Safe automatic one-lead electrocardiogram analysis in screening for atrial fibrillation.

Aims: Screening for atrial fibrillation (AF) using intermittent electrocardiogram (ECG) recordings can identify individuals at risk of AF-related morbidity in particular stroke. We aimed to validate the performance of an AF screening algorithm compared with manual ECG analysis by specially trained nurses and physicians (gold standard) in 30 s intermittent one-lead ECG recordings.

Methods and results: The STROKESTOP study is a mass-screening study for AF using intermittent ECG recordings. All individuals in the study without known AF registered a 30-s ECG recording in Lead I two times daily for 2 weeks, and all ECGs were manually interpreted. A computerized algorithm was used to analyse 80 149 ECG recordings in 3209 individuals. The computerized algorithm annotated 87.1% (n = 69 789) of the recordings as sinus rhythm/minor rhythm disturbances. The manual interpretation (gold standard) was that 69 758 ECGs were normal, making the negative predictive value of the algorithm 99.9%. The number of ECGs requiring manual interpretation in order to find one pathological ECG was reduced from 288 to 35. Atrial fibrillation was diagnosed in 84 patients by manual interpretation, in all of whom the algorithm indicated pathology. On an ECG level, 278 ECGs were manually interpreted as AF, and of these the algorithm annotated 272 ECGs as pathological (sensitivity 97.8%).

Conclusion: Automatic ECG screening using a computerized algorithm safely identifies normal ECGs in Lead I and reduces the need for manual evaluation of individual ECGs with >85% with 100% sensitivity on an individual basis.