[Predicting atrial fibrillation through a sinus-rhythm electrocardiogram; useful or not?]

Area of Science:

• Cardiology
• Neurology
• Artificial Intelligence in Medicine

Background:

• Atrial fibrillation (AF) detection is critical in cryptogenic stroke patients to guide oral anticoagulation therapy, preventing recurrent ischemic events.
• Permanent AF is detectable via electrocardiogram (ECG), but paroxysmal AF often requires prolonged monitoring.
• Identifying AF in stroke patients is essential for appropriate treatment selection over anti-platelet therapy.

Purpose of the Study:

• To describe an artificial intelligence (AI) and big data-driven algorithm designed to detect a 'footprint' of atrial fibrillation (AF) on a 12-lead ECG during sinus rhythm.
• To evaluate the utility of this AI algorithm for identifying AF in patients with cryptogenic stroke.

Main Methods:

• Development of an algorithm using artificial intelligence and big data analytics.
• The algorithm analyzes 12-lead ECGs recorded during sinus rhythm to identify subtle indicators of underlying atrial fibrillation.
• Evaluation of the algorithm's performance in a post-stroke population with a low prevalence of undetected AF.

Main Results:

• The AI algorithm aims to detect a 'footprint' of AF on ECGs during sinus rhythm, offering a potential non-invasive detection method.
• The algorithm's 'black-box' nature prevents ruling out bias from patient characteristics or medications.
• Moderate test specificity and a low prevalence (10%) of undetected AF in the study population resulted in a low positive predictive value, limiting its clinical utility for initiating anticoagulation.

Conclusions:

• While the AI algorithm shows a promising premise for detecting AF 'footprints' on ECGs, its current performance is insufficient for clinical decision-making in cryptogenic stroke patients.
• The low positive predictive value, stemming from moderate specificity and low AF prevalence, renders the algorithm not useful for guiding anticoagulation therapy initiation.
• Further refinement and validation are necessary to overcome limitations related to potential bias and improve predictive accuracy for clinical application.