AI uses 12-lead ECGs to predict long-term stroke risk

 Do you really think your competent? doctor can get this into your hospital and accurately predict your next stroke without being confused by the factors that created your current stroke?

AI uses 12-lead ECGs to predict long-term stroke risk

Researchers have developed a new artificial intelligence (AI) model capable of reading 12-lead electrocardiograms (ECGs) and predicting a patient’s long-term stroke risk, sharing their findings in JACC.[1]

"Stroke remains a leading cause of death and disability worldwide,” wrote first author Rahul Mahajan MD, PhD, a researcher with the department of neurology at Brigham and Women’s Hospital (BWH), and colleagues. “Although age-adjusted rates of stroke have declined, the absolute number of strokes occurring annually continues to increase. Recurrent strokes account for only one-fourth of stroke events, and ischemic stroke is the most common type, highlighting a gap in primary prevention.”

The group trained its AI model, a convolutional neural network, with ECG data from more than 100,000 patients. The mean age was 57 years old, and 52% of patients were men. Patients were treated at BWH, Massachusetts General Hospital (MGH) or Beth Israel Deaconess Medical Center (BIDMC).

Overall, the AI model—ECG2Stroke—was associated with an area under the ROC curve (AUC) of 0.795 for MGH patients, 0.774 for BWH patients and 0.772 for BIDMC patients. In all cases, those AUC totals were comparable to the Framingham Stroke Risk Profile. 

In addition, the authors noted that their AI model was effective when looking at patients with or those without atrial fibrillation. 

“ECG2Stroke demonstrated consistent predictive utility across three test sets, including two large independent datasets spanning clinically and demographically varied populations,” the authors wrote. “Among individuals with available clinical data, ECG2Stroke provided similar discrimination of stroke risk compared with the validated FSRP clinical risk model. ECG2Stroke showed strong associations with clinical and ECG correlates of atrial dysfunction as well as cardio-embolic stroke, pointing to plausible risk mechanisms related to atrial cardiopathy.”

Mahajan et al. also explored which ECG features appeared to influence a patient’s stroke risk the most. 

“Examination of ECG neural network saliency maps demonstrated that variation in the ECG waveform near the region of the P-wave had the greatest effect on model predictions,” the authors wrote. “We examined associations of the ECG neural network component of ECG2Stroke with available ECG-based atrial substrate markers and found moderate correlation with PR interval and modest, but significant correlations with P-wave duration and P-wave axis.”

Reviewing these data, the group highlighted the potential of using ECG-based AI algorithms for “scalable stroke risk stratification.”

“Although all patients may benefit from lifestyle choices to promote cerebrovascular health, long-term stroke risk estimation offers the potential to efficiently prioritize individuals for intensification of general cardiovascular and stroke-specific primary prevention efforts, including improved control of known atherosclerotic cardiovascular disease risk factors or consideration of rhythm monitoring.”

Click here for the full study in JACC, an American College of Cardiology journal.