On modelling RR tails in heart rate variability studies: An extreme value analysis

RR distributions with tails larger than the Gaussian have been proved to be an independent predictor of cardiac mortality in chronic heart failure patients. Within this context, extreme value theory provides a powerful tool to quantify the probability of a long RR occurrence, through the statistical characterization of the RR tail distribution. Here, tail characterization does not rely on the Gaussian assumption but by fitting the Generalized Pareto distribution (GPd) to the excesses above a properly chosen high threshold, and through the analysis of its corresponding tail index, denoted as γ. The new approach is illustrated with a 24-h RR recording from a normal subject and a Congestive Heart Failure (CHF) patient. Wavelet analysis allowed to reconstruct one signal containing the RR power traditionally related to respiratory rhythm (~ 0.25 Hz) and another to sympathetic baroreflex activity (~ 0.1 Hz). The fitted distributions for the normal subject do not reject the hypothesis of γ = 0 for both LF and HF while γ > 0 for the CHF patient. Thus, the CHF distributions are heavy-tailed, indicating a non-negligible probability that a very long RR interval can occur. In a forthcoming study, it will be assessed the impact of these preliminary findings in CHF mortality prediction.