Prevention & lipidsCohort study
Deep learned representations of the resting 12-lead electrocardiogram to predict at peak exercise
- Retrospective cohort study
- external validation
Authors' conclusion
Deep learning-enabled analysis of the resting 12-lead ECG can estimate exercise capacity (V˙O2peak) at scale to enable efficient cardiovascular risk stratification.
Study context
- Population
- Individuals undergoing cardiopulmonary exercise testing (CPET) at Massachusetts General Hospital; external sample BWH; primary care patients without CPET for outcomesdevelopment set
- Participants
- 1,891 individuals
- Exposure
- Basic vs Basic + Electrocardiogram (ECG) Parameters vs Deep ECG-V˙O2
- Outcome
- Accuracy of V˙O2peak estimation from resting 12-lead ECG
Main finding
As reported in the abstract
V˙O2peak estimation accuracy (Deep ECG-V˙O2)
| Group | MGH Test r | MGH Test mean absolute error (MAE) | BWH Test r | BWH Test MAE |
|---|---|---|---|---|
| Deep ECG-V˙O2 | 0.845 (95% CI 0.817-0.870) | 5.84 (95% CI 5.39-6.29) | 0.552 (95% CI 0.509-0.592) | 6.49 (95% CI 6.21-6.67) |
- Deep ECG-V˙O2 outperformed Wasserman, Jones and FRIEND reference equations
- P < 0.01
Observational data: associations, not effects.
Further findings
| Outcome | HR | |
|---|---|---|
| Incident disease with Deep ECG-V˙O2 estimated V˙O2peak <14 mL/kg/min (84 718 primary care patients without CPET) — Estimated V˙O2peak <14 mL/kg/min vs Basic · incident AF · proportional hazards models | 1.36 (95% CI 1.21–1.54) | |
| Incident disease with Deep ECG-V˙O2 estimated V˙O2peak <14 mL/kg/min (84 718 primary care patients without CPET) — Estimated V˙O2peak <14 mL/kg/min vs Basic · MI · proportional hazards models | 1.21 (CI 1.02–1.45) | |
| Incident disease with Deep ECG-V˙O2 estimated V˙O2peak <14 mL/kg/min (84 718 primary care patients without CPET) — Estimated V˙O2peak <14 mL/kg/min vs Basic · HF · proportional hazards models | 1.67 (CI 1.49–1.88) | |
| Incident disease with Deep ECG-V˙O2 estimated V˙O2peak <14 mL/kg/min (84 718 primary care patients without CPET) — Estimated V˙O2peak <14 mL/kg/min vs Basic · death · proportional hazards models | 1.84 (CI 1.68–2.03) | |
| 0.512Hazard ratio (log scale) |
Accuracy in BWH Test excluding individuals with heart failure (HF)
- r (HF excluded, BWH Test)
- 0.628, 95% CI 0.567-0.682
- MAE (HF excluded, BWH Test)
- 5.97, 95% CI 5.57-6.37
Deep ECG-V˙O2 vs Wasserman, Jones and FRIEND reference equations
- Deep ECG-V˙O2 outperformed the Wasserman, Jones and FRIEND reference equations (comparisons of correlation)
- P < 0.01
Original publication
Deep learned representations of the resting 12-lead electrocardiogram to predict at peak exercise. Eur J Prev Cardiol. Published 5 October 2023. doi:10.1093/eurjpc/zwad321
Summarised from the published abstract; every number is checked against it