Electrocardiogram-Based Artificial Intelligence to Identify Coronary Artery Disease
Authors' conclusion
Artificial intelligence-enabled analysis of the ECG may facilitate identification of individuals with possible undiagnosed CAD and inform downstream testing and preventive measures.
Study context
- Population
- Individuals with electrocardiograms (ECGs) in a primary care setting; training at Massachusetts General Hospital (MGH), testing at MGH, Brigham and Women's Hospital (BWH) and UK BiobankECG2CAD trained on 764,670 ECGs representing 137,199 individuals at MGH, evaluated in 3 test sets · Training (MGH, individuals) 137,199 · MGH test set (6,051 cases) 18,706 · BWH test set (27,898 cases) 88,270 · UK Biobank test set (1,509 cases) 42,147
- Index test or model
- ECG2CAD compared against model of age and sex, and Pooled Cohort Equations, for discrimination of prevalent coronary artery disease (CAD) defined using diagnostic codes
- Target condition
- Prevalent CAD defined using diagnostic codes
Methodological details3 test sets: MGH, BWH, UK Biobank, BWH primary care cohort for incident CAD-related events
- 3 test sets: MGH, BWH, UK Biobank
- BWH primary care cohort for incident CAD-related events
Performance
As reported in the abstract
| Test or model | AUROC | Area under the precision-recall curve (AUPRC) |
|---|---|---|
| MGH | 0.782 | 0.639 |
| BWH | 0.747 | 0.588 |
| UK Biobank | 0.760 | 0.155 |
AUROC
MGH0.782
BWH0.747
UK Biobank0.760
Area under the precision-recall curve (AUPRC)
MGH0.639
BWH0.588
UK Biobank0.155
Further findings
| Outcome | HR (95% CI) | |
|---|---|---|
| Adverse events (myocardial infarction, heart failure, all-cause mortality), BWH primary care cohortHighest quintile of ECG2CAD riskvs Low-risk groupMyocardial infarction | 5.59 (4.76–6.56) | |
| Adverse events (myocardial infarction, heart failure, all-cause mortality), BWH primary care cohortHighest quintile of ECG2CAD riskvs Low-risk groupHeart failure | 10.49 (7.96–13.84) | |
| Adverse events (myocardial infarction, heart failure, all-cause mortality), BWH primary care cohortHighest quintile of ECG2CAD riskvs Low-risk groupAll-cause mortality | 2.68 (2.32–3.10) | |
| 0.050.21520Hazard ratio (log scale) |
Incremental discrimination of prevalent CAD vs models based on age and sex or Pooled Cohort Equations (MGH and BWH)
- ECG2CAD discriminated prevalent CAD incrementally vs models based on age and sex or Pooled Cohort Equations in MGH and BWH
- P < 0.01
Model performance across subgroups (BWH primary care subset)
- Model performance was consistent across subgroups
Original publication
Electrocardiogram-Based Artificial Intelligence to Identify Coronary Artery Disease. JACC Adv. Published 31 July 2025. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →