A Meta-Learner Framework to Estimate Individualized Treatment Effects for Survival Outcomes

• Activity Number: 378 - LiDS Student Paper Award Winners: Topic-Contributed Papers
• Type: Topic Contributed
• Date/Time: Wednesday, August 10, 2022 : 8:30 AM to 10:20 AM
• Sponsor: Lifetime Data Science Section
• Author(s): Na Bo* and Yue Wei and Lang Zeng and Ying Ding and CHAERYON KANG
• Companies: University of Pittsburgh and University of Pittsburgh and University of Pittsburgh and University of Pittsburgh and Dept. of Biostatistics, University of Pittsburgh
• Keywords: age-related macular degeneration; counterfactual outcome; heterogeneous treatment effect; precision medicine; treatment rule recommendation

Abstract

One important aspect of precision medicine is to allow physicians to choose the most suitable treatment for their patients, which requires an understanding of the heterogeneity of treatment effects from a patient-centric view. With large amount of genetic data being generated, a full picture of individuals' characteristics is forming. Recent development using machine learning methods within the counterfactual framework shows great potential in analyzing such data. In this research, we develop a meta-learner approach to estimate individual treatment effect (ITE) for survival outcome. We consider two algorithms, T-learner and X-learner, each combined with three machine learning methods: random survival forest, Bayesian accelerate failure time model, and deep survival neural network. The performance of these methods is compared through simulations. We then apply the methods on a randomized clinical trial (RCT), the AREDS study for age-related macular degeneration, to estimate ITEs and identify genetic variants that contribute to the heterogeneous treatment effects. The resulting treatment recommendation rules are applied to a subsequent RCT, AREDS2, as an external validation.