Artificial intelligence applied to electrocardiograms (ECG-AI) may offer a scalable way to identify individuals at high risk of developing heart failure (HF), potentially improving on current clinical risk estimators, according to a new pooled cohort analysis from the HeartShare/Accelerating Medicines Partnership (AMP) Heart Failure program.¹

The study aimed to assess whether ECG-AI models designed to detect systolic and diastolic dysfunction could enhance the prediction of incident HF when added to the PREVENT-HF (Predicting Risk of Cardiovascular Disease EVENTs–Heart Failure) clinical risk equation.²

Methodology

This analysis pooled baseline clinical and ECG data from 14,126 participants without prevalent cardiovascular disease from three large population-based cohorts: the Framingham Heart Study (FHS), the Multi-Ethnic Study of Atherosclerosis (MESA), and the Cardiovascular Health Study (CHS).¹

Researchers used two previously validated ECG-AI algorithms: one designed to detect systolic dysfunction (ECG-AI LEF, based on left ventricular ejection fraction ≤40%) and another for diastolic dysfunction (ECG-AI DD, based on grade 2 or 3 dysfunction). The risk of incident HF was estimated using these models and the PREVENT-HF equation, both individually and in combination. The primary outcome was the development of incident HF over 1, 3, 5, and 10 years.

Results

At baseline, 11.9% of participants screened positive on a composite ECG-AI model (positive on either ECG-AI LEF or ECG-AI DD).¹ Participants with a positive composite ECG-AI screen had a markedly higher risk of developing HF compared to those with a negative screen, with a risk ratio of 23.8 at 3 years and 10.6 at 10 years.

The addition of the ECG-AI composite model to the PREVENT-HF score significantly improved risk discrimination. The combined model achieved a C-statistic of 0.898 at 10 years for predicting incident HF. Furthermore, adding ECG-AI to PREVENT-HF resulted in significant net reclassification improvement (NRI). At a PREVENT-HF risk threshold of 20%, the one-directional NRI was 0.344 at 1 year and 0.327 at 10 years, correctly reclassifying a substantial portion of individuals as high risk who would have been missed by the clinical score alone.

In Practice

These findings suggest that integrating ECG-AI with established clinical risk scores can refine the identification of individuals at high near-term risk for HF. The study's authors concluded, "The addition of ECG-AI to PREVENT-HF improved discrimination of near-term HF risk. ECG-AI may enable population-level HF risk stratification and facilitate targeted prevention strategies."¹ This enhanced predictive ability could help target preventive therapies, such as SGLT2 inhibitors, to patients most likely to benefit, potentially improving population health outcomes.

This study was funded by the Accelerating Medicines Partnership Heart Failure (AMP HF) Project.

References

1. Desai AS, Pandey A, Suratekar R, et al. Predicting Heart Failure From 12-Lead ECGs Using AI: A HeartShare/AMP-HF Pooled Cohort Analysis. JACC. 2026;87(8):990-1005. https://doi.org/10.1016/j.jacc.2025.10.065

2. Khan SS, Coresh J, Pencina MJ, et al. Novel prediction equations for absolute risk assessment of total cardiovascular disease incorporating cardiovascular-kidney-metabolic health: a scientific statement from the American Heart Association. Circulation. 2023;148(24):1982-2004. https://doi.org/10.1161/CIR.0000000000001191

Disclaimer

The information presented in this article is for educational purposes only. Any quotes included reflect the opinions of the individual quoted, and do not necessarily reflect the views of the publisher. The publisher does not guarantee the accuracy or completeness of the content and accepts no responsibility for any errors, or any consequences arising from its use.