Online Program

Mediation Analysis with Time Failure Outcome and Error Prone Mediator

Ross Prentice, Department of Biostatistics, University of Washington 
*Cheng Zheng, Department of Biostatistics, University of Washington 
Xiao-Hua Andrew Zhou, University of Washington 

Keywords: Causal Inference, Survival, Measurement Error, Regression Calibration, Expected Estimating Equation

Mediation analysis is an important part to understand why an intervention works. Most mediation analysis literature focused on either binary or continuous outcome. Recently, possible ways to define direct and indirect effects for causal mediation analysis with survival outcome is proposed by VanderWeele. However, current methods mainly rely on the assumption of sequential ignorability. We proposed an additive hazard type model for failure time outcome without requirement of sequential ignorability. We estimated causal parameter of interest via inverse censoring probability weighted (IPCW) estimating equation. To handle the measurement error in the mediator, we proposed to use either regression calibration or estimated estimating equation technique. The estimators were compared with the regression method assuming sequential ignorability by simulation studies. We applied our method on WHI study to evaluate whether the dietary modification intervention decrease breast cancer risk via decrease total energy consumption measured by food frequency questionnaire (FFQ).