Learning Optimal Group-structured Individualized Treatment Rules with Many Treatments
Haixu Ma, Donglin Zeng, Yufeng Liu; 24(102):1−48, 2023.
Abstract
In recent years, there has been increased attention towards data-driven individualized decision making problems. One specific area of focus is determining the optimal Individualized Treatment Rule (ITR), which maximizes the expected specified outcome across heterogeneous patient-specific characteristics. Existing methods primarily address binary or a moderate number of treatment arms, often neglecting potential treatment effect structure. However, the effectiveness of these methods diminishes when the number of treatment arms is large. To address this issue, we propose GRoup Outcome Weighted Learning (GROWL), which estimates the latent structure in the treatment space and the optimal group-structured ITRs through a single optimization. For estimating group-structured ITRs, we utilize the Reinforced Angle based Multicategory Support Vector Machines (RAMSVM) within the weighted angle based multi-class classification framework. We establish Fisher consistency, the excess risk bound, and the convergence rate of the value function to provide a theoretical guarantee for GROWL. Empirical results from simulation studies and real data analysis demonstrate that GROWL outperforms several other existing methods.
[abs]