AI聊天机器人作为区域麻醉临床决策支持:e-Consults新时代是否已准备就绪?
该试点研究评估了四种流行的大语言模型聊天机器人在区域麻醉临床场景和管理问题中的表现,发现所有平台在临床判断、全面性和决策依据方面均存在不足,复杂场景下更为明显,且存在虚构内容。结论认为基于大语言模型的聊天机器人尚不能真正辅助区域麻醉的基本或复杂临床场景。
要点
- 四种大语言模型聊天机器人在区域麻醉临床决策中均存在临床判断、全面性和决策依据的不足。
- 复杂临床场景下聊天机器人的表现缺陷更为明显。
- 聊天机器人的回答中存在虚构内容,尚不能真正辅助临床决策。
核心结论
- 基于大语言模型的聊天机器人应用于区域麻醉亚专科,尚不能真正辅助基本或复杂临床场景的决策。
研究设计与样本
- 试点研究,纳入四种流行的大语言模型聊天机器人,使用相同的基本和复杂临床场景及常见管理问题进行提示,并进行区域麻醉亚专科的定性评估。
主要发现
- 所有四种平台在临床判断、全面性和决策依据方面均表现出弱点,复杂场景下更为明显。
- 回答中存在虚构内容的证据。
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 AND OBJECTIVE: There is growing interest to further incorporate artificial intelligence into clinical practice. Perioperative decision-making for patients considered for regional anesthesia is complex, and the use of artificial intelligence as a clinical decision support tool in this context is underinvestigated. We aim to evaluate large language models (LLMs) when prompted with clinical scenarios and management questions in the subspecialty of regional anesthesiology. METHODS: A pilot study was conducted with four popular LLM-based chatbots. They were prompted with the same basic and complex clinical scenarios and common management questions followed by a qualitative review specific to regional anesthesiology. RESULTS: All four platforms demonstrated weaknesses in clinical judgment, thoroughness, and rationale in decision-making that were more pronounced for the complex scenario. There is evidence of confabulations in the responses. CONCLUSIONS: LLM-based chatbots applied to the subspecialty of regional anesthesiology are not ready to truly assist in basic or complex clinical scenarios.
引用
- Corpman D, Auyong DB, Walters D, et al. Using AI chatbots as clinical decision support for regional anesthesia: is the new era of "e-Consults" ready for primetime? Reg Anesth Pain Med. 2025. PMID: 40610221. doi:10.1136/rapm-2025-106850
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。