BACKGROUND
Although the evaluation of left ventricular ejection fraction (LVEF) in patients with atrial fibrillation (AF) or atrial flutter (AFL) is crucial for appropriate medical management, echocardiography is often required.
OBJECTIVE
This study aimed to investigate deep learning approaches to predict reduced LVEF (<50%) in patients with AF/AFL electrocardiograms (ECGs) and easily obtainable clinical information.
METHODS
Patients with 12-lead ECG of AF/AFL and echocardiography were divided into those with LVEF<50% and ≥50%. A deep learning model was based on a convolutional neural network. ECG signals, ECG features, and clinical features (demographic information, comorbidities, blood cell counts, and blood test results) were collected for training. A hold-out test dataset was constructed using a different recruitment period. Five-fold cross-validation and calibration plots were used to evaluate performance.
RESULTS
15,683 patients were analyzed (mean age, 70.0±11.7 years; 61.2% men), with 82.2% having LVEF≥50% and 17.8% having LVEF<50%. Patients with reduced LVEF were more likely to be men and have diabetes, ischemic heart disease, and chronic kidney disease. Using ECG signals alone, the model predicted reduced LVEF with the area under the receiver operating characteristics curve (AUROC) of 0.80, area under the precision-recall curve (AUPRC) of 0.51, sensitivity of 0.70, and specificity of 0.75. Additional training with ECG features and clinical features significantly improved AUROC (0.82 vs. 0.80, p=0.041), AUPRC (0.55 vs. 0.51, P<0.001), and sensitivity (0.77 vs. 0.70, P<0.001). Among the subgroups, the best overall performance was observed in patients with ischemic heart disease (AUROC of 0.87, AUPRC of 0.91, F1-score of 0.79, and sensitivity of 0.94).
CONCLUSIONS
The deep learning prediction of reduced LVEF was feasible using AF/AFL ECGs and readily available clinical information. This model might be helpful for early determination of optimal therapeutics for patients with AF/AFL.
CLINICALTRIAL
None.