儿科临床诊疗中生成式人工智能工具开发与实施建议:政策声明
该政策声明指出生成式人工智能(GenAI)在儿科临床决策支持、文档记录和医学教育中有应用前景,但真实世界验证有限,且在大语言模型(LLM)应用于儿科时表现常不如成人专科。声明围绕准确性、偏倚、隐私、监管和公平性提出开发、机构治理与监管层面的建议,强调儿科特异性评估、上市后监测和透明披露。
要点
- 开发者应优先使用多样化儿科数据集,主动处理偏倚并落实数据隐私与安全保护。
- 医疗机构应建立严格验证方案、遵守隐私法规并保持清晰的人类监督。
- 监管机构应强制儿科特异性评估和上市后监测,治理框架需支持跨学科合作与GenAI素养教育。
核心结论
- GenAI在儿科临床诊疗中具有变革潜力,但须通过适应性政策确保伦理、公平和循证实施。
研究设计与样本
- 美国儿科学会(Pediatrics)发布的政策声明,基于对GenAI在儿科应用现状的综述提出建议。
主要发现
- LLM在儿科应用中的表现常不如成人医学专科,引发对其在儿童和青少年照护中就绪度的疑问。
- 偏倚或非代表性训练数据可能加剧不同种族、民族、性别、语言、能力和社会经济地位患者间的健康差异。
关键数值与推荐
- 开发者应优先使用多样化儿科数据集、主动处理偏倚并实施强数据隐私与安全保护。
- 医疗机构应建立严格验证方案、确保遵守隐私法规并保持清晰的人类监督。
- 监管与监督机构应执行儿科特异性评估并要求上市后监测。
- 治理框架应支持跨学科合作、透明度和教育举措,使儿科医生具备GenAI素养。
- 对患者照护中GenAI的参与进行合理披露,以建立和加强信任。
临床怎么用
- 儿科医生应了解GenAI的局限性和偏倚风险,在临床使用中保持人类监督并合理披露GenAI参与。
- 机构应制定验证和隐私合规流程,教育项目应提升儿科医生GenAI素养。
局限性
- GenAI在儿科的真实世界验证仍然有限,相关担忧涉及准确性、偏倚、可靠性、可持续性和耐久性。
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
The integration of generative artificial intelligence (GenAI) into pediatric health care offers exciting opportunities alongside critical challenges. GenAI tools, like large language models (LLMs), show promise in several health care applications, including but not limited to clinical decision support, documentation, and medical education across a wide range of pediatric subspecialties. However, real-world validation remains limited, and concerns persist around accuracy, bias, reliability, sustainability, and durability. Studies indicate that LLMs, when applied to pediatrics, often underperform compared with adult medical specialties, raising questions about their readiness for use in the care of children and adolescents. Moreover, the risk of exacerbating health disparities among patients of various races, ethnicities, genders, languages, abilities, and socioeconomic statuses because of biased or nonrepresentative training data underscores the need for rigorous oversight and accountability. This policy outlines recommendations for the safe, equitable, and effective use of GenAI in pediatric settings. Developers should prioritize diverse pediatric data sets, proactively address bias, and implement strong data privacy and security safeguards. Health care institutions must establish rigorous validation protocols, ensure compliance with privacy regulations, and maintain clear human oversight. Relevant regulatory and oversight bodies should enforce pediatric-specific evaluations and require postmarket surveillance. Governance frameworks must support interdisciplinary collaboration, transparency, and education initiatives to equip pediatricians with GenAI literacy. Reasonable disclosure of GenAI involvement in patient care is essential to establish and strengthen trust. GenAI policies pertaining to the care of children and adolescents must remain adaptive to ensure ethical, equitable, and evidence-based implementation. Despite these numerous challenges, GenAI holds transformative potential to improve pediatric health outcomes, and this policy seeks to empower developers, researchers, and clinicians to responsibly build and harness this technology.
引用
- Suresh S, Hunter RB, Periard B, et al. Recommendations for the Development and Implementation of Generative Artificial Intelligence Tools in Pediatric Clinical Care: Policy Statement. Pediatrics. 2026. PMID: 42827276. doi:10.1542/peds.2026-079037
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。