Heart tracing model ranks heart failure risk across populations
In UK Biobank, the model outperformed three established clinical risk scores, with 6.9% to 22.5% relative improvements in risk-ranking performance.
Limit: Prediction was assessed, not whether using the model improved care or prevented heart failure
Edited by Boris Djordjevic · AgingResearch.News
Researchers studied patients with high blood pressure, a previous heart attack, or both in a retrospective, cross-sectional observational analysis. They developed an artificial intelligence model using 168,925 patients from the Harvard–Emory dataset. It combined electrocardiograms, recordings of the heart’s electrical activity, with age and sex.
The team evaluated it internally in 18,769 patients and externally in three datasets: MIMIC-IV (29,197), Alberta (97,462) and UK Biobank (18,292). Its C-index, a measure of how well it ranked patients by risk, was 0.765 internally, 0.720 in MIMIC-IV, 0.725 in Alberta and 0.762 in UK Biobank. In UK Biobank, it outperformed three established clinical scores for predicting heart failure over five years. Those scores used conventional clinical information, medications and blood measurements.
Why it matters
For people with existing heart disease, estimating the risk of heart failure may help guide earlier preventive care. The authors suggested that widely available heart recordings could support this approach.
Caveats
This retrospective, cross-sectional analysis assessed prediction, not whether using the model improved care or prevented heart failure. The findings concern patients with high blood pressure, previous heart attacks or both, and were reported in a conference abstract.
