Causal Mediation Analysis on Failure Time Outcome Without Sequential Ignorability

• Activity Number: 609
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Social Statistics Section
• Author(s): Cheng Zheng*+ and Xiao-Hua Andrew Zhou and Ross Prentice
• Companies: University of Washington and University of Washington and University of Washington
• Keywords: Causal Inference ; Inverse Censoring Probability Weighting ; Generalized Estimating Equation ; Additive Hazard Model ; Regression Calibration

Abstract

Mediation analysis is an important issue in social and biomedical sciences as it helps to understand why an intervention works. Most mediation analysis works on binary or continuous outcome. Recently, method for causal mediation analysis for failure time outcome was proposed relying on the assumption of sequential ignorability. To allow potential confounders between the mediator and the outcome, we proposed estimators for additive hazard model for mediation analysis with survival outcome. The estimators were compared with the regression method assuming sequential ignorability in numerical studies. As an estimating equation based method, our method allows time varying effect and can handle correlated data and error prone mediator with standard technique. To handle the case where mediator is measured with error, we explored the performance of regression calibration type strategy. Our method allows time varying effect and can be applied to other applications such as adherence and surrogate evaluation.