Paper Type

Complete

Paper Number

PACIS2026-1992

Description

Large language model (LLM)-based agents have shown promise in supporting healthcare consultations, yet their effectiveness remains limited in community healthcare contexts where patients often provide incomplete, ambiguous, or non-technical descriptions of symptoms. These conditions introduce significant incomplete information and uncertainty during multi-turn consultations. To address these challenges, we introduce Extent-Med, an entropy-guided dynamic decision-making framework designed for multi-turn medical consultation under incomplete information in community healthcare. The proposed approach can structure patient narratives, grounds reasoning in external medical knowledge, and prioritizes high-information symptoms during sequential interactions. We evaluate Extent-Med on both public medical dialogue datasets and real-world scenarios. Experimental results demonstrate that our system significantly outperforms baseline LLM-based agents and knowledge-augmented variants. This research contributes to the literature on AI-enabled decision support by demonstrating how LLM-based agents can be designed to manage uncertainty in conversational healthcare settings. The findings provide practical insights for developing trustworthy AI systems for medical consultation.

Comments

14-Healthcare

Download

Share

COinS
 
Jul 5th, 12:00 AM

Extent-Med: A Dynamic Decision-Making Approach under Incomplete Information for Community Healthcare

Large language model (LLM)-based agents have shown promise in supporting healthcare consultations, yet their effectiveness remains limited in community healthcare contexts where patients often provide incomplete, ambiguous, or non-technical descriptions of symptoms. These conditions introduce significant incomplete information and uncertainty during multi-turn consultations. To address these challenges, we introduce Extent-Med, an entropy-guided dynamic decision-making framework designed for multi-turn medical consultation under incomplete information in community healthcare. The proposed approach can structure patient narratives, grounds reasoning in external medical knowledge, and prioritizes high-information symptoms during sequential interactions. We evaluate Extent-Med on both public medical dialogue datasets and real-world scenarios. Experimental results demonstrate that our system significantly outperforms baseline LLM-based agents and knowledge-augmented variants. This research contributes to the literature on AI-enabled decision support by demonstrating how LLM-based agents can be designed to manage uncertainty in conversational healthcare settings. The findings provide practical insights for developing trustworthy AI systems for medical consultation.