Clin Res Cardiol (2023).

Deep learning based prediction of 3-year atrial fibrillation incidence from 1-lead ECGs
J. P. Bremer1, M. Neyazi1, M. Knorr1, J. Brederecke1, F. M. Ojeda Echevarria1, S. Blankenberg1, R. Schnabel1
1Klinik für Kardiologie, Universitäres Herz- und Gefäßzentrum Hamburg, Hamburg;
Introduction:  Wearables capable of deriving biomarkers such as 1-lead ECGs become largely available, therefore offering the potential to screen for prevalent and incident cardiovascular disease within the population. Screening for prevalent atrial fibrillation (AF) from such devices is already widely applied. It has remained unclear whether predicting incident AF using 1-lead ECGs is possible. The assessment of  AF risk before clinical manifestation may permit prevention and early detection and thus slow or limit the onset and progression of the disease.

Purpose: Utilizing data from the UK-Biobank, we examined the possibility to predict three-year AF incidence from  single-lead ECGs.

Methods: The population-based UK-Biobank,  initiated in 2006, provides data from 500,000 participants. For this study 100,551 1-lead ECGs from 91,898 unique participants were utilized, including 57,897 4-lead resting ECGs during the initial resting phase of ergometry from 2006-2010 (Exam1; median age 56.7 years, 46.2% men) and 9,826 at the first follow-up from 2010-2013 (Exam2; median age 61.3 years, 49.0% men) and 32,828 12-lead resting ECGs during the follow-up imaging study from 2014-2020 (Exam3; median age 63.5 years, 48.4% men). The 12-lead and 4-lead ECGs share information from lead I, which was extracted and used for all further analyses. Here we employ a 1D-Convolutional Neural Network for predicting the 3-year risk of incident AF. We compare it with the clinically used CHARGE-AF Score. The Neural Network is employed in a nested 6-fold cross-validation scheme. The model is currently undergoing further external validation.

Results: Our model provided a good detection of 3-year incident AF at all three examinations (AUC 0.729, 0.727, 0.705), while the CHARGE-AF Score provides better results at Exam1 (0.773), worse at Exam2 (0.692) and equal at Exam3 (0.717). A combination of the CHARGE-AF Score and the Neural Network performed better at all three examinations (AUC 0.811, 0.741, 0.753) than the individual methods.

Conclusion: Using a Neural Network model to predict incident AF from 1-lead ECGs, we provide further evidence for the added value of clinically applied deep learning-based methods. By combining the model predictions on ECG data with the well established CHARGE-AF score, 3-year AF incidence can be predicted with high accuracy.