Optimal Treatment Regime Estimation for a Target Population with Summary Statistics Only

• Activity Number: 3 - New Developments and Challenges for Dynamic Individualized Treatments
• Type: Invited
• Date/Time: Sunday, August 7, 2022 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Wenbin Lu* and Shu Yang and Jianing Chu
• Companies: North Carolina State University and North Carolina State University and North Carolina State University
• Keywords: Optimal Treatment Regime; Personalized Medicine ; Summary Statistics; Target Population; Covariates Heterogeneity

Abstract

Personalized treatment decision has recently attracted a great deal of attention. Various methods have been developed for estimating optimal treatment regime based on some source data. However, an important question to answer is whether the obtained optimal treatment regime from the source data can be applied to a target population. There are a few challenges. First, the target population may be heterogeneous in covariate distributions from the population where the source data came from. Second, due to the limited availability of data, for example, individual level patient data can't be shared due to the privacy protection, only summary statistics of the target population are available. To address these challenges, in this work, we develop a class of optimal treatment regime estimation methods for a target population with summary statistics only based on a source data. Both empirical performance and theoretical properties of the proposed estimators are examined.