CurrentCardio
Get Weekly

Heart failure & cardiomyopathiesCohort study

Artificial Intelligence–Enabled Prediction of Heart Failure Risk From Single-Lead Electrocardiograms

JAMA CardiolPublished

  • Retrospective cohort study
  • multinational
  • 3 cohorts (Yale New Haven Health System (YNHHS), UK Biobank (UKB), ELSA-Brasil)
  • Data analysis September 2023 to February 2025
Read the original article ↗

Authors' conclusion

Across multinational cohorts, a noise-adapted AI-ECG model estimated HF risk using lead I ECGs

Study context

Population
Individuals without heart failure (HF) at baseline with outpatient electrocardiograms (ECGs); lead I ECGs isolated
Participants
192,667 / 42,141 / 13,454
Exposure
AI-ECG vs PCP-HF and PREVENT risk equations for new-onset HF
  • YNHHSn = 192,667
  • UKBn = 42,141
  • ELSA-Brasiln = 13,454
Outcome
New-onset HF (first HF hospitalization); discrimination of AI-ECG-defined LVSD risk

Main finding

As reported in the abstract

Discrimination (C statistic) of AI-ECG for new-onset HF

YNHHS

0.723

C statistic · 95% CI 0.694–0.752

UKB

0.736

C statistic · 95% CI 0.606–0.867

ELSA-Brasil

0.828

C statistic · 95% CI 0.692–0.964

Positive AI-ECG screen for left ventricular systolic dysfunction (LVSD): higher HF risk
3- to 7-fold
Each 0.1 increment in model probability: higher hazard across cohorts (independent of age, sex, comorbidities, competing risk of death)
27% to 65%

Observational data: associations, not effects.

Further findings

Incident HF

median (IQR) follow-up YNHHS 4.6 (2.8-6.6) years · UKB 3.1 (2.1-4.5) years · ELSA-Brasil 4.2 (3.7-4.5) years

Incremental discrimination of AI-ECG added to PCP-HF and PREVENT

Harrel C statistic difference, added to PCP-HF
0.080-0.107
Harrel C statistic difference, added to PREVENT
0.069-0.094
Integrated discrimination improvement vs PCP-HF
0.091 to 0.205
Integrated discrimination improvement vs PREVENT
0.068 to 0.192
Net reclassification improvement vs PCP-HF
18.2% to 47.2%
Net reclassification improvement vs PREVENT
11.8% to 47.5%
  • ranges across cohorts

Original publication

Artificial Intelligence–Enabled Prediction of Heart Failure Risk From Single-Lead Electrocardiograms. JAMA Cardiol. Published 1 June 2025. doi:10.1001/jamacardio.2025.0492
Read the original article ↗PubMedCite RIS · BibTeX

Summarised from the published abstract; every number is checked against it