CMStatistics 2021: Start Registration
View Submission - CMStatistics
Title: Simulating hypothetical interventions on multiple mediators: Extending methods and practical guidance Authors:  Margarita Moreno-Betancur - University of Melbourne and Murdoch Children's Research Institute (Australia) [presenting]
John B Carlin - Murdoch Childrens Research Institute and University of Melbourne (Australia)
Abstract: Many research questions concern the pathways presumed to mediate an association, particularly in epidemiological research. Invariably, the translational intent of such research is to inform potential intervention targets, but until recently mediation effect definitions did not acknowledge this interventional intent. Recent work proposed a novel framework conceptualising mediation effects by mapping to a hypothetical ``target'' randomised trial evaluating mediator interventions. This approach is particularly relevant for mediators that do not correspond to well-defined interventions, which arise frequently, and perhaps can only be addressed by considering hypothetical interventions that would shift the mediators' distributions. The approach proposes specifying a target trial to capture the research question in the form of (a measure of) the impact of shifting joint mediator distributions to user-specified distributions. These estimand assumptions are distinguished from identifiability assumptions, which are needed to emulate the effects with the observed data. By its nature, the approach is context-specific. Drawing on learnings from applications to several longitudinal cohort studies, some of which are already published, further developments of the method are presented with a focus on alternative approaches to effect definition according to the question in diverse contexts. A workflow that will assist researchers in applying the method in practice is described.