利用人工智能解读妇科细胞病理报告:面向患者门户的探索性研究
本研究让ChatGPT简化15份基于Bethesda系统的妇科细胞病理报告,6名病理医生用5分Likert量表评价180份解释。多数解释被评准确、完整且无感知伤害,提示2优于提示1,但结论基于病理医生主观意见而非患者理解。
要点
- ChatGPT经适当提示可生成准确、完整的妇科细胞病理报告简化解释。
- 提示2(「最重要的信息」)评价优于提示1(「向儿童解释」)。
- 结论基于病理医生意见,需直接测量患者理解与后续行动。
核心结论
- 经适当提示,AI聊天机器人有潜力为患者提供准确、简单的妇科细胞病理报告摘要,伤害极小或无伤害。
研究设计与样本
- 探索性研究:一名细胞病理医生基于Bethesda系统创建15份妇科细胞病理报告,两名细胞病理医生筛选提示,六名病理医生用5分Likert量表评价180份ChatGPT解释。
主要发现
- 六名病理医生提供180份评价,绝大多数解释被评为准确、事实完整且未感知为可能伤害患者,尤其使用提示2时。
- 提示1生成的解释评分较低。
关键数值与推荐
- 评价使用5分Likert量表,评估事实准确性、相关医学信息纳入及对患者潜在伤害。
- 研究采用提示1「用简单语言向儿童解释此医学报告」和提示2「我应从病理报告知道的最重要事项是什么」。
临床怎么用
- 病理科智能体可参考提示2设计患者友好的报告简化输出,但需注意结论基于病理医生意见而非患者理解。
局限性
- 发现基于病理医生意见,是患者理解的间接测量。
- 未来研究应直接测量患者理解度和预期随访行动。
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
BACKGROUND: The 21st Century Cures Act allows patients to have immediate access to their medical records. However, it is documented that health literacy levels in patient portal users vary, and results shared with patients may require further explanation. The objective of this exploratory study was to evaluate if an artificial intelligence (AI) chatbot assistant (ChatGPT; Open AI, Inc.) may be used to simplify gynecologic cytopathology reports for patients. The authors also assessed pathologists' opinions on the quality of simplified pathology reports generated with ChatGPT. METHODS: An experienced cytopathologist created 15 unique gynecologic cytopathology reports based on The Bethesda System for Reporting Cervical Cytology. Two cytopathologists evaluated multiple possible prompts to ask ChatGPT for simplified, summarized explanations of cytopathology reports. The two prompts selected for this study were prompt 1 ("Explain this medical report to a child using simple language") and prompt 2 ("What are the most important things I should know from my pathology report?"). Six pathologists received a questionnaire to evaluate factual accuracy, the inclusion of relevant medical information, and potential harm to the patient. They were asked to score each simplified explanation using a five-point Likert scale. Pathologists were also asked to list any missing information and potential physical and/or psychological harm. The scores were tabulated and analyzed. RESULTS: Six board-certified pathologists provided 180 evaluations of ChatGPT's explanations. The vast majority were scored as accurate, factually complete, and not perceived as a possible cause of harm for the patients, especially when using prompt 2. Explanations from ChatGPT for prompt 1 were scored less favorably. CONCLUSIONS: This study provides insight into the field of automated simplification of pathology reports using ChatGPT. With proper prompting, AI chatbots have potential to serve as powerful assistants to patients desiring accurate, simple summaries of their gynecologic cytopathology reports, with minimal or no harm. These findings are based on pathologists' opinions-an indirect measure of patient understanding. Future studies should directly measure patient comprehension and intended follow-up actions to ensure that AI-simplified reports are not only medically accurate but truly patient-centered.
引用
- Lichtenberg RS, Mon KS, Mehrotra S, et al. Leveraging artificial intelligence to decipher gynecologic cytopathology reports: Insights from an exploratory study for possible use in patient portals. Cancer Cytopathol. 2026. PMID: 41663331. doi:10.1002/cncy.70077
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。