Time Domain Analysis
Time domain analysis is based on the statistical interpretation of R-R time interval values. Time domain measures of HRV (
Table 17-1) are closely related to the total variance of the heart signal.
^{3},
^{8} The most common index of overall HRV is the standard deviation of all R-R intervals (SDNN), typically involving 80,000 to 150,000 heart period values in a 24-hour recording. Long-term variability, such as that reflecting normal circadian influence over a 24-hour period, is best reflected by two measures based on partitioning the full recording into sequential 5-minute segments. Each segment typically contains 300 to 500 R-R intervals, and there would be 288 such segments in a 24-hour recording. The SDANN is defined as the standard deviation of the means of the R-R intervals in each 5-minute segment, whereas the complementary SDNN index is the mean of the standard deviations of the R-R intervals in each 5-minute segment.
Short-term time domain measures of HRV are derived from the differences of successive normal R-R intervals. They are highly correlated and are considered to provide good estimates of PSNS activity.
^{3} Short-term measures include the square root of the mean squared difference of successive normal R-R intervals (rmsSD) and the percentage of successive normal R-R intervals that change by more than 50 milliseconds compared with the total number of R-R intervals (pNN50).
Frequency Domain Analysis
Frequency domain analysis, or spectral analysis, is an elegant method for studying the rhythmic components in an R-R interval sequence and presents intriguing possibilities for disentangling PSNS and SNS influences on the heart.
^{9} A plot of the power spectral density of HRV versus frequency describes how the variances of the frequency components of the heart signal are distributed.
^{3}
Both parametric and nonparametric methods common to time series analysis have been used to estimate the power spectral density. The most common methods are the discrete Fourier transform (DFT) (nonparametric) and autoregressive (AR) (parametric) time series models. The AR model-based spectrum is usually less computationally efficient than the DFT, but it can be applied to data sequences of arbitrary length, including very short segments. The AR approach tends to produce a spectrum that is statistically more stable than that produced by the DFT but requires assumptions about the time series model.
^{3}
The total area under the curve of the power spectral density versus frequency plot is equal to the total statistical variance, or the power of the signal. These power (variance) distributions are calculated for defined frequency bands and are interpreted as an estimate of the variance of the HRV signal within that band (
Table 17-2). There are two major spectral components seen in HRV data: the high-frequency (HF) (0.15-0.40 Hz) component and the low-frequency (LF) (0.04-0.15 Hz) component (
Fig. 17-2). The HF component is associated with respiration
^{8},
^{10} and is considered to reflect the relative input of the PSNS. The basis of the LF component is more controversial and may be the result of both SNS and PSNS activity input.
^{11} The LF to HF ratio (LF:HF) has been regarded as reflecting the balance between the mixed PSNS and SNS activity input to the PSNS activity input.
^{3} The spectral HF and the LF:HF are often reported together in nursing research studies seeking to explore the joint contribution of the SNS and the PSNS branches to HRV phenomena. Studies of very low-frequency and ultra-low-frequency ranges have also been conducted but require long uninterrupted sampling periods and specialized methods of analysis. In addition, the clinical interpretation of findings in these frequency ranges remains controversial.
^{3}
In common with other variance-like measures, the within-subject HRV band power estimates are often reexpressed using the natural logarithm transform to reduce distributional skewness before use in statistical procedures. HRV quantitative band power summary indices (LF, HF, etc.) computed using the AR or DFT methods should be virtually identical.
^{12}
Several derived measures can easily be computed from these spectral band summaries (see
Table 17-2). Normalized variants of the LF and HF indices are often defined by dividing the power in each band by the total power, with the result expressed as a percentage.
^{13}
It should be pointed out that the HRV spectrum and spectrum-based band power (variance) summary statistics, like all HRV measures, are defined over blocks of R-R intervals; thus, their meaning is not localized to a particular instant in time or to a particular beat. Typical block window lengths in clinical and research applications range from 2 minutes to 24 hours. Spectra derived from shorter blocks are more localized in time and are more likely to be internally stationary but may have less frequency resolution, especially with respect to slower rhythm patterns. HRV spectra based on very long individual blocks (e.g., 24 hours) will have the ability to resolve very slow rhythmic patterns but will almost certainly span nonstationary data segments and heterogeneous latent autonomic states.