AI Heart Age Models Predict Death Risk Better Than Chronological Age
Three AI models analyzing ECGs showed remarkable agreement in predicting cardiovascular events and mortality risk.
Summary
Researchers tested three AI models that estimate "heart age" from electrocardiograms in over 7,000 people from Norway's Tromsø Study. All three models showed strong agreement and accurately predicted heart attacks, strokes, and death risk. People whose heart age exceeded their chronological age by one standard deviation had 27-48% higher risk of cardiovascular events and death. The models were within 6-8 years of actual age on average, demonstrating that ECG-based heart age is a reliable biomarker for cardiovascular health across different AI approaches.
Detailed Summary
Your heart's electrical activity may reveal more about your longevity than your birth certificate. This groundbreaking study validates that AI-powered "heart age" calculations from simple electrocardiograms can accurately predict cardiovascular disease and death risk.
Researchers analyzed 7,108 participants from Norway's renowned Tromsø Study, comparing three different convolutional neural networks that estimate heart age from ECG patterns. Despite being trained on different datasets and using different architectures, all three AI models showed remarkable consistency.
The models predicted chronological age within 6-8 years on average, with correlations around 0.71-0.73. Most importantly, when someone's estimated heart age exceeded their chronological age, it strongly predicted future health problems. For every standard deviation increase in this "delta age" gap, participants faced 27-50% higher risks of heart attacks, strokes, cardiovascular death, and all-cause mortality.
This validation across multiple AI systems suggests ECG-based heart age is a robust biomarker that could revolutionize preventive medicine. Unlike expensive imaging or blood tests, ECGs are quick, inexpensive, and widely available. The technology could enable early identification of cardiovascular risk, allowing for timely interventions through lifestyle changes, medications, or closer monitoring.
The study's strength lies in testing multiple independent AI models in a real-world population, proving the concept's reliability. However, the cohort was predominantly Norwegian, and longer follow-up periods would strengthen the findings. This research brings us closer to personalized longevity medicine where simple, accessible tests reveal biological age and guide targeted interventions.
Key Findings
- Three AI models estimated heart age within 6-8 years of chronological age with 86% agreement
- Higher heart age predicted 27-48% increased risk of heart attacks, strokes and death
- ECG-based heart age showed consistent predictive power across different AI architectures
- Simple electrocardiograms may serve as accessible biomarkers for cardiovascular aging
Methodology
Observational study of 7,108 participants from Norway's Tromsø Study cohort. Researchers compared three published convolutional neural networks for ECG-based heart age estimation. Cox regression models assessed associations with cardiovascular outcomes and mortality over follow-up period.
Study Limitations
Study population was predominantly Norwegian, potentially limiting generalizability to other ethnicities. Follow-up duration not specified, and longer observation periods would strengthen mortality predictions. External validation in diverse populations needed.
Enjoyed this summary?
Get the latest longevity research delivered to your inbox every week.