Title: Mediation analysis with high-dimensional mediators
Authors: Rhian Daniel - Cardiff University (United Kingdom) [presenting]
Abstract: In many modern biomedical applications, interest lies in decomposing the effect of an exposure, eg a genetic variant, on an outcome, eg cardiovascular disease, into its effect via a large number of mediators, eg blood protein and metabolite measures. Such an endeavour poses formidable methodological challenges. First, the mediators are likely to be highly-correlated according to an unknown causal structure, including unmeasured common causes of one mediator and another. Second, the identification of the usual natural path-specific effects in such a setting would rely on many ``cross-world independence'' assumptions, which are impossible to justify. Third, the usual parametric regression estimation approaches would rely on a huge number of (uncheckable, in practice) modelling assumptions. We propose that the first two problems be overcome by focusing on interventional multiple mediator effects and the third by using data-adaptive (machine learning) estimation. We will outline a few such possible approaches (including one based on targeted minimum loss-based estimation), compare their properties in a simulation study, and illustrate their use on a motivating application using data from the UCLEB consortium investigating the metabolomic pathways through which six common cardiovascular SNPs act.