文献速递生殖医学

大语言模型回答不孕患者常见问题的准确性与共情能力初探

Can large language models provide accurate and empathetic answers to the most frequently asked questions by infertile patients? A pilot study.
PubMed 文献 · Reprod Biomed Online · 2025-08-14

本研究让两个临床医生和两个胚胎学家回答13个生殖医学常见问题,同时将相同问题及相同文本长度限制输入免费在线大语言模型(LLM),由四名评估者盲法评价质量与准确性,并由一名心理学家评价共情。结果显示LLM答案平均字数显著多于人类(P<0.001),平均质量与专业人员无统计学差异,无完全异常答案,仅少数含错误或不适当信息;胚胎学家答案排名显著高于LLM(P=0.02),临床医生则无此差异;心理学家在26个问题中有14个选择LLM答案最具共情或清晰度。

要点

  • LLM回答不孕患者常见问题的平均质量与人类专业人员无统计学差异
  • 胚胎学家回答质量显著高于LLM(P=0.02),临床医生与LLM无差异
  • LLM在26个问题中有14个被心理学家评为最具共情或清晰度

核心结论

  • LLM可作为辅助生殖中心回答患者常见问题的教育工具,但需谨慎评估和监管以防止不准确信息传播。

研究设计与样本

  • 单中心、双盲、前瞻性初步研究,涉及两名临床医生和两名胚胎学家回答各自领域的13个常见问题。
  • 相同问题以相同文本长度限制输入免费在线LLM,由四名评估者(两名妇科医生和两名胚胎学家按主题)盲法评价质量和准确性。
  • 一名心理学家评价共情。

主要发现

  • LLM答案平均字数显著高于人类(P<0.001)。
  • LLM与专业人员的平均答案质量无统计学差异。
  • LLM无完全异常答案,仅少数含错误或不适当信息或被评估者评为很差。
  • 胚胎学家答案排名显著高于LLM(P=0.02),临床医生与LLM无显著差异。
  • 心理学家在26个问题中有14个选择LLM答案最具共情、清晰或两者兼有。

临床怎么用

  • 可在辅助生殖中心将LLM作为回答患者常见问题的教育工具,但需对输出内容进行审核和监管。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

原文

RESEARCH QUESTION: Is the quality, relevance and empathy of the answers provided by large language models (LLMs) in response to the most frequently asked patient questions in reproductive medicine comparable to those provided by human specialists? DESIGN: This monocentric, double blind, prospective study involved two clinicians and two embryologists who answered 13 frequently asked questions in their respective field. The same questions were asked to a free online LLM, with the same constraint of text length as practitioners. All answers were blindly evaluated by four assessors (two gynaecologists and two embryologists depending on the topic) for quality and accuracy. A psychologist also evaluated empathy. RESULTS: The mean number of words per answer was significantly higher (P < 0.001) for LLM than for humans. The average quality of answers was not statistically different between LLM and professionals. No answer provided by LLM was evaluated as completely aberrant, and only a minority contained false or inappropriate information or was scored as being very poor by assessors. Answers provided by embryologists, but not clinicians, ranked significantly higher (P = 0.02) than LLM. The psychologist chose LLM answers as most empathetic, clear, or both, in 14 out of 26 questions. CONCLUSIONS: LLMs could be used as an educational tool within assisted reproductive technology centres to answer frequently asked patient questions. Although the potential applications of LLMs' capabilities in answering medical questions are numerous, this should be carefully evaluated and regulated to prevent the dissemination of inaccurate information to patients.

来源:Reprod Biomed Online,查看原文。版权归原作者所有。

引用

  1. Gély L, Chaillot M, Fréour T, et al. Can large language models provide accurate and empathetic answers to the most frequently asked questions by infertile patients? A pilot study. Reprod Biomed Online. 2025. PMID: 41771212. doi:10.1016/j.rbmo.2025.105221

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。