|
Heart
Rate Variability
Kenneth C. Bilchick, M.D.;
Ronald D. Berger, M.D., Ph.D.
J Cardiovasc Electrophysiol.
2006;17(6):691-694. ?2006 Blackwell Publishing
Posted
IntroductionDuring normal sinus rhythm, the heart rate (HR) varies from beat to beat.Heart rate variability (HRV) results from the dynamic interplay between the multiple physiologic mechanisms that regulate the instantaneous HR.Since short-term HR regulation is predominantly governed by sympathetic and parasympathetic neural activity, examination of HR fluctuations provides a window to observe the state and integrity of the autonomic nervous system. This review focuses on methodology and interpretation of HRV measures. Time Domain Measures of HRVMeasurement of HRV first requires detection of each heart beat. For the purpose of assessing autonomic regulatory effects on HR, it makes most physiologic sense to detect the occurrence of sinus nodal events or P waves. In practice, this is technically difficult without intracardiac atrial electrograms, so HRV measurement is usually based on the sequence of RR intervals. This practice neglects the potential presence of fluctuations in PR interval due to modulation of AV nodal conduction.Premature ventricular contractions (PVCs) and premature atrial contractions (PACs) represent additional confounders in assessing autonomic regulation of HR. Removal of the effects of PVCs and PACs is discussed below. A straightforward and useful metric of HRV,termed the SDNN, is the standard deviation of all normal RR intervals(those measured between consecutive sinus beats). The SDNN may be easily calculated from a 24-hour Holter monitor. In calculating SDNN,any RR interval that begins or ends with a PAC or PVC is simply deleted from the sequence. SDNN is typically measured over 24 hours and reported in units of ms. Results derived from shorter (or longer)periods should not be compared to values for the accepted normal range,which are based on 24-hour records. This is because HRV is not a stationary process, i.e., a process in which the mean and variance are independent of record length. For example, for 24-hour recordings, SDNN cutoff values separating better and worse outcomes are usually 70-100 msec, whereas for 5-minute recordings, the corresponding cutoff value is on the order of 30 msec. Two variants of the SDNN, created by dividing the 24-hour monitoring period into 5-minute segments, are the SDNN index and the SDANN index(both with units in ms). The SDNN index is the mean of all the 5-minute standard deviations of NN (normal RR) intervals during the 24-hour period (i.e., the mean of 288 NN standard deviations), while the SDANN index is the standard deviation of all the 5-minute NN interval means(i.e., the standard deviation of 288 NN means). The HRV indices discussed thus far are called time domain measures because they are based on the time-series of normal RR intervals. Other noteworthy time domain indices are the r-MSSD and the pNN50. The r-MSSD(units in ms), or the root-mean-square successive difference,calculates the square root of the mean of the squared differences between successive NN intervals over 24 hours. The pNN50 (percentage units) calculates the percentage of differences between successive NN intervals over 24 hours that are greater than 50 ms. Both of these indices measure short-term variation in the NN interval because they are entirely based on comparisons between successive beats. Of note, all the HRV indices described above, except pNN50, have units of time (ms) and thus, strictly speaking, are measures of variability in RR interval, not HR. HR and RR interval are reciprocals of each other,or to be exact, HR = 60,000/RR, where HR has units of beats per minute(bpm), and RR has units of ms. Fluctuations in RR interval and HR are closely related, but not in a linear way, since the reciprocal is not a linear operation. Thus, a doubling in RR interval variability does not mean HRV would necessarily double if measured from the sequence of corresponding instantaneous HR values. Time domain measurements are traditionally calculated from the RR (or NN) interval sequence, even though instantaneous HR may be more closely tied to autonomic tone and,therefore, have greater physiologic significance than RR interval. Frequency Domain Measures of HRVAdditional insight into the nature of HR fluctuations may be gained by analyzing the fluctuations in the frequency domain. By analogy, one learns only about the overall power of a star by measuring the intensity of light emanating from it, but by separating the light into its component colors with a prism, one may learn about the composition of chemical reactions within the star. HRV may similarly be broken into the frequency components that compose the overall variability. Frequency domain analysis is performed by taking a series of numbers along the time axis and computing the Fourier transform. A computationally efficient algorithm for calculating the Fourier transform, called the fast Fourier transform (FFT), can be carried out when the number of time samples is exactly a power of 2 (e.g., 256, 512, 1,024, etc.) If the analysis is performed using N time samples denoted x(n), then the FFT yields N/2 complex numbers, denoted X(f), spaced evenly along the frequency axis, from 0 to half the sampling rate. Thus, if 1,024 RR intervals are used, the FFT will provide 