Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 372 - SPEED: SPAAC SESSION IV
Type: Topic-Contributed
Date/Time: Thursday, August 12, 2021 : 12:00 PM to 1:50 PM
Sponsor: Biometrics Section
Abstract #318617
Title: Semiparametric Q-Learning for Optimal Dynamic Treatment Regime with Right-Censored Survival Outcome and Missing Treatment Data
Author(s): Lingyun Lyu* and Yu Cheng and Abdus S Wahed
Companies: Department of Biostatistics, University of Pittsburgh and University of Pittsburgh and Department of Biostatistics, University of Pittsburgh
Keywords: Cox proportional hazard model; Weighted hot-deck multiple imputation; Missing data; Optimal dynamic treatment regimes; Precision medicine
Abstract:

An optimal dynamic treatment regime (DTR) is a sequence of treatment decisions that yields the best expected outcome. Limited work has been reported for estimating optimal DTRs for survival outcome with right-censoring, and most of the existing work used parametric models. We propose a new approach to estimate optimal DTRs in the survival setting using Q-learning, where we posit Cox proportional hazard (CPH) model to estimate the treatment rule for each stage, and then use the weighted hot-deck multiple imputation (MI) method to predict the optimal potential survival time. We extend the method to the incomplete data setting by using inverse probability weighting and MI. Our method offers several advantages. First, the hot-deck imputation method does not rely on parametric models, thus the predicted potential survival time is potentially less sensitive to model misspecification. Second, only plausible potential survival times can be imputed. Third, use of CPH model provides more flexibility regarding the shape of the baseline hazard. Extensive simulation studies were conducted to show the effectiveness of our method. The approach was applied to optimize DTR for leukemia treatment.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2021 program