IPPM; SyPPM; SoCPPM
收藏资源简介:
该研究发布了三个专家标注的临床响应数据集,旨在评估大语言模型(LLM)在患者门户消息响应草拟任务中的对齐性能。IPPM和SoCPPM包含真实患者消息与EHR摘要,分别模拟理想条件和实际临床场景;SyPPM为公开的半合成数据集,均围绕8类临床主题(如共情、药物询问)构建。数据集总计700条样本,每条包含患者消息、EHR摘要及临床医生响应,通过混合方法(主题分析、专家研讨)标注生成。其核心应用于优化LLM在医患沟通中的可靠性,减少临床医生编辑负担,提升响应效率与安全性。
This study releases three expert-annotated clinical response datasets intended to evaluate the alignment performance of Large Language Models (LLMs) in the task of drafting responses to patient portal messages. IPPM and SoCPPM contain real patient messages and EHR summaries, simulating ideal clinical conditions and real-world clinical scenarios respectively; SyPPM is a public semi-synthetic dataset constructed around 8 clinical themes including empathy and medication inquiries. The datasets consist of a total of 700 samples, each containing patient messages, EHR summaries, and clinician responses, which were annotated and generated through mixed methods such as thematic analysis and expert workshops. Their core applications are to optimize the reliability of LLMs in doctor-patient communication, reduce the editing burden on clinicians, and enhance response efficiency and safety.




