Chapter 16. Graphical Analysis of Heart Rate Patterns to Assess Cardiac Autonomic Function

A continuous electrocardiogram, whether from a Holter recording, an intensive care unit monitor, an overnight polysomnogram, or even a short-term recording, provides a signal that can yield information about the morphology and time of onset of each heartbeat. This information, exported as a "beat file" provides the basis for multiple ways of quantifying and categorizing heart rate variability (HRV), in most cases based on intervals between normal to normal (N-N) heartbeats only. The various methods for quantifying HRV (eg, time domain, frequency domain, nonlinear) and their relationship to cardiac autonomic function have been described in multiple excellent reviews elsewhere.1-3 Less appreciated is the power of using graphical images, also derived from beat files, to obtain information about normal and abnormal cardiac autonomic function, sinus node function, and sleep-disordered breathing.

The periods between successive heartbeats can be converted to a time series of instantaneous heart rates (60,000 milliseconds in a minute/time between beats in ms). Heart rate (HR) patterns can be examined on multiple scales, each providing both unique and overlapping information. HR itself can be plotted on a beat-by-beat basis, or HR averages and ranges over longer periods can be plotted. Power spectral analysis can mathematically deconstruct heartbeat patterns into their underlying rhythmic components using fast Fourier transforms (FFTs).3 The structure of heartbeat patterns can be examined using Poincaré plots, which are plots of the interval between every pair of successive beats versus the next pair.

The current review will illustrate the types of information potentially available from these aspects of graphical HRV analysis (ie, 5-minute averaged HR patterns, hourly power spectral analysis, hourly Poincaré plots, and beat-by-beat HR tachograms). To accomplish this, we will primarily use representative plots from recordings selected from our database of subjects with and without known cardiovascular disease. From among the healthy subjects, we will examine graphical HRV in a younger adult with high HRV (the standard deviation of all normal-to-normal interbeat intervals [SDNN] = 198), an older adult with high HRV (SDNN = 167), and an older adult with low HRV (SDNN = 65). From among those with known cardiovascular disease, we will examine graphical HRV in a subject with very low HRV but normal HR patterns (SDNN = 63), a subject with periods of abnormal HR patterns (SDNN = 99), a subject with significant sleep-disordered breathing HR patterns (SDNN = 49), and a subject with atrial fibrillation (SDNN = 211).

Commercial Holter scanner reports often show plots of 5-minute averaged HR patterns. The presence or relative absence of a circadian rhythm of HR is clearly visible on these plots. Under normal circumstances, a clear decrease in HR during the night, a distinct rise in HR on awakening, and a relatively higher HR during the daytime are seen. Lack of circadian rhythm of HR is the primary determinant of low values for total HRV and suggests severe autonomic dysfunction and/or a complete lack of physical activity. In Fig. 16–1, the average HRs and their ranges for every 5 minutes are shown for four healthy subjects. The circadian rhythm of HR can be appreciated qualitatively by following the center line of the plots. A qualitatively normal circadian rhythm is clearly seen for the younger subject (see Fig. 16–1A) and the older subjects with high HRV (see Fig. 16–1B) and average HRV (see Fig. 16–1C), and approximate bed and wake times can be seen from the decrease in HR in the evening and the sharp increase in the morning, but in the older adult with low HRV (see Fig. 16–1D), circadian rhythm is markedly attenuated and bed time is less clear.

Figure 16–2 shows 5-minute averaged HR plots for subjects with known cardiovascular disease. The patients in Fig. 16–2A,B have a diminished circadian rhythm, but a circadian rhythm is clearly visible in Fig. 16–2C. No circadian rhythm for HR is visible for the subject with atrial fibrillation (see Fig. 16–2D).

In Figs. 16–1 and 16–2, the top line of each plot shows the highest instantaneous HR in each 5-minute period, and the bottom line shows the lowest. The range of HRs within each period is a rough surrogate for the amount of HRV within each period. When the subjects with high HRV (see Fig. 16–1A,B) are compared with the subjects with lower HRV (see Fig. 16–1C,D), HR ranges are clearly less in subjects with lower HRV. HR ranges are still lower in the subject shown in Fig. 16–2A. Thus, examination of the range of HRs within each 5-minute period provides a qualitative sense of the amount of HRV present.

