The accuracy of beat-interval based algorithms for detecting atrial fibrillation

Automated detection of Atrial Fibrillation (AF) from the surface electrocardiogram (ECG) remains a challenge. Some have suggested that a major source of false positives from R-R interval based AF algorithms are ectopic beats and/or other supraventricular arrhythmias. However, this has not been thoroughly investigated. This study aims to evaluate the accuracy of four commonly implemented R-R Interval based AF algorithms (1) The coefficient of variance, (2) Root Mean Square of the Successive Differences, (3) Turning Point Ratio (TPR) and (4) Shannon Entropy. All four algorithms were tested on R-R interval data from patients in normal sinus rhythm, during atrial fibrillation, with ectopic beats and with supraventricular tachycardia (SVT). Receiver operating characteristic analysis was used to determine the performance of each algorithm over different analysis segment lengths ranging from 30 to 120 beats. When comparing algorithm results, a clear reduction in algorithm performance was found in patients with ectopic beats and SVT. This must be taken into consideration when designing and evaluating algorithms for automated AF detection.