Arrhythmias & electrophysiology

ECG Signatures and Long-Term Ischemic Stroke Risk: A Deep Learning Analysis of 200,000 Patients

J Am Coll CardiolPublished Prognostic model development and validation study

Authors' conclusion

ECG-based artificial intelligence (AI) can predict 10-year ischemic stroke with performance similar to a validated clinical score, possibly by encoding markers of abnormal atrial substrate linked to cardioembolism.

Study context

Population
Patients receiving longitudinal care at Massachusetts General Hospital (MGH), with external samples from Brigham and Women's Hospital (BWH) and Beth Israel Deaconess Medical Center (BIDMC)development n = 101,496
Comparison
ECG2Stroke vs revised Framingham Stroke Risk Profile
Outcome
Incident ischemic stroke: discrimination (AUC) and calibration (ICI)
Follow-up
10 years
Methodological detailsdeep learning (convolutional neural network), external validation in 2 independent samples
  • Deep learning (convolutional neural network)
  • External validation in 2 independent samples

Performance

As reported in the abstract

10-year incident ischemic stroke, discrimination and calibration of ECG2Stroke
AUC, MGH Test (n = 4,771)
0.795
AUC, BWH (n = 68,884)
0.774
AUC, BIDMC (n = 29,882)
0.772
ICI, MGH Test
0.030
ICI, BWH
0.005
ICI, BIDMC
0.026

Further findings

OutcomeCause-specific HR (95% CI)
Cardioembolic stroke · per 1-SD of logit-transformed probability2.17 (1.64–2.87)
OutcomeECG2StrokeFSRPEstimate
10-year AUC vs FSRP (patients with available data)
AUC MGH/BWH Test0.7910.779
AUC BIDMC0.7450.728

10-year AUC vs Framingham Stroke Risk Profile (FSRP) (patients with available data)

ECG2Stroke similar to FSRP

Cardioembolic stroke

Not associated with noncardioembolic stroke

Subgroups incl. with and without AF

Stratification persisted across subgroups

Original publication

ECG Signatures and Long-Term Ischemic Stroke Risk: A Deep Learning Analysis of 200,000 Patients. J Am Coll Cardiol. Published 13 May 2026. doi:10.1016/j.jacc.2026.03.084

Summarised from the published abstract; every number is checked against it

Get them every week

New studies on this topic in My Week and in your Monday e-mail.

We e-mail you a sign-in link and a code. No password, no spam.