Power spectral analysis of heart rate variability by autoregressive modelling and fast Fourier transform: a comparative study.

OBJECTIVE: To compare the results from autoregressive modelling (ARM) and from fast Fourier transform (FFT), the most commonly used methods for the analysis of short-term heart rate variability in the frequency domain. METHODS &
RESULTS: RR interval and respiratory activity were recorded in the supine and standing positions under standardized laboratory conditions in a population-based sample of 614 subjects. The low-(LF) and high-frequency (HF) components of heart rate variability were identified by power spectral analysis, by use of FFT, with application of two sets of frequency ranges, and by ARM; LF and HF power were expressed in both normalized (%) and absolute units (ms2). The RR interval, its variance and the HF power decreased from the supine to the standing position (P < 0.001). The LF power increased on standing when expressed in normalized units, but decreased in absolute units, whereas the LF-to-HF ratio increased (P < 0.001). On the low side of the spectrum, FFT slightly overestimated the LF component obtained with ARM, when the predefined frequency range was 0.05-0.15 Hz (P < 0.001); the underestimation of LF in the frequency range 0.07-0.14 Hz was more pronounced, particularly in the erect position (P < 0.001). Both FFT methods overestimated (P < 0.001) the ARM HF component, more so for the 0.15-0.50 Hz range than for the 0.14-0.35 Hz range. Finally, we observed considerable within-subject differences between methods, which were estimated by calculation of the limits of agreement.
CONCLUSIONS: Different methods for spectral decomposition of short-term heart rate variability yield similar qualitative results, but the quantitative results differ between ARM and FFT, and within the FFT method according to the selected frequency range.