Multi-LLM Collaboration for Medication Recommendation

2025-12-04

AI Summary

To address hallucination, inconsistent reasoning, and instability in multi-model ensembles when applying large language models (LLMs) to clinical drug recommendation, this paper proposes a reliable LLM-collaborative reasoning framework. The method introduces “LLM Chemistry”—a novel metric quantifying inter-model collaborative compatibility—alongside a chemistry-inspired interaction modeling mechanism to enable synergistic patient feature extraction and personalized medication generation across multiple LLMs. It further incorporates dynamic calibration and error suppression to mitigate error amplification and interference. Experiments on real-world clinical briefing data demonstrate that the framework significantly outperforms both single-LLM baselines and naive multi-LLM ensembles in recommendation accuracy, logical consistency, and output credibility. By establishing a principled, interpretable, and robust collaboration paradigm, this work advances trustworthy, AI-augmented clinical decision support.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Search and Optimization: Metareasoning and Metaheuristics

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

As healthcare increasingly turns to AI for scalable and trustworthy clinical decision support, ensuring reliability in model reasoning remains a critical challenge. Individual large language models (LLMs) are susceptible to hallucinations and inconsistency, whereas naive ensembles of models often fail to deliver stable and credible recommendations. Building on our previous work on LLM Chemistry, which quantifies the collaborative compatibility among LLMs, we apply this framework to improve the reliability in medication recommendation from brief clinical vignettes. Our approach leverages multi-LLM collaboration guided by Chemistry-inspired interaction modeling, enabling ensembles that are effective (exploiting complementary strengths), stable (producing consistent quality), and calibrated (minimizing interference and error amplification). We evaluate our Chemistry-based Multi-LLM collaboration strategy on real-world clinical scenarios to investigate whether such interaction-aware ensembles can generate credible, patient-specific medication recommendations. Preliminary results are encouraging, suggesting that LLM Chemistry-guided collaboration may offer a promising path toward reliable and trustworthy AI assistants in clinical practice.

Problem

• Improves reliability of medication recommendation from clinical vignettes
• Addresses hallucinations and inconsistency in individual LLM models
• Enables effective, stable, and calibrated multi-LLM collaboration

Innovation

• Multi-LLM collaboration guided by Chemistry-inspired modeling
• Ensembles exploiting complementary strengths for stable recommendations
• Interaction-aware ensembles minimizing interference and error amplification