AI turns routine ECGs into stroke risk predictors

 Is your competent? doctor up-to-date and bringing this into the hospital?

AI turns routine ECGs into stroke risk predictors

Using 12-lead ECGs from thousands of patients, researchers developed and validated an artificial intelligence (AI) model, dubbed ECG2Stroke, that accurately predicted 10-year risk of ischaemic stroke and demonstrated discrimination comparable to the established Framingham Stroke Risk Profile.

The model’s risk signals were strongly associated with atrial electrical abnormalities -- particularly P-wave features -- and showed a specific link to cardioembolic stroke.

The findings were published in the Journal of the American College of Cardiology.

“Existing tools to identify which patients are at the highest risk of stroke often require cumbersome clinical score calculations, are not easily scalable, and are therefore not used widely in routine practice,” said Rahul Mahajan, MD, Mass General Brigham, Boston, Massachusetts.

To find an alternative, the researchers developed and validated a deep learning model (ECG2Stroke) using 12-lead ECG data from more than 101,000 patients at Massachusetts General Hospital to estimate 10-year risk of ischemic stroke, integrating neural network outputs with age and sex in a survival model. The model was then externally tested across independent cohorts from Brigham and Women’s Hospital and Beth Israel Deaconess Medical Center, where its performance was assessed for discrimination, calibration, and comparison against the established Framingham Stroke Risk Profile.

Across validation datasets totalling tens of thousands of patients and thousands of stroke events, ECG2Stroke demonstrated moderate predictive performance with area under the receiver operating characteristic curve values around 0.77 to 0.80 and low calibration error, performing similarly to the Framingham score.

Model interpretability analyses indicated that risk predictions were driven in part by P-wave features on the ECG and were strongly associated with cardioembolic stroke, suggesting the system may be capturing markers of underlying atrial pathology relevant to stroke risk stratification.

“If confirmed after prospective, real-world studies, tools like this could identify which patients should be prioritised for intensive prevention efforts,” said coauthor Shaan Khurshid, MD, Mass General Brigham Heart and Vascular Institute. “The tool could also be helpful in driving future mechanistic research into abnormalities in the upper chambers of the heart and links to stroke.”

Reference: https://www.jacc.org/doi/10.1016/j.jacc.2026.03.084

SOURCE: Mass General Brigham