ChatGPT在重症肺炎患者家属教育和沟通的适当性与准确性研究

Appropriateness and Comprehensiveness of Using ChatGPT for Diabetes Patient Education and Communication

目的检验ChatGPT 3.5在线人工智能平台对重症肺炎家属关心的肺炎诊治、治疗方案和患者管理等问题回复的准确性及适当性,探讨其在重症肺炎家属教育和沟通中的潜在应用。方法参考当地三甲医院医师相关临床经验与患者在线咨询平台,制订一个涵盖重症肺炎诊治、患者管理、治疗方案等方面相关10个问题的重症肺炎家属信息需求问卷。可靠性使用专家权威程度进行衡量。将每个问题询问ChatGPT 3.53次,得出3个回复。问卷中每个问题的3个回复都由来自多个中心的专家进行独立主观地评分,且专家可对问题回复提供进一步的补充意见和建议。结果最终纳入7名专家。经过计算,专家对于重症肺炎相关的临床知识及诊治的熟悉程度Cs为0.91,专家对ChatGPT 3.5回答重症肺炎患者家属问题的准确性的判断依据Ca为0.97、权威系数Cr为0.94,7位专家的评价ICC值为0.659,大多数回复中最常见的质量等级是“好”或“非常好”(92.86%)。结论ChatGPT 3.5对重症肺炎家属相关问题的回答质量相对准确可靠,ChatGPT有望作为帮助及时评估和满足患者家属需求,促进患者良好临床结局的辅助信息工具而发挥作用。

Objective To evaluate the accuracy and appropriateness of responses from the ChatGPT 3.5 online AI platform to concerns of family members of severe pneumonia patients regarding pneumonia diagnosis,medication selection,and patient management.Furthermore,to explore its potential applications in educating and communicating with the families of severe pneumonia patients.Methods By referencing the clinical experiences of local tertiary hospital doctors and online patient consultation platforms,a questionnaire was devised encompassing 10 queries related to severe pneumonia diagnosis,patient management,and medication selection.The reliability was assessed through expert authority.Each question was posed to ChatGPT 3.5 three times,yielding three responses.The three responses to each question in the questionnaire were independently and subjectively scored by experts from multiple centers,who could also provide additional feedback and suggestions on the responses.Results Seven experts were ultimately included.Upon calculation,the familiarity level(Cs)of experts with clinical knowledge and treatment concerning severe pneumonia was 0.91.The basis(Ca)for experts'judgments on the accuracy of ChatGPT 3.5's responses to severe pneumonia family members'questions was 0.97,and the authority coefficient(Cr)was 0.94.The Inter-Class Correlation(ICC)value of the evaluations by the seven experts was 0.659,with the most common quality ratings being"good"or"very good"(92.86%).Conclusions ChatGPT 3.5 demonstrated relatively accurate and reliable response quality to queries related to family members of severe pneumonia patients.ChatGPT holds promise as an auxiliary information tool to help assess and meet the needs of patient families promptly,thus fostering positive clinical outcomes for patients.