNN to study DNA methylation's mediation effect of childhood trauma on cortisol stress reactivity reveals previously undiscovered relationships by providing a comprehensive profile of the relationship at various quantiles.
