Multi-treatment classification weighted learning for estimating optimal treatment regimes

We propose a Multi-treatment Classification Weighted Learning (MCWL) framework for estimating optimal treatment regimes (OTRs) with multiple treatment options. Our approach reformulates treatment selection as a multi-class weighted classification problem through data space expansion and employs augmented inverse probability weighting estimators (AIPWE) for contrast functions to achieve double robustness. This ensures consistent identification of the optimal treatment regime even under model misspecification. The framework is flexible and compatible with a wide range of machine learning classifiers. Extensive simulations and a real-data application demonstrate that MCWL outperforms existing methods in accuracy and robustness across diverse scenarios.