Comprehensive Summary
In this study, researchers developed a convolutional neural network that can predict an atrial fibrillation event during sinus rhythm from standard 10-second, 12-lead electrocardiographs (ECGs). 649,931 sinus rhythm ECGs from 180,922 patients were obtained from the Mayo Clinic ECG Laboratory and were determined to be positive or negative for atrial fibrillation, with the positive criterion encompassing any ECG with at least one atrial fibrillation or atrial flutter present. The ECGs were divided in a 7:1:2 ratio for training, internal validation, and testing sets, respectively. For patients with multiple ECGs, a “window of interest” was defined as any ECGs taken 31 days prior to a patient's first detected atrial fibrillation. The neural network’s statistical optimization was performed using the Keras package. Diagnostic measurements including area under the curve (AUC) of receiver operating characteristic (ROC), and accuracy, were taken for both the first ECG score in the window of interest (main analysis) and the highest score in the window of interest (secondary analysis). The network proved to have high AUC and accuracy, with the main analysis yielding an AUC of 0.87 and accuracy of 79.4%, and the secondary analysis yielding an AUC of 0.90 and accuracy of 83.3%.
Outcomes and Implications
Almost a third of embolic strokes occur with no attributable cause, and many of these such strokes are associated with atrial fibrillation. Early detection of atrial fibrillation is key to preventing these strokes, as anticoagulant use has only shown to be helpful prior to a stroke event and has even been associated with an increased risk of bleeding if administered after. However, current atrial fibrillation screening methods are quite challenging and require prolonged monitoring, and no ECG structure offers a high enough predictive value to be of significance in a clinical setting. A neural network can detect subtle ECG characteristics to accurately predict atrial fibrillation presence, which can allow patients to receive anticoagulant treatment sooner and prevent a future stroke.