烧伤护理中的人工智能与算法偏见:伦理挑战文献综述
该叙述性综述综合2010至2025年文献,探讨烧伤护理中人工智能(AI)应用的伦理问题。研究发现算法常对少数族裔、社会经济弱势群体、女性、老年人及复杂合并症患者误分类,而烧伤特异性模型鲜少评估亚组表现或报告人口构成。综述主张在模型全生命周期中关注公平与伦理保障。
要点
- AI在烧伤分诊、深度评估、预后评分、疼痛管理和远程医疗中应用渐增,但存在透明度、问责和偏见等伦理隐患。
- 跨医学领域算法常对少数族裔、弱势群体、女性、老年人及复杂合并症患者误判,烧伤模型鲜少评估亚组表现。
- 需在模型设计、报告、验证和部署后监测中贯彻公平导向,强调正义、不伤害、透明、问责和利益相关者参与。
核心结论
- 当前AI在烧伤护理中的发展轨迹可能嵌入并放大本已脆弱烧伤人群中的不平等。
- 负责任地采用AI需超越技术性能,在模型全生命周期中明确关注公平与伦理保障。
研究设计与样本
- 叙述性综述,系统检索PubMed、Embase和Scopus,纳入报告算法开发或部署并讨论或可推断公平性、公正性或可解释性的研究。
- 文献覆盖2010至2025年,补充重症监护、急诊医学、放射学、皮肤病学和肿瘤学证据。
主要发现
- 跨医学领域算法常对种族和族裔少数群体、社会经济弱势患者、女性、老年人及复杂合并症个体误分类结局。
- 烧伤特异性模型鲜少评估亚组表现或报告人口构成。
- 常见问题包括数据集不具代表性、建模流程不透明及缺乏正式偏见审计。
- 伦理分析碎片化,鲜少基于生物医学伦理或AI治理的既定框架。
临床怎么用
- 在烧伤护理中采用AI时,应实施公平导向的设计、报告、验证和部署后监测。
- 需强调正义、不伤害、透明、问责和利益相关者参与等伦理原则。
局限性
- 为叙述性综述,非系统评价,可能存在选择偏倚。
- 烧伤特异性文献中关于亚组表现和人口构成的数据报告有限。
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
Artificial intelligence (AI) is increasingly integrated into burn care for triage, burn-depth assessment, prognostic scoring, pain management, and telemedicine-enabled resource allocation. These tools promise greater efficiency and precision, yet raise substantial ethical concerns regarding transparency, accountability, and bias. This narrative review synthesizes literature from 2010 to 2025 on AI, predictive modeling, and digital tools in burn care, supplemented by evidence from critical care, emergency medicine, radiology, dermatology, and oncology. Systematic searches of PubMed, Embase, and Scopus identified studies that reported algorithm development or deployment and discussed or allowed inference about equity, fairness, or interpretability. Across medical domains, algorithms frequently misclassified outcomes for racial and ethnic minorities, socioeconomically disadvantaged patients, women, older adults, and individuals with complex comorbidities, while burn-specific models rarely evaluated subgroup performance or reported demographic composition. Common problems included unrepresentative datasets, opaque modeling pipelines, and the absence of formal bias audits. Ethical analyses were fragmented and seldom grounded in established frameworks of biomedical ethics or AI governance. This review argues that current trajectories risk embedding and amplifying inequities in an already vulnerable burn population. It proposes concrete strategies for fairness-oriented design, reporting, validation, and post-deployment monitoring, emphasizing justice, nonmaleficence, transparency, accountability, and stakeholder engagement. Responsible adoption of AI in burn care will require moving beyond technical performance alone toward explicit attention to equity and ethical safeguards throughout the model life cycle.
引用
- Khorsandi J, Ahmed AB, Mirharooni J, et al. Artificial Intelligence and Algorithmic Bias in Burn Care: A Literature Review of Ethical Challenges. J Burn Care Res. 2026. PMID: 42340329. doi:10.1093/jbcr/irag105
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。