基于人工智能心电图检测未诊断肝硬化:一项实用性整群随机临床试验
这项实用性整群随机试验将98个初级保健团队分为干预组(可查看心电图机器学习结果,123名临床医生)和常规护理组(122名临床医生),共纳入15596名接受12导联心电图的成人。干预组在180天内新诊断晚期慢性肝病(CLD)的比例为1.0%,对照组为0.5%(OR 2.09,95%CI 1.22-3.55,P=0.007),提示基于心电图机器学习模型可能有助于在常规初级保健中发现晚期CLD病例。
要点
- 干预组晚期CLD新诊断率1.0%,对照组0.5%(OR 2.09,95%CI 1.22-3.55,P=0.007)。
- 心电图机器学习阳性患者中,干预组晚期CLD 4.4%,对照组1.1%(OR 4.37,95%CI 1.94-9.88,P<0.001)。
- 干预组任何纤维化检出率1.7%,对照组0.5%(OR 3.17,95%CI 1.86-5.40,P<0.001)。
核心结论
- 基于心电图机器学习模型,并结合基于风险因素的靶向检测,可能有助于在常规初级保健中发现晚期慢性肝病病例。
研究设计与样本
- 实用性整群随机临床试验,98个初级保健团队随机分为干预组(可查看心电图机器学习结果,123名临床医生)和常规护理组(122名临床医生)。
- 共15596名成人接受12导联心电图并符合纳入标准(干预组8034人,对照组7562人)。
- 主要终点为心电图后180天内经序贯肝病评估确认的新诊断晚期纤维化CLD。
主要发现
- 总体队列中,干预组晚期CLD新诊断率为1.0%,对照组为0.5%(OR 2.09,95%CI 1.22-3.55,P=0.007)。
- 心电图机器学习阳性患者中,干预组晚期CLD发生率为4.4%,对照组为1.1%(OR 4.37,95%CI 1.94-9.88,P<0.001)。
- 总体队列中,干预组任何纤维化检出率为1.7%,对照组为0.5%(OR 3.17,95%CI 1.86-5.40,P<0.001)。
- 心电图机器学习阳性患者中,干预组任何纤维化检出率为8.4%,对照组为1.1%(OR 8.03,95%CI 3.50-18.4,P<0.001)。
临床怎么用
- 作者认为,基于心电图的机器学习模型,并结合基于风险因素的靶向检测,可能有助于在常规初级保健中发现晚期CLD病例。
局限性
- 诊断率低于流行病学估计值,可能反映了临床医生对人工智能驱动建议的依从性存在差异。
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
Advanced chronic liver disease (CLD) affects 2-5% of the general population, and accessible screening tools are needed in primary care. Here we conducted a pragmatic trial to assess whether an electrocardiogram (ECG)-based machine learning (ECG-ML) model enables early detection of advanced CLD. In this trial, 98 primary care teams were cluster randomized to intervention (access to ECG-ML results; 123 clinicians) or usual care (122 clinicians). Clinicians in the intervention arm were notified of a positive ECG-ML result, indicating higher risk of advanced CLD. The primary endpoint was new diagnosis of CLD with advanced fibrosis within 180 days of ECG, confirmed by sequential liver disease assessments. A total of 15,596 adults underwent 12-lead ECGs as part of routine care and met inclusion criteria (N = 8,034 intervention and N = 7,562 control). The intervention significantly increased new diagnoses of advanced CLD in the overall cohort (1.0% versus 0.5% in the control arm; odds ratio (OR) 2.09, 95% confidence interval (CI) 1.22-3.55, P = 0.007). Among ECG-ML-positive patients, advanced CLD was more frequent in the intervention arm (4.4% versus 1.1%; OR 4.37, 95% CI 1.94-9.88, P < 0.001). The intervention also increased the detection of any fibrosis (secondary endpoint) in the overall cohort (1.7% versus 0.5%; OR 3.17, 95% CI 1.86-5.40, P < 0.001) and among ECG-ML-positive patients (8.4% versus 1.1%; OR 8.03, 95% CI 3.50-18.4, P < 0.001). The diagnostic yield below epidemiological estimates probably reflects variable clinician adherence to artificial intelligence-driven recommendations. These results demonstrate that an ECG-based machine learning model, followed by targeted testing based on risk factors, may aid case finding of advanced CLD in routine primary care. ClinicalTrials.gov registration: NCT05782283 .
引用
- Simonetto DA, Rushlow D, Liu K, et al. Detection of undiagnosed liver cirrhosis via AI-enabled electrocardiogram: a pragmatic, cluster-randomized clinical trial. Nat Med. 2025. PMID: 41408416. doi:10.1038/s41591-025-04058-y
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。