Abstract
Fourier power spectra of heart rate variability (HRV) exhibit 1/f-type scaling. Recently, to evaluate such scaling properties, detrended fluctuation analysis (DFA) has been widely used. In those studies, short-term (<11 beats) and long-term (>11 beats) scaling exponents have been commonly estimated, since those indices were defined in the original paper on DFA. However, the scaling range selection assuming the crossover point located on 11 beats is not based on systematic studies. Actually, spurious scaling exponents due to overfitting are sometimes observed. To address this problem, we here revisited the scaling properties in healthy subjects (n=122) and congestive heart failure patients (n=108). In particular, we address the following problem: (1) Which is the physiologically meaningful time unit, the subject's heartbeat or physical time? (2) Are the scaling ranges of HRV invariant? Although, in previous studies, the subject's heartbeat has been used as a standard time unit, we also studied re-sampled HRV time series. Our analysis revealed that most crossover points of the scaling behavior were located not on 11 beats, but on the very low frequency range (more than 25 seconds).
