Point process modeling of R-R interval dynamics during atrial fibrillation

Atrial fibrillation (AF) is the most common arrhythmia, and one of the main causes of ictus and strokes. Effective treatments for AF are still unknown, as its effects on the heart substrate have not been clearly quantified yet. One of the main lines of investigation aims at characterizing ventricular response by looking at its effects on heartbeat interval dynamics. Most of the standard approaches have focused on RR interval (RRI) histogram parameters albeit with several shortcomings, such as bin width dependence or lack of attention to the time-varying dynamical structure. In this study, we model heartbeat interval series as a history-dependent inverse Gaussian (HDIG) point process where the history for each RRI prediction is a linear regression of the previous RRIs. As opposed to classical nonparametric methods, the heart rate (HR) variability features derived from the proposed parametric model provide a physiologically more consistent characterization during AF, and can clearly discriminate AF from sinus rhythm (SR) subjects. Analysis of 36 patients affected by persistent AF and 18 controls shows that RRI distributions are more right-skewed and affected by higher variability during AF (skewness of 0.63±0.29 in AF and of 0.17±0.16 in SR, p=7.6 · 10^-8; HR standard deviation of 18.73±10.64 bpm in AF and 4.66±4.75 bpm in SR, p=2.3 · 10^-6). Our results demonstrate that we can extract valuable information associated with AF from RRI series by using a point process framework.