Atrial fibrillation detection evaluation - performance measures

Atrial fibrillation is the most commonly sustained arrhythmia in clinical practice worldwide. It´s essential that AF detection algorithms should be powerful. This paper is addressed to study the proper performance metrics for evaluation.The current ANSI/AAMI standard recommends using two metrics - sensitivity and positive predictive value (PPV). We argue with this combination of indicators of diagnostic performance. We reviewed the possible metrics for evaluation applied in the diverse fields. F-measure, diagnostic odds ratio, accuracy, kappa coefficient, regression and correlation coefficients were closely reviewed. Sensitivity and specificity are best paired performance measures. We propose to utilize sensitivity, specificity, aggregate phi correlation coefficient and aggregate kappa coefficient as performance measures of atrial fibrillation detection algorithms.