However, these HR ranges must be interpreted with caution, and examination of the HR ranges and circadian patterns of these ranges, as shown in these plots, could arouse suspicion of additional underlying problems beyond the lack of normal circadian rhythm. The clearest example is Fig. 16–2D (atrial fibrillation). Note the complete lack of a circadian rhythm and the extremely wide HR range. Abnormalities in cardiac rhythm may also be suspected in Fig. 16–2B,C, where there are periods of low HR range interspersed with periods of much higher HR ranges.

Another way of displaying HR patterns is by frequency domain or power spectral analysis. This involves using FFTs or autoregressive methods to decompose the variations in HR into underlying frequencies that together constitute the whole HR pattern, roughly analogous to decomposing a single note in a symphony into the underlying sounds of the different instruments that are playing at once.1,3 Normally, this process is performed on short segments of the signal (often 5 minutes) and then averaged over time. Only N-N interbeat intervals are used. If <80% of the time of segment is accounted from by N-N interbeat intervals (ie, >20% noise or ectopic beats), that segment is not usable for power spectral analysis. The reason for this is that the beat before and the beat after a nonnormal beat is excluded in power spectral analysis; thus, having >20% noise in a 5-minute segment means excluding >60% of the beats in that segment a priori. In a display of the results of power spectral analysis, the amount of variance (or amplitude, which is the square root of variance) at each underlying frequency is plotted on the y-axis, and the underlying frequency is plotted on the x-axis. Thus, the greater the HRV is, the greater the variance in HR and, therefore, the larger the area under the power spectral curve.

The various frequency domain HRV measures found in the literature are measures of the area under the FFT curve (or power) in groups of predefined underlying frequencies or "bands." These are usually reported as ultra low–frequency (ULF) power, very low–frequency (VLF) power, low-frequency (LF) power, and high-frequency (HF) power.3 ULF power reflects variance in HR with a period of between 5 minutes and 24 hours and primarily reflects circadian HR patterns and long-term activity.3 VLF power (variations from 20-second to 5-minute cycles) is believed to reflect the activity of the renin-angiotensin and parasympathetic systems, although it is exaggerated by periodic breathing patterns.4 LF power (variations from 3-9 cycles/min) reflects the combined activity of the sympathetic and parasympathetic nervous systems and especially the activity of the baroreflex.3 Beat-to-beat HR changes are quantified by HF power (variations at respiratory frequencies) and primarily reflect parasympathetic modulation of HR.3 Differences in the amount of power in the different FFT bands are often used to compare cardiac autonomic function between subjects or within subjects after interventions, but as will be emphasized throughout this chapter, without graphical analysis of HR patterns, these numbers may prove to be misleading.5

Many commercial Holter scanning programs provide a single summarized plot of HRV power averaged over the entire recording. This plot is usually displayed as either amplitude versus frequency or ln amplitude versus ln frequency. The linear scale is more easily interpreted. Although 24-hour HRV plots are useful, more detailed information about changing autonomic function during the recording period can be obtained by plotting the HRV power spectral plot on an hourly basis. This, too, is available in some commercial Holter scanning systems. Notably, however, when hourly plots are used, information about the amount of ULF power is lost. Some of the information that can be gleaned by examining hourly FFTs is described in the following sections.

Bed and Wake Times

Just as the circadian rhythm plots can suggest bed and wake times, examination of hourly power spectral plots can also identify these times in most people and sometimes even distinguish between quiet rest and sleep. The hour during which the subject became supine can often be identified as the hour when a distinct peak appears in the HF band. This peak appears because when people lie down, sympathetic activity declines and HR is under predominantly parasympathetic control, which leads to a large increase in respiratory sinus arrhythmia, which is vagally mediated.6 Respiratory sinus arrhythmia, which is a speeding up of HR on inhalation and slowing down on exhalation, obviously occurs at the same frequency as respiration (ie, usually between 9 and 24 cycles/min). This frequency corresponds to the underlying oscillations of the HF power spectral HRV band (0.15-0.4 Hz). Moreover, the center frequency of this peak corresponds to the average breathing frequency during that period, and the magnitude of this peak reflects the degree of parasympathetic control of HR during that time. Often, supine rest can be distinguished from sleep because the center frequency of this peak declines at sleep onset compared with quiet rest and remains at that value during the night. Similarly, upon arising, this peak generally disappears. The reappearance of the HF peak during the middle of the day suggests a nap.

