AHA Scientific Sessions 2026 AbstractNov 9, 2026, 00:49 GMT

Heart tracing model better flags high levels of heart scarring

It scored 0.674 versus 0.597 for a clinical model on a measure of how well it distinguished people with high heart muscle scarring.

Paper by Michael J. Chilazi, Danielle F. Pace, Youssef Makarem, +8Victor D. Nauffal · Massachusetts General Hospital · Brigham and Women's Hospital

A heart cross-section with extensive scarring sits beside a smaller, lightly scarred heart section and a group of upright, faceless people.

Our read

Biomarkers

Studied:: Cohort study of 8,048 individuals in the UK Biobank, against a comprehensive clinical model (Independent replication)
What changed:: AUROC for classifying high fibrosis burden was 0.674 with the model versus 0.597 for a clinical model
The catch:: The model estimated a scan-based proxy rather than measuring scar tissue directly

Editor of record: Boris Djordjevic · AgingResearch.News

Paper

UK Biobank participants—8,048 people averaging 65.7 years old, 52% female—formed the test group in an observational study of a heart-scarring prediction model. Researchers trained the model using over 50,000 pairs of magnetic resonance heart scans and electrocardiograms, recordings of the heart’s electrical activity. The model used these electrical recordings to estimate a scan-based measure of heart muscle scarring. It distinguished high scarring better than a model using clinical information.

Higher estimates were associated with existing cardiovascular disease and new disease over five years. These disease associations also held in an external group of 102,038 Mass General Brigham primary care patients averaging 57.4 years old, 58% female. The analyses accounted for demographics, other illnesses and medication use.

Why it matters

Heart muscle scarring contributes to cardiovascular disease, a concern as people age. The authors suggest that an accessible estimate may help assess cardiovascular risk and guide disease prevention.

Caveats

The model estimated a scan-based proxy rather than measuring scar tissue directly; external validation assessed disease associations rather than scarring estimates. The findings were reported in a conference abstract.