Optimal G-Estimation Mediation Analyses Under Departure from Sequential Ignorability

• Activity Number: 451
• Type: Topic Contributed
• Date/Time: Wednesday, August 4, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Rongmei Zhang*+ and Marshall M. Joffe and Thomas R. Ten Have
• Companies: University of Pennsylvania and University of Pennsylvania and University of Pennsylvania School of Medicine
• Keywords: causal inference ; structural nested model ; randomized trials ; post-randomization ; weighting

Abstract

We address several questions relating to the use of standard and causal approaches to estimating the joint and mediation effects of randomized treatments and a post-randomization variable. Unlike standard analyses, our approach does not assume that the post-randomization variable is also randomly assigned to individuals in addition to the randomized baseline intervention (i.e., sequential ignorability). G-estimation is used to obtain consistent estimators for parameters in a structural nested distribution model in which we allow interaction between the randomized intervention and the post-randomization factor. We show how to obtain the efficient estimators. We conduct simulations to compare the performance between standard and causal approaches, and between our optimal approach and simpler g-estimators. We apply our approach to a randomized cognitive therapy trial.