Fine-grained List-wise Alignment for Generative Medication Recommendation

TL;DR

FLAME introduces a fine-grained list-wise alignment framework for LLMs to improve medication recommendations by modeling drug interactions and patient data, achieving state-of-the-art results.

Contribution

The paper presents FLAME, a novel list-wise alignment method with step-wise policy optimization and enhanced patient modeling for safer, more accurate medication recommendations.

Findings

Achieves state-of-the-art accuracy on benchmark datasets.

Provides controllable safety-accuracy trade-offs.

Demonstrates strong generalization across clinical scenarios.

Abstract

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.