The 24 hourly power spectral amplitude plots for the healthy subjects originally shown in Fig. 16–1 are seen in Fig. 16–3. As indicated on the figure, hours with clearly visible HR peaks corresponding to times in bed and hours without such peaks corresponding to times out of bed can clearly be identified for all but the subject with low HRV (see Fig. 16–3D).

Figure 16–4 shows 24 hourly spectral amplitude plots for the individuals with known cardiovascular disease originally shown in Fig. 16–2. Although the HF peak is extremely small for the subject with low HRV originally seen in Fig. 16–2A, close inspection can identify times in bed. In Fig. 16–4B, there are hours during the night with clear HF peaks that correspond to time in bed, but the HRV pattern is unclear in the hours immediately preceding and following them. The subject in Fig. 16–2C has clearly defined HF peaks during normal sleep hours, starting at 00:00, but additional HF peaks with a higher center frequency at 18:00 to 23:00 suggest that this subject was lying down for much of the time. Not surprisingly, no HF peak can be identified for the subject with atrial fibrillation.

Abnormalities in the HF Portion of the Power Spectral Plot

Under normal circumstances, the HF peak of the power spectral plot, especially during sleep, is visible and distinct, as shown in most of the plots. As seen in Fig. 16–2A, even in patients with extremely low HRV, a clear, if small, HF peak can often be identified for the nighttime period, suggesting diminished but persevered vagal control of the heart. The absence of any such peak (as in Fig. 16–1D) suggests extreme parasympathetic dysfunction. Abnormalities in the shape of the HF peak suggest an underlying problem. For example, a disorganized peak in the HF band that has no clear center frequency (as in Fig. 16–2B) might suggest extremely irregular respiration or suggest the presence of either erratic sinus rhythm (ie, a high degree of sinus arrhythmia of nonrespiratory origin) or significant scanning error due to uneven detection of beat onsets.5,7

Sleep-Disordered Breathing HR Patterns in the Power Spectral Plot

The association of sleep-disordered breathing with excessive sympathetic activation and the subsequent improvement in cardiac autonomic function with treatment have been well documented.8 Clearly, sleep-disordered breathing is associated with multiple adverse outcomes, and just as clearly, it is very much underdiagnosed.8-10 Graphical analyses of HRV offer a simple way to screen for this underlying condition. When sleep-disordered breathing is present, the patient experiences repeated episodes of loss of airflow and then abrupt awakenings to resume breathing. Unless autonomic dysfunction is extremely severe, each of these awakenings is accompanied by a sharp increase in HR, sometimes by as much as 25 to 30 beats per minute. Similarly, although generally briefer and more frequent, HR arousals can be caused by periodic limb movements (usually around every 20 seconds). Because sleep-disordered events occur at a frequency of between every 20 seconds to approximately every 90 seconds, this creates an oscillation in HR in the VLF band (which includes all HR oscillations at between every 20 seconds and every 5 minutes). This oscillation is also referred to as cyclic variation of HR (CVHR). Not surprisingly, this oscillation adds power at that underlying frequency in the HRV power spectral plot, and the center frequency of that peak also indicates the underlying frequency of the CVHR. The plot from the older man with high HRV (see Fig. 16–3A) has considerable power in the VLF band during the night. In addition, there is a suggestion of a distinct VLF peak during different hours of the night. The plot from the older healthy man with average HRV (see Fig. 16–3C) shows considerably more VLF power in the 05:00, 06:00, and 08:00 hours than during the rest of the night, suggesting the presence of sleep-disordered breathing HR patterns during that period. The plots from the older healthy man with low HRV in Fig. 16–3D also suggest the presence of a peak in the VLF band between 00:00 and 03:00. In Fig. 16–4C, the plots from 01:00 and 02:00 show large VLF peaks corresponding to periods of sleep-disordered breathing that are not visible in the other examples but are relatively common in recordings from older adults and from cardiac patients. Although hourly power spectral plots are most useful for detecting CVHR peaks, often, if significant sleep-disordered breathing or periodic limb movements are present, this large VLF peak can be seen on the single averaged 24-hour FFT plot.

