多智能体LLM系统用于临床问诊训练评估
原标题：Evaluating Scaffolding-Oriented Multi-Agent Large Language Model System for Clinical Interview Training
AI 摘要
研究团队开发了一个面向临床问诊训练的脚手架式多智能体LLM系统（AI标准化病人），包含病人智能体、导师智能体和回合级评估智能体。在100名医学生的随机对照试验中，多智能体脚手架条件相比结构化非LLM对照组提升了最终考试成绩，尤其在沟通、共情表达和病史采集行为上改善显著，但诊断准确率无显著差异。团队还发布了多专家标注数据集以支持后续研究。
正文节选
Evaluating Scaffolding-Oriented Multi-Agent Large Language Model System for Clinical Interview Training Abstract Clinical education must prepare medical students to conduct safe, coherent, and patient-centered interviews under conditions of uncertainty. Traditional standardized patient (SP) training is resource-intensive and difficult to scale, while case-based learning alone does not reproduce the real-time communicative demands of a consultation. We developed a scaffolding-oriented multi-agent