B-in01-07 an artificial intelligence-based model for prediction of atrial fibrillation from single-lead sinus rhythm ecgs enabling screening

Screening for atrial fibrillation (AF) is currently recommended in the ESC guidelines. To identify individuals that might benefit most from screening, age cut-offs or addition of co-morbidity/biomarkers are often used. A machine-learning model could aid in identifying patterns in normal sinus rhythm ECGs that indicate high risk for having intermittent AF. The aim of this research project was to study if an artificial intelligence-based model could predict who would benefit from prolonged screening for AF using intermittent ECGs based on a single sinus rhythm one-lead ECG. A convolutional neural network model was trained and evaluated using data from three AF screening studies: STROKESTOP I, STROKESTOP II and SAFER. In all three studies, one-lead intermittent ECG was used for at least two weeks in order to detect AF. A total of 443,875 ECGs from 13,080 patients aged ≥65 years were included in the analysis. The training set consisted of 247,463 ECGs from 80% of participants in SAFER and STROKESTOP II. ECGs from the remaining 20% of participants in STROKESTOP II and SAFER and all ECGs from STROKESTOP I were included in the validation set for the model. The area under the ROC curve (AUC) was measured. Sensitivity was set to 75% and specificity was calculated. From a single time-point ECG the artificial intelligence-based algorithm predicted intermittent AF with an AUC of 0.68 (95% CI 0.67-0.70) and specificity of 48% (CI 48-49) in the whole validation set. In STROKESTOP I the algorithm predicted AF with an AUC of 0.64 (CI 0.62-0.66) and specificity of 43% (CI 40-46), and similarly in STROKESTOP II with an AUC of 0.63 (CI 0.60-0.66) and specificity of 40% (CI 35-46). Better results were seen in the SAFER study where the age distribution was wider, AUC 0.83 (CI 0.81-0.86) and specificity 76% (CI 67-82). An artificial intelligence-enabled network has the ability to predict future AF from a sinus rhythm single-lead ECG in an age homogenous group. In a screening program the algorithm may be used as an interim step to identify individuals that might benefit from screening. This would reduce the number of individuals requiring prolonged screening and increase feasibility.