512 complex values in the frequency domain spaced evenly from 0 cycles/beat to the alternans frequency (one cycle every other beat). The squared magnitude of each of the complex values from the FFT is then computed (real part squared plus imaginary part squared), as this yields the power spectral density(PSD) function, denoted P(f). Thus, P(f) = |X(f)|2. PVCs and PACs have a particularly insidious effect on the PSD function. If they are not edited out, the effect of a sudden brief change in RR interval (or HR) is to add substantial power density at all frequencies, often overwhelming the true variability that is being assessed. On the other hand, if the RR intervals corresponding to the ectopic beats are simply deleted, then individual frequency components become altered because the deletion of time advances the phase, but by a different amount at each frequency. This has the effect of reducing the calculated power density at some frequency components and augmenting it at others. A simple solution to this problem is to replace the RR intervals affected by an ectopic beat with the same number of RR intervals of value equal to the mean of those replaced. If the PSD is calculated from a series of RR intervals or instantaneous HR using one value for each beat, then the frequency axis of the PSD will have units of cycles per beat. The frequency axis values can be divided by the average RR interval duration (expressed in seconds) to convert the units to Hz (cycles per second). Alternatively, the RR interval series may be resampled to provide instantaneous HR values at a 4 Hz sampling rate (Fig. 1), which are then used to calculate the PSD function.[1]In the latter case, the frequency axis will already have units of Hz.There is an additional advantage in resampling the RR interval sequence into evenly spaced time samples. The raw RR interval series contains exactly one sample per beat, but each beat occurs after an uneven amount of time from the prior beat. The time sampling associated with the raw RR interval series is thus uneven. This has an effect of warping the frequency axis when expressed in Hz, even after conversion from cycles/beat to Hz. 
Figure 1. The PSD is usually integrated within specific frequency bands, since fluctuations within each band are mediated by specific physiologic mechanisms. In seminal work in the early 1980s, Akselrod et al.[2]showed that the low frequency (LF) band (0.04-0.15 Hz) is related to both sympathetic and parasympathetic modulation, and the high frequency(HF) band (0.15-0.40 Hz) is governed almost exclusively by parasympathetic effects. The ratio of LF to HF power is often used as a metric of sympathetic?parasympathetic balance. It is important to note,however, that the main driver of HRV in the HF band is respiration,which produces the vagally mediated respiratory sinus arrhythmia. The magnitude of HF power is highly dependent on the depth of respiration,which often varies greatly from one recording epoch to another. Unless the depth of respiration is taken into account, assessment of autonomic balance by HRV measurement alone may be quite difficult. Relationships Between Time and Frequency Domain IndicesAn important relationship between the time domain
and the frequency domain is provided by Parseval's theorem. This states
that the sum of squares measured among the time samples is equal to the
sum of squares of the Fourier transform results, when frequency samples
are included from 0 to the sampling frequency, denoted fs.
Thus,  Since P(f) = |X(f)|2 and the power spectrum is symmetrical about the point at half the sampling frequency, we can express Parseval's theorem as:  The left side of this equation is the variance, or standard deviation squared, of the process (after the mean has been subtracted out), and the right side is twice the total area under the PSD curve. Thus, for a sequence of normal RR intervals, the square of SDNN must equal exactly twice the total power found by integrating the PSD function, derived from the same sequence of RR intervals. Bigger et al.[3]carefully studied the extent to which specific time and frequency domain HRV indices correlate. Not surprisingly, SDNN was highly correlated with ln(total power) (r = 0.96). r-MSSD and pNN50, both measures of rapid change from one beat to the next, were highly correlated with ln(HF power) (r = 0.92 and r = 0.89, respectively),while SDANN index, a measure of gradual change, correlated with ln(ultra-LF power) (r = 0.96). Non Linear Indices of HRVAn important finding in 1982 by Kobayashi and Musha led to a new paradigm of quantifying HRV. These investigators reported the results of spectral HRV analysis from data collected over 10 hours in a healthy male and thus were able to study frequencies as low as 0.0001 Hz. The power density versus frequency spectrum plotted on log-log axes,surprisingly, was found to be nearly a straight line, leading to the establishment of the "power law" that P = Cfß with ß = −1. In other words, the power density of HR fluctuations falls with frequency as 1/f. It has been hypothesized that failure of an individual's HRV spectrum to conform to this 1/f power law is pathologic. This hypothesis led to investigations into how to best quantify the power-law behavior of an HRV spectrum, which produced the following two nonlinear indices used today: detrended fluctuation analysis (DFA) slope for short segments,denoted as αs, and approximate entropy of HR fluctuations,denoted as ApEn. There have been a number of studies of these nonlinear indices, and they have generally yielded inconsistent results. The use of the nonlinear indices is worthy of further investigation. Prognostic Value of HRV in Myocardial Infarction and Heart FailureAfter some early reports suggesting that low HRV after myocardial infarction might indicate a worse prognosis, the Multicenter Post infarction Group published in 1987 the results of the first large-scale study on the subject. Over an average follow-up period of 2.5 years, they showed that patients with SDNN < 50 ms had a 5.3 times higher (34%)mortality than those with SDNN > 100 ms (9%).