A0840
Title: Estimating heterogeneous treatment effects with survival outcomes via deep survival learner
Authors: Yuming Sun - College of William and Mary (United States) [presenting]
Abstract: Understanding how treatment effects vary across patient subgroups is critical for precision medicine. Estimating the conditional average treatment effect (CATE) in survival settings poses unique challenges, as existing methods often rely on rigid model assumptions and overlook temporal dependencies inherent in survival data. To address these limitations, the deep survival learner (DSL) is proposed, a novel framework for estimating heterogeneous treatment effects under right censoring. DSL extends the doubly robust learner by integrating recurrent neural networks as base learners, enabling flexible modeling of complex, time-dependent relationships between baseline covariates and survival outcomes. This design enhances both robustness to model misspecification and efficiency in CATE estimation. DSL is assessed through extensive simulations, demonstrating its improved accuracy over existing approaches. DSL is further applied to the Boston Lung Cancer Study (BLCS) to evaluate the impact of chemotherapy among patients with non-small cell lung cancer. The results reveal substantial heterogeneity in treatment effects, highlighting the potential of DSL to inform individualized treatment strategies in clinical practice.
