Hot off the Press

Debra L. Beck, MSc, and Eugene Braunwald, MD

Date Published: June 30, 2025

Current heart failure (HF) risk stratification strategies require comprehensive clinical evaluation, including laboratory testing and detailed clinical assessments, limiting their accessibility and scalability. In this study, artificial intelligence (AI) applied to electrocardiogram (ECG) images was examined as a strategy to predict HF risk.

This was a multinational cohort study conducted across three demographically and geographically distinct populations, all without baseline HF: the Yale New Haven Health System (YNHHS; n=231,285), UK Biobank (UKB; n=42,141), and the Brazilian Longitudinal Study of Adult Health (ELSA-Brasil; n=13,454). The AI-ECG model was a previously developed deep learning algorithm trained to detect left ventricular systolic dysfunction from 12-lead ECG images, deployed without any further modification or fine-tuning.

A positive AI-ECG screen was associated with dramatically elevated risk of incident heart failure across all three cohorts. In YNHHS, over a median follow-up of 4.5 years, 4,472 patients (1.9%) developed primary HF. After adjusting for age and sex, a positive AI-ECG screen conferred a 3.88-fold higher risk of incident HF (p<0.001). The association remained robust after accounting for the competing risk of death, with an adjusted hazard ratio of 3.54 (p<0.05).

In the UK Biobank, despite only 46 HF events (0.1%) over 3.1 years of follow-up, the effect was even more pronounced, with a positive screen associated with a 12.85-fold higher risk (p<0.001). Similarly, in ELSA-Brasil, with 31 HF events over 4.2 years, the age- and sex-adjusted hazard ratio was 23.50 (p<0.001).

The study demonstrated a clear dose-response relationship between AI-ECG probability scores and HF risk. In YNHHS, each 0.1 increment in model output probability was associated with a 36% higher hazard of incident HF (adjusted HR 1.36, 95% confidence interval [CI] 1.35-1.38). Screen-positive patients with probabilities between 0.1-0.5 had a 3.31-fold higher risk (95% CI 3.08-3.55), while those with probabilities between 0.5-1.0 had a 7.11-fold higher risk (95% CI 6.42-7.88) compared to screen-negative patients.

The AI-ECG model showed relative specificity for HF risk compared to other cardiovascular outcomes. While positive screens were associated with increased risk of acute myocardial infarction, stroke, and death, the magnitude of association was consistently smaller than for HF, suggesting the model captures heart failure-specific pathophysiology rather than general cardiovascular risk.

Summary

This multinational study demonstrates that an AI model applied to 12-lead ECG images can effectively identify individuals at elevated risk of developing HF, representing a novel digital biomarker for HF risk stratification. The approach offers significant advantages in terms of accessibility, scalability, and practical implementation compared to existing risk assessment tools, potentially enabling widespread screening and early identification of high-risk individuals who could benefit from preventive interventions.

References

https://pubmed.ncbi.nlm.nih.gov/39804243

Braunwald’s Heart Disease: A Textbook of Cardiovascular Medicine remains the most trusted reference in the field and the leading source of reliable cardiology information for practitioners and trainees worldwide. Written and edited by global experts in the field, this award-winning text is an unparalleled multimedia reference for every aspect of this complex and fast-changing area. Dr. Braunwald curates a monthly online update program featuring “Hot off the Press” and “Late-Breaking Clinical Trials.”

Learn more about Braunwald’s Heart Disease: A Textbook of Cardiovascular Medicine