Prediction of PAF Attacks using Time-Domain Measures of Heart Rate Variability

Abstract

Paroxysmal atrial fibrillation (PAF) is the mainly encountered type of arrhythmia and there is no validated method to predict a PAF attack before it occurs. In this study, predicting the PAF event was aimed using time-domain heart rate variability (HRV) measures in k- nearest neighbor (k-nn) classifier. Traditional time-domain HRV measures were analyzed in every 5-minute segments from 49 normal subjects, 25 patients with PAF attack and 25 patients with no attack within 45 minutes. All features were investigated whether they showed statistically significance. Significant features were classified by k-nn for odd numbers of neighbors between 1 and 19. This setup was run with two different configurations as study 1 to discriminate patients with PAF attack from normals and patients with no attack, and study 2 to discriminate patients with PAF attack from patients with no attack. SDNN, RMSSD and pNN50 measures were found to show statistically significant differences with p less than 0.05 in segments of 0-5 min, 2.5-7.5 min and 5-10 min intervals only. The maximum classification accuracy was obtained in the time interval of 2.5-7.5 minutes with %79 for Study 1 and just before attack with %80 for Study 2 in the time interval of 0-5 minutes. Results showed that the prediction of PAF events was possible when the classification between normal subjects from PAF patients was accurate. PAF attack can be determined 2.5 minutes earlier by simple classifier algorithms.