Dealing with complex problems of confounding in mediation analysis

Mediation analysis is frequently utilized in diverse scientific fields such as psychology, sociology and epidemiology, to develop insight into the causal mechanism whereby an exposure affects an outcome. It concerns the study of indirect effects of that exposure that are mediated through a given intermediate variable or mediator, and/or the study of the remaining direct effect. Despite its popularity, the traditional approach to mediation analysis proceeds in a predominantly heuristic fashion, which can largely be ascribed to the lack of precise definitions of direct and indirect effect in the traditional mediation analysis literatures. Moreover, problems of confounding bias have been largely ignored. James Robins, Sander Greenland and Judea Pearl laid the foundations for a rigorous approach towards mediation analysis, which is based on counterfactuals. They gave precise definitions of direct and indirect effect and elucidated the kind of data that must be collected in order to control for confounding bias. In addition, they provided generic ways to decompose a total effect into a direct and indirect effect that is not tied to a specific statistical model. In this presentation, after a brief review of some of these developments, I will concentrate on the partly unsolved methodological challenges that arise when confounders of the mediator-outcome association are affected by the exposure. In particular, I will present results on the the identification of (natural) direct and indirect effects in such settings, and on the estimation of (controlled) direct effects, thereby focussing on matched case-control studies and/or survival analysis.