A1057
Title: Optimal dynamic treatment regimes for high-dimensional accelerated failure time model
Authors: Wei Zhao - Shandong University (China) [presenting]
Abstract: Optimizing treatment sequences for chronic disease patients is a critical challenge in precision medicine. The purpose is to address this challenge by developing a novel method for estimating dynamic treatment regimes (DTRs) based on a high-dimensional accelerated failure time model. The approach leverages survival time, high-dimensional clinical features collected from patients followed through multiple intervention stages, assuming that survival time at each stage follows an accelerated failure time model. The objective is to estimate treatment decision rules at each stage and identify the optimal DTR that maximizes expected outcomes. Backward induction and counterfactual survival time are employed within the high-dimensional accelerated failure time model estimation. A solution sequence is iteratively obtained using SDAR combined with an l0 penalty. It is theoretically demonstrated that, under a mild regularity condition on the covariate matrix, the estimation error decays exponentially to the optimal bound with high probability. Simulation studies and real data analysis demonstrate that the proposed approach achieves superior estimation and decision accuracy compared to Lasso and other existing methods.
