DischargeSim
收藏资源简介:
DischargeSim是一个用于评估大型语言模型(LLM)在出院教育中扮演个性化出院导师能力的模拟基准。它模拟了医生代理和具有不同心理社会背景的病人代理之间的多轮对话。数据集围绕六个临床验证的出院主题进行构建,包括诊断、检查和治疗、返院指标、药物、出院后治疗和随访。每个主题都作为一个特定阶段的教育目标。医生代理必须不仅要提供准确的信息,还要根据病人代理的心理社会背景(如健康素养、教育程度、情绪状态)应用适当的策略,如简化、情绪支持和共同决策。DischargeSim支持三个相互关联的评价层次:对话质量、出院总结生成和病人理解。
DischargeSim is a simulated benchmark designed to evaluate the capability of large language models (LLMs) to serve as personalized discharge tutors for patient discharge education. It simulates multi-turn conversational interactions between a physician agent and patient agents with diverse psychosocial backgrounds. The dataset is constructed around six clinically validated discharge-related topics, including diagnosis, examinations and treatments, readmission indicators, medications, post-discharge care, and follow-up visits. Each topic functions as an educational goal for a specific phase of discharge education. The physician agent must not only provide accurate medical information but also apply appropriate communication strategies tailored to the patient agent’s psychosocial characteristics, such as health literacy, educational level, and emotional state—including simplifying medical explanations, offering emotional support, and engaging in shared decision-making. DischargeSim supports three interrelated evaluation dimensions: dialogue quality, discharge summary generation, and patient comprehension.
DischargeSim 数据集概述
数据集简介
DischargeSim 是一个医疗出院模拟框架,用于创建真实的医患互动以模拟医院出院场景。
核心功能
- 使用 AI 代理扮演医生和患者角色来模拟医疗出院对话
- 医生代理指导患者理解诊断、治疗、药物和随访护理计划
- 生成结构化输出包括完整对话日志、AHRQ 格式的访后计划和临床笔记格式的出院摘要
技术实现
- 支持 Python 3.9+ 环境
- 可使用 vLLM 运行本地模型
- 支持 OpenAI API 调用
输出内容
history.csv- 医患完整对话日志plan.jsonl- AHRQ 格式的结构化访后计划summary.jsonl- 临床文档格式的出院摘要
许可协议
CC-BY-NC 4.0 许可证

- 1DischargeSim: A Simulation Benchmark for Educational Doctor-Patient Communication at Discharge美国马萨诸塞大学安姆斯特分校计算机科学与信息学院 · 2025年



