D-Learning to Estimate Optimal Individual Treatment Rules

• Activity Number: 231 - Biopharmaceutical Section Student Papers
• Type: Contributed
• Date/Time: Monday, July 31, 2017 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Zhengling Qi* and Yufeng Liu
• Companies: University of North Carolina, Chapel Hill and University of North Carolina
• Keywords: Individual Treatment Rule ; Multi-armed learning ; prescriptive variable selection

Abstract

Recent Exploration of optimal individual treatment rule (ITR) for patients has attracted a lot of attentions due to the potential heterogeneous response of patients to different treatments. An optimal ITR is a decision function based on patients' characteristics that maximizes the expected clinical outcome. Current literature mainly focuses model-based and classification-based methods. Model-based methods rely on the estimation of conditional means of outcome instead of targeting decision boundaries which they may yield suboptimal decisions. In contrast, although directly targeting optimal ITR using a weighted classification problem, classification-based methods rely on the correct weight for each subject but may create problems due to model misspecification. To overcome the potential drawbacks of these methods, we propose a simple and flexible one-step method to directly learn (D-learning) the optimal ITR without model and weight specifications. Selection of prescriptive variables in D-learning can be naturally incorporated as well. We show estimation consistency and establish error bounds for D-learning. Numerical studies demonstrate the the competitive performance of D-learning.