Estimation of frequency shift in cardiovascular variability signals.

Spectral analysis of heart rate (HR) and blood pressure (BP) oscillations has traditionally concentrated on spectral power, although a shift in spectral frequency characterises the variability better than power in some cases. Experimental data were obtained from 14 healthy males in control and pharmacological blockade conditions. When parasympathetic control was reduced, LF oscillations of HR and BP tend to shift towards lower frequencies. Three parameters were compared to estimate the spectral shift within the low frequency (LF, 0.04-0.15 Hz) band in HR and BP variability: mean (fmean), median (fmed), and central frequency (fc). Parameter variance (pSTD) and sensitivity to noise were also estimated using realistic HR, systolic BP (SBP) and diastolic BP (DBP) data. fmean showed the lowest parameter variance both for an autoregressive (AR) method (SBP pSTD 3.1 vs 4.8 vs 4.7 mHz for fmean, fmed and fc, respectively; p<0.001) and an FFT method (SBP pSTD 4.7 vs 7.7 mHz for fmean and fmed, respectively; p<0.001). Furthermore, fmean was least sensitive to noise. fc showed the poorest performance being especially sensitive to noise. To analyse the spectral shift, fmean is preferred, since it performs better than fc, which has been used in most previous studies. To quantify the frequency of oscillations in cardiovascular signals, the mean frequency is recommended, with analyses across different spectral bands.