ECG2Stroke Predicts 10-Year Stroke Risk From 10-Second ECGs, Matching Standard Scores
Updated
Updated · Earth.com · May 19
ECG2Stroke Predicts 10-Year Stroke Risk From 10-Second ECGs, Matching Standard Scores
4 articles · Updated · Earth.com · May 19
More than 200,000 ECGs were used to train and validate ECG2Stroke, a deep-learning model that estimates 10-year ischemic stroke risk from a routine 10-second 12-lead ECG plus age and sex.
Across three hospitals, the model scored about 0.78 on a 0-to-1 accuracy scale, roughly matching the long-used Framingham Stroke Risk Profile without requiring blood pressure, cholesterol, diabetes or smoking data.
Cardioembolic stroke drove the signal: patients with high ECG2Stroke scores had more than double the odds of that subtype, while links to other stroke mechanisms largely disappeared.
P-wave patterns in the ECG carried much of the model's weight, pointing to atrial dysfunction as a likely source of the risk signal even in patients without documented atrial fibrillation.
The JACC study suggests routine ECGs could help clinics flag patients for earlier stroke prevention, though prospective real-world trials are still needed before broad clinical use.
Your ECG predicts a future stroke. Should you start powerful drugs now, even if you feel perfectly healthy?
A new AI predicts stroke risk for wealthy nations. Will this medical marvel leave the rest of the world further behind?
Revolutionizing Stroke Prevention: AI-Driven 10-Year Risk Prediction from Standard ECGs
Overview
ECG2Stroke is a groundbreaking AI model developed by Mass General Brigham and the Broad Institute, marking a major transformation in stroke risk prediction. Unlike traditional methods that rely on complex and hard-to-scale clinical scores, ECG2Stroke uses just a standard 10-second ECG, age, and sex to identify people at high risk of ischemic stroke. This streamlined and accessible approach overcomes the practical challenges of older tools, making it easier to integrate into routine medical practice and enabling early, large-scale identification of at-risk individuals.