Beyond Role-Playing: A Gen AI-Driven Health Consultation Training System

Abstract

Recent advances in large language models enable virtual patient simulations to generate natural language dialogue for healthcare education. However, most existing implementations focus solely on dialogue interaction, lacking structured feedback to explain how individual utterances impact the consultation. This study presents an online health consultation training system for school nurse education. The system integrates generative artificial intelligence child avatars, an interaction mechanism based on trust, and structured reflection support. Featuring three child personality profiles, the system evaluates each learner utterance in real time to produce transparent trust score changes and textual rationales. A formative evaluation with five university students demonstrated high overall usability and positive perceptions of the reflection interface. Furthermore, interaction log analysis revealed that empathic strategies consistently produced high trust increases, whereas premature directives caused sharp trust decreases. These findings demonstrate that the system effectively addresses traditional peer role-playing limitations by reproducing diverse child behaviors, capturing dynamic trust building processes, and providing actionable feedback for structured reflection.