Deep Reinforcement Learning-Driven Efficacy-Toxicity Balance Optimization Strategy for Personalized Drug Combination in Cancer Patients

Precision medicine demands the careful optimization of dose regimens in cancer therapy, particularly when adjusting doses to balance therapeutic effectiveness against adverse toxicities. In this study, we introduce a deep reinforcement learning (DRL) framework that leverages multimodal patient data to optimize personalized drug combination strategies, aiming to maximize efficacy while minimizing toxicity. The DRL agent employs multi-objective reward functions to identify optimal treatment strategies by integrating genomic, clinical, and pharmacokinetic data through an advanced feature engineering pipeline. This approach was evaluated in a cohort of 2,847 cancer patients encompassing a diverse range of tumor types. Experimental results demonstrate that the algorithm improved predicted treatment response by 23.4% compared to conventional methods, while reducing serious adverse events by 18.7%. These findings highlight a significant advancement in computational approaches for personalized therapy optimization, providing clinically interpretable outputs to guide patient-specific treatment decisions.