Abnormal Organization of HRV Patterns on FFT

Not all FFT plots show a normal distribution of underlying rhythms in the different HRV bands. Aside from plots where there is little or no area under the curve, which indicate severely depressed autonomic function, other plots simply show an abnormal distribution of spectral power. Figure 16–4B,D shows examples of clearly abnormal plots. A side-by-side comparison of the normal and abnormal plots permits one to intuitively appreciate the differences. Abnormal plots are less organized. They have more area under the curve at frequencies above the HF band. They lack the characteristic HF peak or, in some cases, have a broad peak in the "wrong" place. This abnormal distribution of spectral power could be due to extremely inaccurate Holter scanning but more often indicates underlying irregularities in HR patterns that do not reflect normal cardiac autonomic function (erratic rhythm, as mentioned earlier) and may reflect higher risk of cardiovascular death. In both the Cardiovascular Health Study (CHS)5 and the Cardiac Arrhythmia Suppression Trial (CAST),11 we have reported that a higher prevalence of hours with erratic rhythm HR patterns is associated with increased risk of cardiovascular mortality. In the CAST study, a worsening of erratic rhythm HR patterns in association with antiarrhythmic therapy was also associated with increased risk of mortality.11

Although the precise origin of this abnormal rhythm has not been determined, it may reflect a form of sinus node dysfunction. However, from the perspective of using HRV to characterize cardiac autonomic function, as we have emphasized, these abnormal patterns are associated with higher HRV numbers that do not translate into "better" autonomic function.5 Therefore, it is recommended that the underlying distribution of the HRV power spectrum be examined on an hourly basis and that clearly abnormal cases be excluded to avoid confounding HRV results. Because, as seen in this example, patients with periods of abnormal HR patterns do not necessarily have this rhythm throughout their entire recording, the question of how to adjust for this in determining HRV as a measure of autonomic function remains open. Clearly though, when this abnormal rhythm is prevalent during the recording, some HRV measures will not be meaningful. This topic will be discussed in greater detail at the end of this chapter.

A Poincaré plot is generated by plotting the time between two heartbeats versus the time between the next two for each pair of intervals on the recording. Although some investigators generate these plots from all R-R intervals, it can be more informative to use plots of N-N interbeat intervals only, because otherwise it is impossible to determine whether outliers are due to ectopy or erratic rhythm. Indeed, some investigators do deal with this by filtering or ignoring these beats, potentially measuring HRV only during the periods of normal sinus rhythm. This filtering, however, could result in the loss of potentially important information.

Just as with any other graphic display, hourly Poincaré plots show the amount of HRV, as seen in the area of the plots, in greater detail than a single 24-hour plot; more importantly, they provide a clear way to visualize the structure of the HR pattern at different times during the recording. Normal hourly plots are somewhat comet shaped or perhaps ellipsoid, although hours that contain both sleep and active periods might have two clusters. Abnormal plots are more disorganized, ranging from plots with a clearly visible central part and some outlier points to plots that seem splattered all over. Highly disorganized Poincaré plots have also been referred to as "complex" plots. Prior studies have supported the clinical importance of abnormal Poincaré plots generated from 24-hour R-R interval data.12-14 These studies, have demonstrated an association between the presence of complex Poincaré plots and mortality among congestive heart failure patients. Abnormal Poincaré plots have also been associated with risk of ventricular tachycardia,15 and in another study from the same group, in post–coronary artery bypass graft patients, more random HR patterns measured on postoperative day 1 predicted ischemic events.16 Thus, Poincaré plots are also a sensitive way to identify recordings with an abnormal HR pattern in patients who could be at increased risk of cardiovascular events and in whom higher values for some HRV measures will be misleading.7

Figure 16–5 shows the hourly Poincaré plots for the subjects originally shown in Figs. 16–1 and 16–3. All of the plots are normally organized, although the young subject with high HRV (see Fig. 16–5A) clearly has more HRV during the daytime than the older subject with high HRV (see Fig. 16–5B). The decreased HRV in the older adult with average HRV (see Fig. 16–5C) and the even more depressed HRV in the older adult with low HRV are clearly visible as reduced area in the Poincaré plots.

Figure 16–6 shows the hourly Poincaré plots for the subjects with known cardiovascular disease originally shown in Figs. 16–2 and 16–4. Although the decreased HRV reflecting blunted cardiac autonomic function is clearly visible, the shapes of the Poincaré plots in Fig. 16–6A,C are unremarkable. However, the plots in Fig. 16–6B reflect a combination of periods of essential normal sinus rhythm and periods of highly disorganized/erratic sinus rhythm. The plots for the patient with atrial fibrillation (see Fig. 16–6D) are, as would be expected, completely abnormal.

