人工智能与目标试验模拟:迈向可扩展且可信的真实世界证据
本文为评论文章,探讨人工智能如何支持目标试验模拟(TTE)的各个组成部分,以生成更可扩展和可信的真实世界证据(RWE),同时指出此类应用可能引入的方法学风险、威胁因果有效性的失败机制及所需的保障措施。文章还提出了实证评估和方法学开发的优先方向,以支持符合临床和监管期望的AI增强RWE生成。
要点
- AI可支持TTE从定义研究人群到分析结局的多个环节,提升RWE生成的可扩展性。
- 应用AI于TTE需警惕偏倚、缺乏透明度和过度依赖自动化方法等风险。
- AI应辅助而非替代研究者,需适当验证、人工监督和保障措施以保持方法学有效性。
核心结论
- AI可支持TTE的不同组成部分,使RWE生成更可扩展和可信,但需考虑方法学风险及保障措施。
研究设计与样本
- 本文为评论文章,系统审视AI在TTE中的应用潜力、方法学风险及保障需求。
主要发现
- AI可能支持TTE从定义研究人群到分析结局的多个环节。
- AI应用可能引入偏倚、缺乏透明度和过度依赖自动化方法等风险。
- 需识别威胁因果有效性的失败机制并采取保障措施以保持方法学有效性、透明度和人工监督。
临床怎么用
- AI应辅助而非替代研究者,应用时需适当验证、人工监督和保障措施。
局限性
- AI在TTE中的应用尚未得到系统关注,需优先开展实证评估和方法学开发。
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 OBJECTIVES: Randomized controlled trials remain the gold standard for causal inference but are often resource-intensive, time-consuming, and operationally complex to conduct. The growing availability of real-world data has created new opportunities for generating real-world evidence (RWE), and target trial emulation (TTE) provides a principled framework for designing observational studies that approximate a target trial. At the same time, artificial intelligence (AI) is increasingly applied in health research, yet its application within TTE has received limited systematic attention. CONCLUSION: In this Commentary, we examine how AI may support different components of TTE to enable more scalable and credible RWE generation, while considering the methodological risks such applications may introduce. We highlight potential failure mechanisms that may threaten causal validity and the safeguards needed to preserve methodological validity, transparency, and human oversight. We also outline priorities for empirical evaluation and methodological development to support AI-enhanced RWE generation aligned with clinical and regulatory expectations. PLAIN LANGUAGE SUMMARY: TTE is a way of using routinely collected health-care data to design observational studies that more closely resemble clinical trials. In this Commentary, we discuss how AI could support different components of TTE, from defining study populations to analyzing outcomes. We also highlight important risks, including bias, lack of transparency, and overreliance on automated methods. AI should support, rather than replace, researchers. With appropriate validation, human oversight, and safeguards, it may help make RWE more efficient, reproducible, and trustworthy.
引用
- Liu T, Wang Y. Artificial intelligence and target trial emulation: toward scalable and credible real-world evidence. J Clin Epidemiol. 2026. PMID: 42716449. doi:10.1016/j.jclinepi.2026.112501
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。