Mediation Analysis With Exposure–Mediator Interaction and Covariate Measurement Error Under the Additive Hazards Model

Abstract

Causal mediation analysis is a useful tool to examine how an exposure variable causally affects an outcome variable through an intermediate variable. In recent years, there is increasing research interest in mediation analysis with survival data. The existing literature usually requires accurate measurements of the mediator and the confounders, which is infeasible in many biomedical and social science studies. Ignoring measurement errors may lead to misleading inference results. Furthermore, the current identification results of causal effects under the additive hazards model are limited to the scenario with no exposure–mediator interaction, which can be unappealing in mediation analysis. In this paper, we derive the identification results of direct and indirect effects under the additive hazards model in the presence of exposure–mediator interaction. Furthermore, we propose a corrected approach to adjust for the impact of measurement error in the mediator and the confounders and obtain consistent estimations of the direct and indirect effects. The performance of the proposed method is studied in simulation studies and a real data study.