Abstract
Autoregressive (AR) models have been used in heart rate variability (HRV) analysis. In the conventional AR model, a Gaussian noise with a homogeneous variance are assumed to be the driving stochastic force. If this assumption is reasonable for the observed HRV time series, the HRV dynamics is fully characterized by its power spectrum. However, reasonability of the assumption has not been systematically examined. To address this problem, we studied variance heterogeneity, called heteroscedasticity, of HRV using a generalized autoregressive conditional heteroscedasticity (GARCH) model. In this study, we analyzed 24 hour HRV time series measured in healthy subjects (n=171) and in congestive heart failure (CHF) patients (n=108), and estimate the parameters of the GARCH model from the observed time series. In both healthy and CHF groups, variance heterogeneity of the HRV time series was observed. Furthermore, we found a statistically significant difference on the estimated parameters of the GARCH model between healthy and CHF groups.
