AI fitness estimates from heart recordings linked to death risk
In 5,398 patients, estimates were off by an average of 4.87 milliliters of oxygen per kilogram per minute compared with exercise-test measurements.
Our read
Biomarkers- Studied:
- Cohort study of 5,398 patients who underwent clinically indicated CPET, against the lowest quartile
- What changed:
- Death from any cause was 54% lower in the highest estimated fitness quartile than in the lowest
- The catch:
- The cohort was mostly White and excluded people with pacemaker-controlled heart rhythms
- Why it’s aging research:
- It builds or tests a tool for aging research.
Editor of record: Boris Djordjevic · AgingResearch.News
Patients aged 18 or older who underwent clinically indicated exercise testing formed an observational cohort used to develop an artificial intelligence model. It estimated maximum oxygen uptake—a measure of fitness—from a resting electrocardiogram, a recording of the heart’s electrical activity. The 5,398-patient test group averaged 54.4 years of age; 61.5% were male and 88.2% were White. The model outperformed two established age-based fitness equations.
Over a median follow-up of 7.2 years, 1,395 patients (25.8%) died. Patients in the highest quarter of estimated fitness had a 54% lower relative risk of death than those in the lowest quarter. This association remained after accounting for cardiovascular disease and fitness measured during exercise testing.
Why it matters
Fitness predicts cardiovascular disease and death. The work explores whether a resting heart test can help assess this aspect of long-term health without exercise testing.
Caveats
Patients with higher estimated fitness may have been healthier in other ways the analysis did not account for. Reported in a conference abstract, the results came from a mostly White clinical-testing population that excluded heart or lung transplant recipients and people with pacemaker-controlled heart rhythms, among other exclusions.
The paper
Mayo Clinic
AHA Scientific Sessions 2026 Abstract, Nov 2, 2026
Presented at AHA Scientific Sessions 2026, abstract 18023
Comments
Be the first to comment.