A routine heart test may help flag stroke risk years before symptoms appear
A research team co-led by investigators at Mass General Brigham and the Broad Institute has developed an artificial intelligence model that uses a standard electrocardiogram, or ECG, to estimate a patient’s risk of stroke up to 10 years into the future. The system, called ECG2Stroke, was trained and validated on data from more than 200,000 patients and is designed to work from a single 10-second ECG alongside a patient’s age and sex.
The work points to a potentially scalable way to identify people who might otherwise be missed by traditional screening tools. Stroke prevention often depends on finding elevated risk early enough to act, but clinical risk scoring can be cumbersome and may not be used consistently in routine care. By relying on a widely available, non-invasive test that is already common in cardiology, the researchers argue that AI could help close that gap.
What the model learned from the ECG
Rather than depending on a long list of clinical variables, ECG2Stroke looks for subtle waveform patterns in the heart’s electrical activity. The researchers said the model performed similarly to a validated clinical risk score across hospitals and patient subgroups, despite using a much narrower set of inputs. That matters because ECGs are inexpensive, fast, and already embedded in everyday clinical workflows.
The model was developed with patient data from Massachusetts General Hospital and then tested in patients from Brigham and Women’s Hospital and Beth Israel Deaconess Medical Center. That multi-hospital validation gives the findings more weight than a single-site proof of concept, though it still stops short of real-world deployment.
Strongest signal: cardioembolic stroke
Among the most important findings was the model’s accuracy in predicting cardioembolic stroke, a subtype caused when blood clots form in the heart, then travel to the brain. The researchers said ECG features related to dysfunction in the atria, the upper chambers of the heart, had an outsized influence on predictions. That is clinically relevant because cardioembolic strokes can often be prevented with blood thinners if high-risk patients are identified in time.
In practical terms, the model appears to be picking up traces of cardiac vulnerability that may not be obvious in standard reading of an ECG. If those signals hold up in prospective studies, the tool could help clinicians prioritize patients for more intensive monitoring or preventive treatment.
Why this could matter in practice
The appeal of ECG2Stroke is not just its performance but its workflow fit. Existing stroke risk tools may be accurate, but they are not always easy to operationalize at scale. A system that can run automatically on an ECG already collected in clinical care could be deployed more broadly, especially in health systems looking for low-friction ways to identify preventable risk.
That does not mean the model is ready to change care on its own. The authors were explicit that prospective, real-world confirmation is still needed. Predictive performance in retrospective datasets is an important milestone, but not the same as proving that clinicians can use the tool effectively, safely, and equitably in live care settings.
There is also a wider question around how such a model would be used. Some patients identified as high risk may need follow-up rhythm monitoring, imaging, or more aggressive management of other cardiovascular risk factors. Others may benefit from watchful surveillance rather than immediate intervention. The value of the tool will depend not only on prediction accuracy, but also on how well it fits into decision-making pathways.
A step toward more passive prevention
Even with those caveats, the study adds to a growing body of work showing that AI can extract clinically meaningful signals from tests already sitting inside the health system. The ECG has long been used to diagnose acute or known heart problems. This research suggests it may also serve as a quiet forecasting tool for future neurological risk.
For stroke medicine, that is a compelling idea. Strokes are often devastating, and prevention is far more effective than treatment after the fact. If a 10-second ECG can help identify patients who deserve closer attention years before an event occurs, it could shift part of stroke prevention from reactive care to earlier, more routine screening. The next question is whether that promise survives contact with everyday medicine.
This article is based on reporting by Medical Xpress. Read the original article.
Originally published on medicalxpress.com
