EcoSta 2017: Start Registration
View Submission - EcoSta2017
A0656
Title: Value function inference in high-dimensional Q-learning for dynamic treatment regimes Authors:  Wensheng Zhu - Northeast Normal University (China) [presenting]
Donglin Zeng - University of North Carolina at Chapel Hill (United States)
Rui Song - North Carolina State University (United States)
Abstract: A dynamic treatment regime is a set of decision rules, in which treatment decisions often need to be tailored over time according to patients responses to previous treatments as well as covariate history. There is a growing interest in development of statistical inference for optimal dynamic treatment regimens, the difficulty of which is the non-regularity problem in the presence of positive zero-treatment effect probabilities, especially when the dimension of covariates is very high. We propose a high-dimensional Q-learning (HQ-learning) to facilitate the inference of optimal values and parameters, which allows us to simultaneously estimate the optimal dynamic treatment regimes and select the important variables that truly contribute to the individual reward. To reduce the effect of non-regularity on the statistical inference of HQ-learning, we remove those subjects with very small treatment effects by using truncated pseudo-outcome in the first-stage penalized regression. The asymptotic properties for the parameter estimators as well as the estimated optimal value function are also established by adjusting for the bias due to the truncation. The simulation studies and real data analysis show that our HQ-learning can simultaneously estimate the optimal dynamic treatment regimens and selected the important variables very well in the presence of high-dimensional case and non-regularity.