CMStatistics 2021: Start Registration
View Submission - CMStatistics
B0196
Title: Nonlinear mediation analysis with high-dimensional mediators whose causal structure is unknown Authors:  Wen Wei Loh - Ghent University (Belgium) [presenting]
Beatrijs Moerkerke - Ghent University (Belgium)
Tom Loeys - Ghent University (Belgium)
Stijn Vansteelandt - Ghent University and London School of Hygiene and Tropical Medicine (Belgium)
Abstract: With multiple possible mediators on the causal pathway from treatment to an outcome, the focus is on the problem of decomposing the effects along multiple possible causal paths through each distinct mediator. Fine-grained decompositions under Pearl's path-specific effects framework necessitate stringent assumptions, such as correctly specifying the causal structure among the mediators, and no unobserved confounding among the mediators. In contrast, interventional direct and indirect effects for multiple mediators can be identified under much weaker conditions. However, current estimation approaches require (correctly) specifying a model for the joint mediator distribution, which can be difficult when there is a high-dimensional set of possibly continuous and non-continuous mediators. We avoid the need to model this distribution, by developing a definition of interventional effects previously suggested for longitudinal mediation. We propose a novel estimation strategy that uses non-parametric estimates of the (counterfactual) mediator distributions. Non-continuous outcomes can be accommodated using non-linear outcome models. Estimation proceeds via Monte Carlo integration. The procedure is used to assess the effect of a microRNA expression on the three-month mortality of brain cancer patients via expression values of multiple genes.