The following is a summary of “Deep Learning of Electrocardiograms in Sinus Rhythm From US Veterans to Predict Atrial Fibrillation,” published in the October 2023 issue of Cardiology by Yuan et al.  

Deep learning of electrocardiograms (ECGs) can detect atrial fibrillation early, preventing stroke. Researchers performed a retrospective study to evaluate the performance of deep learning models in predicting atrial fibrillation (AF) from sinus rhythm ECGs in a large and diverse patient population. 

The study obtained ECGs analyzed from 6 US VA hospital networks and 1 non-VA academic medical center developed a convolutional neural network to predict AF within 31 days of sinus rhythm ECGs. The model’s performance was evaluated on ECGs from all 7 sites.

The analysis included 907,858 ECGs from patients across 6 VA sites. Patients had a mean (SD) age of 62.4 (13.5) years, with 6.4% female and 93.6% male. Their mean (SD) CHA2DS2-VASc score was 1.9 (1.6). Regarding race and ethnicity, 0.2% were American Indian or Alaska Native, 2.7% Asian, 10.7% Black, 4.6% Latinx, 0.7% Native Hawaiian or Other Pacific Islander, 62.4% White, 0.4% of other race/ethnicity (which is not broken down into subcategories in the VA data set) and 18.4% unknown. At the non-VA academic medical center (72,483 ECGs), patients had a mean (SD) age of 59.5 (15.4) years, with 52.5% female and a mean (SD) CHA2DS2-VASc score of 1.6 (1.4). Their racial/ethnic breakdown was: 0.1% American Indian or Alaska Native, 7.9% Asian, 9.4% Black, 2.9% Latinx, 0.03% Native Hawaiian or Other Pacific Islander, 74.8% White, 0.1% other race/ethnicity, and 4.7% unknown.

In the testing of the deep learning model at VA sites, this achieved an AUROC of 0.86 (95% CI, 0.85-0.86), an accuracy of 0.78 (95% CI, 0.77-0.78), and an F1 score of 0.30 (95% CI, 0.30-0.31) for predicting AF within 31 days of a sinus rhythm ECG. At the non-VA site, the model performed even better with an AUROC of 0.93 (95% CI, 0.93-0.94), accuracy of 0.87 (95% CI, 0.86-0.88), and an F1 score of 0.46 (95% CI, 0.44-0.48). The model demonstrated good calibration with a Brier score of 0.02 across all sites. For individuals identified as high risk by the deep learning model, the number needed to screen to detect a positive case of AF was 2.47 for a testing sensitivity of 25% and 11.48 for a sensitivity of 75%. Importantly, the model’s performance remained consistent in Black female patients younger than 65 years or with CHA2DS2-VASc scores of 2 or greater. 

The study found deep learning of outpatient ECGs predicts AF in diverse populations, suggesting its potential for early AF screening. 

Source: jamanetwork.com/journals/jamacardiology/article-abstract/2810388