This study uses a deep learning model to predict the risk of atrial fibrillation from 12-lead ECGs. Their neural network, ECG-AI, has been trained with over 45,000 ECG’s from patients receiving longitudinal primary care at Massachusetts General Hospital and it has been validated in external cohorts from Brigham and Women’s Hospital and the UK Biobank as well. ECG-AI’s findings were assessed in comparison to a pre-existing clinical risk assessment of atrial fibrillation known as the Cohorts for Aging Research and Genomic Epidemiology - Atrial Fibrillation (CHARGE-AF). Each model was assessed both separately and together in a combined CH-AI model where the discrimination, calibration, and reclassification of each model was measured. The ECG-AI model performed comparably to CHARGE-AF (AUC ~0.82 vs 0.80 at MGH) and the combined CH-AI model achieved the highest discrimination across all test sets (AUC up to 0.84). Further analysis determined the P-wave region of the ECG to be the most determining feature for ECG-AI’s prediction. The study concluded that ECG-AI offers more accurate results in identifying the risk of atrial fibrillation within this patient population.