Abstract

Aim: Left ventricular systolic dysfunction occurs frequently in diabetes, and is associated with high risk of deteriorating to heart failure, but it is often silent and echocardiographic screening of all patients is not possible. In this study, the researchers assessed an AI-powered ECG to detect hidden LVF in type-2 DM patients who had no symptoms of LVF. Methods: In a cohort of asymptomatic adults with type 2 diabetes a 12-lead electrocardiogram was paired with echocardiography. A convolutional neural network was trained using only the ECG to classify an LFVEF < 40% and its performance evaluated with a held-out test set when compared to echocardiographic measurements. Results: 8.6% of 4,200 patients had an EF of 40% or lower. The model was obtained with 0.88 (95% CI 0.84 to 0.91) area under the curve, 82% sensitivity, 78% specificity, and consistent performance with sex, age, glycaemic control and diabetes duration, in the test set. Conclusions: AI-based ECG was found to be an effective tool for accurate screening of occult left ventricular dysfunction in asymptomatic diabetic patients, and can be an affordable, scalable method of triaging confirmatory echocardiography.

Keywords: Artificial intelligence, Electrocardiogram, Deep learning, Left ventricular dysfunction, Diabetes mellitus, Screening