Heart failure & cardiomyopathiesResearch article
Artificial Intelligence-Enabled ECG Analysis to Predict Incident Heart Failure
- develop and validate a generalizable and publicly available convolutional neural network
- validated in 3 test sets
Authors' conclusion
ECG2HF is a publicly available 12-lead ECG-based artificial intelligence model that discriminates the risk of future HF with favorable and consistent performance across 3 large health care samples from the northeastern United States. ECG2HF may enable efficient prioritization of high-risk individuals for HF-related preventive measures.
Study context
- Population
- Patients receiving longitudinal ambulatory care (development, Massachusetts General Hospital (MGH)) and individuals aged 30 to 79 years without heart failure (HF) (validation, MGH, Brigham and Women’s Hospital (BWH), Beth Israel Deaconess Medical Center (BIDMC))individuals in the 3 test sets, development in 94,636 patients · MGH test set 13,954 · BWH test set 54,396 · BIDMC test set 25,457
- Participants
- 93,868 patients
- Comparison
- ECG2HF vs Pooled Cohorts Equations to Prevent HF
Main finding
As reported in the abstract
Cumulative risk of HF · 10 years
MGH
4.6% (95% CI, 4.1-5.0)
BWH
5.0% (4.8-5.2)
BIDMC
4.4% (4.1-4.7)
Further findings
| Outcome | Values | Estimate |
|---|---|---|
| Improvement in area under the receiver operating characteristic curve, ECG2HF vs Pooled Cohorts Equations to Prevent HF · 10 years — ECG2HF vs Pooled Cohorts Equations to Prevent HF · MGH/BWH | improvement in area under the receiver operating characteristic curve 0.061 (CI 0.025–0.097) | |
| Improvement in area under the receiver operating characteristic curve, ECG2HF vs Pooled Cohorts Equations to Prevent HF · 10 years — ECG2HF vs Pooled Cohorts Equations to Prevent HF · BIDMC | improvement in area under the receiver operating characteristic curve 0.038 (CI −0.0096 to 0.086) | |
| Net reclassification (NRI) of 10-year HF risk, ECG2HF vs Pooled Cohorts Equations to Prevent HF · MGH/BWH | NRI 0.16 (CI 0.077–0.24) | |
| Net reclassification (NRI) of 10-year HF risk, ECG2HF vs Pooled Cohorts Equations to Prevent HF · BIDMC | NRI 0.23 (CI 0.10–0.35) | |
| Test set characteristics | MGH 441 events · individuals 13 954 · age 57±13 years · women 48% · BWH 1809 events · individuals 54 396 · age 57±13 years · women 55% · BIDMC 901 events · individuals 25 457 · age 57±13 years · women 53% |
- Net reclassification (NRI) of 10-year HF risk, ECG2HF vs Pooled Cohorts Equations to Prevent HF: <10%, 10% to 20%, ≥20% 10-year risk categories
Discrimination of 10-year incident HF by ECG2HF (area under the receiver operating characteristic curve)
- MGH
- 0.86 [0.84-0.87]
- BWH
- 0.85 [0.84-0.86]
- BIDMC
- 0.84 [0.83-0.86]
Original publication
Artificial Intelligence-Enabled ECG Analysis to Predict Incident Heart Failure. Circ Heart Fail. Published 23 February 2026. doi:10.1161/circheartfailure.125.013927
Summarised from the published abstract; every number is checked against it