Fine-grained List-wise Alignment for Generative Medication Recommendation

Fuli Feng (National University of Singapore) · Chenxiao Fan (University of Science and Technology of China) · Chongming Gao (University of Science and Technology of China) · Wentao Shi (University of Science and Technology of China) · Yaxin Gong (University of Science and Technology of China) · Zhao Zihao (University of Science and Technology of China)

clinical decision-makingdrug-drug interactionsfine-grained list-wise alignmentflamegroup relative policy optimizationmedication recommendationsmultimorbiditypatient modelingpoint-wise predictionrepresentation space reward shapingsafety–accuracy trade-offssequential decision processstructured clinical knowledge

Accurate and safe medication recommendations are critical for effective clinical decision-making, especially in multimorbidity cases. However, existing systems rely on point-wise prediction paradigms that overlook synergistic drug effects and potential adverse drug-drug interactions (DDIs). We propose FLAME, a fine-grained list-wise alignment framework for large language models (LLMs), enabling drug-by-drug generation of drug lists. FLAME formulates recommendation as a sequential decision process, where each step adds or removes a single drug. To provide fine-grained learning signals, we devise step-wise Group Relative Policy Optimization (GRPO) with potential-based reward shaping, which explicitly models DDIs and optimizes the contribution of each drug to the overall prescription. Furthermore, FLAME enhances patient modeling by integrating structured clinical knowledge and collaborative information into the representation space of LLMs. Experiments on benchmark datasets demonstrate that FLAME achieves state-of-the-art performance, delivering superior accuracy, controllable safety–accuracy trade-offs, and strong generalization across diverse clinical scenarios. Our code is available at https://github.com/cxfann/Flame.