人工智能在患者教育中的应用:全髋关节置换术后聊天机器人可靠性的比较评估
本研究评估并比较不同人工智能大语言模型(AI LLMs)对全髋关节置换术(THA)术后护理常见患者问题的回答,覆盖可靠性、质量、准确性和可读性四个维度。结果显示Gemini的评分者间信度最高(ICC=0.71),ChatGPT模型为中等至良好(ICC=0.67-0.71);Gemini在可靠性上显著更高(P<0.001),ChatGPT 3.5可读性最佳(FRES中位数=55)。结论认为AI LLMs可作为THA术后患者教育的补充工具,但需临床医生监督以确保准确性和安全性。
要点
- Gemini在THA术后患者教育回答的可靠性上显著优于其他模型(P<0.001)。
- ChatGPT 3.5的可读性最佳,FRES中位数为55。
- AI LLMs可作为患者教育补充工具,但需临床医生监督。
核心结论
- AI LLMs可作为全髋关节置换术后患者教育的补充工具,但使用需临床医生监督以确保准确性和安全性。
研究设计与样本
- 三个AI模型于2025年5月2日至5日生成回答,使用改良DISCERN量表、全球质量量表、准确性量表和Flesch阅读易读性评分进行评估。
主要发现
- 评分者间信度最高为Gemini(ICC=0.71),ChatGPT模型为中等至良好(ICC=0.67-0.71)。
- Gemini在可靠性上显著更高(P<0.001)。
- ChatGPT 3.5可读性最佳(FRES中位数=55)。
- 模型间在可靠性、质量和可读性指标上存在显著差异。
关键数值与推荐
- 可靠性评估采用ICC,Gemini为0.71,ChatGPT模型为0.67-0.71。
- 可读性评估采用FRES,ChatGPT 3.5中位数为55。
- 可靠性比较中Gemini显著更高(P<0.001)。
临床怎么用
- AI LLMs可用于全髋关节置换术后患者教育,但需临床医生监督以确保回答准确安全。
- 未来研究应探索以患者为中心的评价以及结合AI指导与专业监督的混合模式。
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
OBJECTIVE: This study aimed to evaluate and compare the responses of different artificial intelligence-based large language models (AI LLMs) to frequently asked patient questions regarding post-operative care following total hip arthroplasty (THA) across four domains: reliability, quality, accuracy, and readability. DESIGN: Twenty-seven commonly asked questions were identified through Google Trends and expert consensus, covering exercises, activities of daily living, and dislocation precautions. Responses generated by the three AI models (May 2-5, 2025) were assessed using the modified DISCERN scale, Global Quality Scale (GQS), Accuracy Scale, and Flesch Reading Ease Score (FRES). RESULTS: Inter-rater reliability was highest for Gemini (ICC = 0.71), while ChatGPT models demonstrated moderate-to-good reliability (ICC = 0.67-0.71). Gemini scored significantly higher in reliability (P < 0.001), whereas ChatGPT 3.5 achieved the greatest readability (median FRES = 55). Significant differences were observed among models across reliability, quality, and readability metrics. CONCLUSION: AI LLMs can serve as supplementary tools for patient education following THA; however, their use requires clinician oversight to ensure accuracy and safety. Future studies should explore patient-centered evaluations and hybrid approaches combining AI guidance with professional supervision in orthopedic rehabilitation.
引用
- Bozgeyik-Bağdatli S, Bazancir-Apaydin Z, Güney-Deniz H, et al. Artificial Intelligence in Patient Education: A Comparative Evaluation of Chatbot Reliability After Total Hip Arthroplasty. Am J Phys Med Rehabil. 2026. PMID: 42296472. doi:10.1097/PHM.0000000000003061
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。