[4]SDNN was the strongest univariate predictor of mortality and remained the most powerful predictor of mortality even after adjustment for clinical, demographic, other Holter features, and ejection fraction.The authors hypothesized that this was due to increased sympathetic tone and vagal withdrawal, which increased the risk for ventricular fibrillation. In the thrombolytic era, the GISSI-2 group in 1996 studied patients who had received thrombolytics after myocardial infarction and found that patients with SDNN < 70 msec had an adjusted 3-fold increased mortality.[5]The large prospective autonomic tone and reflexes after myocardial infarction (ATRAMI) study with 1,284 post myocardial infarction patients(only 20% treated with beta blockers) followed over an average 21 months further confirmed these results. The investigators showed that SDNN < 70 ms carried a 3.2-fold increased risk of mortality, which was additive when combined with low baroreflex sensitivity and low ejection fraction.[6] The majority of studies in chronic heart failure of both ischemic and nonischemic etiology have shown that low SDNN predicts mortality.Several studies of HRV in heart failure from the late 1990s showed that low SDNN is associated with increased mortality, although only one study demonstrated an increase in sudden death with lower SDNN.[7] HRV was used as a risk stratifying entry criterion in the recent DINAMIT study, which randomized patients between implantable cardioverter defibrillator (ICD) and standard care 6-40 days post myocardial infarction and failed to show a benefit for the ICD. It is unclear whether the negative results of this study are due to lack of benefit of the ICD so soon after infarction, or to an inability of HRV to identify patients who can benefit from ICD implantation. Effect of Cardiovascular Drugs on HRVMost cardiovascular drugs that improve morbidity and mortality, including beta blockers, ACE-inhibitors, and statins, also increase HRV.Metoprolol, quinapril, captopril, enalapril, and atorvastatin have been shown in separate studies to increase HRV. The mechanism by which beta blockers increase HRV is likely complex, since the sympathetic nervous system activation has been shown to increase low-frequency HRV, as discussed earlier.[2]The clinically observed increase in HRV with beta blockers is likely related to the concomitant beneficial effects on the parasympathetic nervous system and renin-angiotensin-aldosterone axis. Beta blockers may diminish the predictive value of HRV after myocardial infarction,at least for sudden death.[8] ConclusionsIn summary, HRV is a well-studied metric in cardiovascular disease. It has been shown convincingly to reflect changes in neurohormonal activation,an important pathophysiologic factor in a number of cardiovascular disease states. Findings from studies in the 1980s and 1990s indicated that SDNN was independently predictive of mortality after acute myocardial infarction and in heart failure. In the current era with widespread use of medications resulting in a more comprehensive blockade of the beta adrenergic and renin-angiotensin-aldosterone axes,the value of traditional HRV measures for predicting sudden death is less clear. There are currently no data showing that HRV is useful in risk stratifying patients for ICDs. The DINAMIT study suggests that HRV cannot be used to reliably predict which patients will benefit from an ICD implant within 40 days after a myocardial infarction. Now that some ICD models record daily measures of HRV, we will have the opportunity to determine if HRV can predict appropriate ICD shocks in patients with conventional implant indications. In interpreting the recent studies of HRV and sudden death, it is important to note that currently no single test or, for that matter combination of tests, has high predictive value for sudden cardiac death. For example, in the recent series of Huikuri, the positive predictive accuracy of the three best metrics, ejection fraction,presence of non sustained ventricular tachycardia, and abnormal signal-averaged ECG, were only 8%, 12%, and 13%, respectively.[8] Other modalities, such as genomics and cardiac MRI, are under investigation for this purpose. CLICK HERE for subscription information about this journal. References
Reprint
Address
Ronald D. Berger, M.D., Ph.D., Department of Medicine, Johns Hopkins University School of Medicine, 600 North Wolfe Street, Carnegie 592, Baltimore, MD 21287. Fax: 410-502-4854; E-mail: rberger@jhmi.edu .gif){kind=spacer} Kenneth C. Bilchick, M.D.,
Ronald D. Berger, M.D., Ph.D., Johns Hopkins Medical Institutions,
Baltimore, Maryland, USA |
