语言、权力与可及性:大型语言模型与医学公平
本文为社论,提出「偏见金字塔」概念框架,阐释大型语言模型在麻醉、围术期医学和重症监护中可能在全球知识生态、机构信息环境和个人临床工作流三个层面产生不公平,并强调通过治理、教育和协调行动在模型嵌入常规实践前实现公平实施。
要点
- 偏见金字塔框架揭示LLMs不公平可源于全球知识生态、机构信息环境和个人临床工作流三个层面。
- 公平应被视为学习型医疗系统的演化属性,可放大或逐步改善不公平。
- 在LLMs嵌入常规围术期和重症实践前,需通过治理、教育和协调行动促进公平实施。
核心结论
- LLMs在麻醉、围术期医学和重症监护中的公平实现取决于其开发、实施、使用和治理方式,而不仅是模型能力。
研究设计与样本
- 社论,提出偏见金字塔概念框架,未报告系统检索或样本量。
主要发现
- 偏见金字塔解释不公平可出现在全球知识生态系统、机构信息环境和个人临床工作流三个层面。
- 框架纳入强化反馈,将公平概念化为学习型医疗系统的演化属性,可放大或逐步改善不公平。
临床怎么用
- 偏见金字塔可作为实践实施框架,指导治理、教育和协调行动,在LLMs嵌入常规围术期和重症实践前促进公平。
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
Large language models (LLMs) are increasingly being integrated into anaesthesia, perioperative medicine, and critical care, creating new opportunities to improve communication, decision support, and access to healthcare knowledge. Realising these opportunities equitably depends not only on model capability but also on how LLMs are developed, implemented, used, and governed within healthcare systems. This editorial introduces the Bias Pyramid, a conceptual framework that explains how inequity can emerge across global knowledge ecosystems, institutional information environments, and individual clinical workflows, and identifies opportunities for equitable implementation. By incorporating reinforcing feedback, the framework conceptualises equity as an evolving property of learning healthcare systems that can either amplify existing inequities or progressively improve equity over successive learning cycles. The Bias Pyramid provides a practical implementation framework to guide governance, education, and coordinated action before LLMs become embedded in routine perioperative and critical care practice.
引用
- Lashin H. Language, power, and access: large language models and equity in medicine. Br J Anaesth. 2026. PMID: 42711214. doi:10.1016/j.bja.2026.08.008
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。