Covariate-Adjusted Value-Guided Subgroup Identification via Boosting

• Activity Number: 322 - Novel Statistical Methods and Applications in Precision Mental Health
• Type: Topic Contributed
• Date/Time: Tuesday, August 9, 2022 : 2:00 PM to 3:50 PM
• Sponsor: Mental Health Statistics Section
• Author(s): Jinchun Zhang* and Pingye Zhang and Junshui Ma and Yue Shentu
• Companies: Merck & Co. and Beigene and Merck & Co. and Daiichi Sankyo Inc.
• Keywords: Subgroup Identification ; Value-function guided; Boosting

Abstract

It is widely recognized that treatment effect could be different across subgroups of patients. Subgroup analysis, which assesses such heterogeneity, may provide useful information for developing personalized therapies. There has been extensive research developing novel statistical methods for subgroup identification. However, the performance of many could also downgrade when large prognostic effects exist. In this paper, we propose a new framework named Covariate-Adjusted Value-guided boosting for subgroup identification (CAVboost) to address this problem. CAVboost’s approach to estimating the treatment effect by using covariates to account for the prognostic effects mimics the idea of using covariates in an ANCOVA estimator. We show that CAVboost can effectively deal with prognostic effects under both continuous and binary outcomes and improves the subgroup identification capability of the original method by Zhang. We applied the method to a randomized clinical trial of Major Depressive Disorder to identify the subgroups that lead to the highest overall value and also investigated the predictive variable importance.