GAP: Graph-Assisted Prompts for Dialogue-based Medication Recommendation

TL;DR

GAP introduces Graph-Assisted Prompts to tackle knowledge-intensive, dialogic medication recommendations by constructing a patient-centric dialogue graph from medical conversations and combining it with external knowledge graphs. It generates neighborhood and path-based prompts driven by the graph to retrieve relevant medical knowledge and constrain LLM reasoning, while KG verification and Internet access mitigate hallucinations. Evaluations on DialMed and diagnostic interviewing show GAP outperforms strong baselines and approaches or matches supervised methods in key metrics, with ablation and error analyses highlighting the importance of its prompt components. This framework advances safe, explainable MDS by embedding graph-structured, patient-specific information into prompt generation and knowledge augmentation, enabling more reliable and context-aware medical guidance.

Abstract

Medication recommendations have become an important task in the healthcare domain, especially in measuring the accuracy and safety of medical dialogue systems (MDS). Different from the recommendation task based on electronic health records (EHRs), dialogue-based medication recommendations require research on the interaction details between patients and doctors, which is crucial but may not exist in EHRs. Recent advancements in large language models (LLM) have extended the medical dialogue domain. These LLMs can interpret patients' intent and provide medical suggestions including medication recommendations, but some challenges are still worth attention. During a multi-turn dialogue, LLMs may ignore the fine-grained medical information or connections across the dialogue turns, which is vital for providing accurate suggestions. Besides, LLMs may generate non-factual responses when there is a lack of domain-specific knowledge, which is more risky in the medical domain. To address these challenges, we propose a \textbf{G}raph-\textbf{A}ssisted \textbf{P}rompts (\textbf{GAP}) framework for dialogue-based medication recommendation. It extracts medical concepts and corresponding states from dialogue to construct an explicitly patient-centric graph, which can describe the neglected but important information. Further, combined with external medical knowledge graphs, GAP can generate abundant queries and prompts, thus retrieving information from multiple sources to reduce the non-factual responses. We evaluate GAP on a dialogue-based medication recommendation dataset and further explore its potential in a more difficult scenario, dynamically diagnostic interviewing. Extensive experiments demonstrate its competitive performance when compared with strong baselines.

Figures (5)

• Figure 1: A medical consultation between a human patient and LLMs. Red, blue, and yellow represent diseases, states, and medication concepts. Losartan is restrained for pregnancy, and Labetalol must be used with caution when the patient suffers from bronchitis.
• Figure 2: Overall diagram of GAP.
• Figure 3: Medication recommendation cases on DialMed. GAP is compared with the Chain-of-Thoughts prompting strategy. Green and red indicate accurate and inaccurate recommendation respectively. The neighborhood prompts and path-based prompts are key elements that assist the recommendation.
• Figure 4: Error distribution on the samples of DialMed.
• Figure 5: Prominent schema used for constructing path-based prompts.