Doubly Robust Fusion of Many Treatments for Policy Learning

Individualized treatment rules/recommendations (ITRs) aim to improve patient outcomes by tailoring treatments to the characteristics of each individual. However, in high-dimensional treatment settings, existing methods face significant challenges due to data sparsity within treatment groups and highly unbalanced covariate distributions across groups. To address these challenges, we propose a novel calibration-weighted treatment fusion procedure that robustly balances covariates across treatment groups and fuses similar treatments using a penalized working model. The fusion procedure ensures the recovery of latent treatment group structures when either the calibration model or the outcome model is correctly specified. In the fused treatment space, practitioners can seamlessly apply state-of-the-art ITR learning methods with the flexibility to utilize a subset of covariates, thereby achieving robustness while addressing practical concerns such as fairness. We establish theoretical guarantees, including consistency, the oracle property of treatment fusion, and regret bounds when integrated with multi-armed ITR learning methods such as policy trees. Simulation studies show superior group recovery and policy value compared to existing approaches. We illustrate the practical utility of our method using EHR-derived data from patients with Chronic Lymphocytic Leukemia and Small Lymphocytic Lymphoma.

Personalized medicine aims to recommend treatments that are tailored to the specific characteristics of each patient. However, when faced with numerous treatment options, deriving individualized treatment rules (ITRs) from limited data can be challenging. Small sample sizes in some treatment groups, coupled with variations across patient populations, often result in unreliable decisions.We introduce a novel statistical method that tackles both challenges by adjusting for baseline differences and combining treatments that yield similar outcomes. This is achieved in two stages: first, by calibrating patient characteristics across treatment groups, and then by grouping together similar treatments. A major benefit is its double robustness—our method remains valid as long as either the treatment model or the outcome model is correct, which provides reliability even amid messy real-world healthcare data.This strategy assists physicians in making optimal use of the data at hand, particularly in intricate clinical contexts with many treatment options. It leads to more consistent and interpretable treatment recommendations, particularly in fields like oncology, where diverse treatment options exist but patient information may be limited. In both simulations and real-world leukemia data, our method enhances the identification of treatment groups and promotes more informed, patient-centered care.