CMStatistics 2021: Start Registration
View Submission - CMStatistics
Title: A doubly robust joint modelling approach of multiple uncausally correlated mediators Authors:  Lijia Wang - University of Waterloo (Canada) [presenting]
Yeying Zhu - University of Waterloo (Canada)
Richard Cook - University of Waterloo (Canada)
Abstract: Causal mediation analysis aims at disentangling the effects of a treatment on an outcome via a variety of paths through either intermediate variables lied alongside the causal pathways (the mediators) or the treatment itself. Recently, mediation analysis on multiple mediators is attracting much attention because of inspirations from reality, where the relationship between the multiple mediators plays an important role when investigating the causal effects. Traditional studies focus on the scenario that the multiple mediators are causally sequentially related. We review and extend another new concept that the multiple mediators are uncausally related, which depicts the phenomenon that the multiple mediators are related given the baseline covariates but their correlation structure cannot be causally ordered or clearly identified. We further provide a copula-based joint modelling approach performing the causal mediation analysis of the multiple uncausally related mediators. A doubly robust approach is also proposed to tackle the model misspecification issue. Theoretical properties and simulation studies are also presented, with the theoretical standard error derived based on the sandwich formula. We finally apply the proposed method on a genetic psychiatric study dataset to identify the causal mediation effect of three DNA methylation loci on the pathway between childhood trauma and stress reactivity.