ESOC 2026 | Prediction of AF using deep learning from a single-lead sinus rhythm electrocardiogram

So the idea of the study comes from the concept that we can predict AF from a sinus ECG. This has been demonstrated by Atia and colleagues in the Mayo Clinic, where they demonstrate very good AUC to predict AFib in the following years from a 12-lead ECG. And with our team at Cardiologs, we developed another proof of concept that can predict AF in a short time horizon from a single-lead ECG. So in this study, we wanted to link this AF risk to stroke risk. So we had a retrospective cohort of 11,000 patients from the Health Data Hub, the French National Registry. And so we had each of these patients had a 24-hour Holter ICG in sinus rhythm, no AF. And so we can make the AI model prediction if the patient was at high or low risk of AF. And then we add the outcome of stroke or systemic embolism at five years. And what we showed is that the high-risk population, high-risk of AF population has a 1.7-fold higher risk of stroke at one year and 2.1-fold higher risk of stroke at four years. So with this study, this opened the door to the concept of prophylactic treatment, anticoagulation, of patients at high risk of AF but who are in sinus rhythm to avoid a stroke as primary or secondary prevention. And for that, we will need a further randomized study.

This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.