Accuracy of algorithms for detection of atrial fibrillation from short duration beat interval recordings

• Published in: Medical engineering & physics Vol. 34; no. 10; pp. 1441 - 1447
• Main Authors: Langley, P., Dewhurst, M., Di Marco, L.Y., Adams, P., Dewhurst, F., Mwita, J.C., Walker, R., Murray, A.
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier Ltd 01.12.2012; Elsevier
• Subjects: Aged; Algorithms; Atrial fibrillation; Atrial Fibrillation - diagnosis; Atrial Fibrillation - physiopathology; Beat interval variability; Biological and medical sciences; Cardiac dysrhythmias; Cardiology. Vascular system; Databases, Factual; Detection; Electrocardiography; Heart; Humans; Medical sciences; Radiology; Receiver operating characteristic; Reproducibility of Results; Residence Characteristics; ROC Curve; Signal Processing, Computer-Assisted; Time Factors; Receiver operating characteristic; Detection; Beat interval variability; Atrial fibrillation; Arrhythmia; Variability; Cardiovascular disease; Receiver; Algorithm; Excitability disorder; Heart disease; Diagnosis; Biomedical engineering
• ISSN: 1350-4533; 1873-4030; 1873-4030
• DOI: 10.1016/j.medengphy.2012.02.002

Atrial fibrillation (AF) is characterised by highly variable beat intervals. The aims of the study were to assess the accuracy of AF detection algorithms from short analysis durations and to validate prospectively the accuracy on a large community-based cohort of elderly subjects. Three algorithms for AF detection were evaluated: coefficient of variation (CV), mean successive difference (Δ) and coefficient of sample entropy (COSEn), using two databases of beat interval recordings: 167 recordings of 300s duration for a range of rhythms acquired in a hospital setting and 2130 recordings of 10s duration acquired in the community. Using the longer recordings receiver operating characteristic (ROC) analysis was used to identify optimal algorithm thresholds and to evaluate analysis durations ranging from 5s to 60s. An ROC area of 93% was obtained at recording duration of 60s but remained above 90% for durations as low as 5s. Prospective analysis on the 2130 recordings gave AF detector sensitivities from 90.5% (CV and Δ) to 95.2% (COSEn), specificities from 89.3% (Δ) to 93.4% (COSEn) and accuracy from 89.3% (Δ) to 93.4% (COSEn), not significantly different to those obtained on the initial database. AF detection algorithms are effective for short analysis durations, offering the prospect of a simple and rapid diagnostic test based on beat intervals alone.