Heart-tracing model predicts two types of heart failure
In Massachusetts General Hospital’s test group, 10-year prediction scores were 0.87 for heart failure with reduced ejection fraction and 0.79 for preserved ejection fraction.
Our read
Biomarkers- Studied:
- Cohort study of individuals without prior HF, 10 years
- What changed:
- 10-year AUROC was 0.87 for incident HFrEF and 0.79 for HFpEF in the MGH test set
- The catch:
- The study did not report how closely predicted risks matched observed event rates
- Why it’s aging research:
- It bears on staying healthy in midlife and later life.
Editor of record: Boris Djordjevic · AgingResearch.News
People without previous heart failure were evaluated in three hospital test groups of 16,658, 75,006 and 29,686 in an observational study. Each group averaged 56 years of age; women made up 55%, 56% and 54%, respectively. Training used data from 97,214 people.
The artificial intelligence model used electrocardiograms, recordings of the heart’s electrical activity, to predict 10-year risk. The two heart failure types differed in ejection fraction, the share of blood expelled by the heart’s main pumping chamber per beat. Reduced meant below 50%; preserved meant at least 50%. The prediction scores measured separation of future cases from noncases. The model outperformed age and sex alone for both types. Among people with available data, it outperformed a model using heart failure risk factors for reduced ejection fraction and performed similarly for preserved ejection fraction.
Why it matters
Heart failure prevention is an important goal as people age. The authors suggest that predicting risk separately for its two forms may help tailor preventive efforts.
Caveats
The observational analysis tested predictions, not whether acting on them prevented heart failure. The conference abstract did not report how closely predicted risks matched observed event rates.
The paper
Massachusetts General Hosp; Harvard Med Sch
AHA Scientific Sessions 2026 Abstract, Nov 2, 2026
Presented at AHA Scientific Sessions 2026, abstract 13523
Comments
Be the first to comment.