Estimating the Optimal Regime to Prolong Survival in Treatment of Prostate Cancer Recurrence Using Flexible Weighting Models

• Activity Number: 625
• Type: Invited
• Date/Time: Thursday, August 4, 2016 : 8:30 AM to 10:20 AM
• Sponsor: ENAR
• Author(s): Jincheng Shen and Lu Wang* and Jeremy M. G. Taylor
• Companies: Harvard T.H. Chan School of Public Health and University of Michigan and University of Michigan
• Keywords: Dynamic Treatment Regime ; Inverse Probability Weighting ; Random Forest ; Causal Inference ; Restricted Mean Survival Time

Abstract

The challenge to find a clinically meaningful optimal dynamic treatment regime from the observational data lies in the fact that the regime being followed by each subject is not well characterized. In this talk, we focused on the scenario where 1) the outcome is time-to-event, and 2) some of the covariates are time-varying, and possibly follow a complicated pattern. We consider a class of dynamic treatment regimes that are determined by the time-varying longitudinal covariates. A novel Random Forest based inverse probability weighting scheme is proposed to adjust for the complexity in the mechanism of adherence in the observational data while still allowing for some patients to remain treatment-free as defined by the regime. The optimal regime is then identified as the one with the largest restricted mean survival time. The performance of the proposed method is evaluated through simulation studies, which are designed to mimic the situation of salvage therapy to reduce the risk of cancer recurrence in prostate cancer. We also apply the method to an observational prostate cancer study.