A plot of instantaneous HR on a beat-to-beat basis and on an appropriate scale provides the most direct possible image of HRV. We have found that plotting HR on a series of 10-minute scales (0-100 beats/min) is excellent for these analyses. Only N-N interbeat intervals are used for this analysis. Figure 16–7 is an example of 1 hour's worth of data—six stacked 10-minute plots during quiet sleep. Respiratory sinus arrhythmia, seen as zig-zag HR patterns on the plots, and occasional arousals from sleep are clearly seen.

HR Tachograms to Identify Sleep Periods

Although the hour of lying down can usually be identified from hourly power spectral plots, a more precise estimate of lying down and getting up times can be made from beat-to-beat HR tachograms. The onset of lying down or sleep can be seen as the beginning of a period of lower and more regular HR and often the onset of a clearly seen HR pattern reflecting respiratory sinus arrhythmia. This clearly seen respiratory sinus arrhythmia pattern during sleep also provides information about breath-by-breath respiratory rate and a qualitative sense of the changing depth of respiration, which will be discussed in greater detail later. Figure 16–8 illustrates the difference in HR patterns between asleep and awake periods, 10 minutes apart, in the same subject (the older adult with average HRV). Note the higher mean HR (indicated by the average HR line) and the far greater irregularity of HR patterns during wake.

HR Tachograms to Identify Sleep-Disordered Breathing

Although the presence of CVHR can often be identified from the hourly power spectral plot as described earlier, HR tachograms clearly identify the magnitude and duration of these cycles of regular arousals during sleep. The pattern of this CVHR often provides clues as to its origin. For example, especially during the early part of the night, repeated low-amplitude, relatively brief CVHR cycles at approximately 20-second intervals are often seen and suggest periodic limb movements. Higher amplitude, longer cycles suggest sleep-disordered breathing. A sharp increase in HR usually occurs when the patient arouses to breathe. HR returns to baseline when the patient falls back to sleep, at which point if there is sleep-disordered breathing, the airway collapses again. These events are associated with significant autonomic activation and have been shown, in some patients, to be association with spectacular surges in blood pressure. Although intuitively one might suspect that larger HR cycles are associated with more severe sleep-disordered breathing, the magnitude of these cycles is also influenced by underlying autonomic function, and in the case of severely impaired autonomic function, severe sleep apnea HR changes might be barely visible on the tachogram. Note that respiratory sinus arrhythmia patterns can persist during the period of airflow obstruction, indicating a continuing respiratory effort. Although this pattern is often referred to as a bradytachy pattern, the commonly held view that obstruction of airflow is always associated with a significant bradycardia is not correct.17 Such bradycardias do occur, especially in the presence of severe oxygen desaturations, but they are not a hallmark of the syndrome. Rather, the bradytachy pattern seen on electrocardiogram reflects the tachycardia of the HR arousal and the return to baseline HR after breathing is resumed.

Figure 16–9 shows 10-minute sections of the HR tachograms from the subjects whose FFT plots in Figs. 16–3 and 16–4 previously suggested the presence of sleep-disordered breathing. In Fig. 16–9A, showing the older adult with high HRV, HR arousals are very sharp, and sleep is disrupted. This suggests either limb movements or perhaps what would be scored as "arousals for no apparent reason." The older adult with average HRV shown in Fig. 16–9B did not have CVHR until the morning hours. This is a typical pattern consistent with the greater amount of rapid eye movement (REM) sleep (with more intense CVHR) toward morning and the possibility that the patient is lying on his back, which increases the likelihood of sleep-disordered breathing. In Fig. 16–9C, the sleep-disordered breathing HR pattern of the older adult with average HRV is seen. The mixture of longer and short HR arousals suggests the possibility of a mixture of sleep-disordered breathing events of varied severity. Finally, Fig. 16–9D shows CVHR from the cardiac patient originally seen in Fig. 16–2C. This CVHR pattern of relatively short HR arousals suggests the presence of hypopneas (periods of decreased airflow) rather than obstructive apneas (full cessation of airflow).

None of the patients selected to illustrate various aspects of HR pattern analysis proved to have either severe obstructive sleep apnea HR patterns or central sleep apnea HR patterns. Therefore, these patterns are illustrated in Fig. 16–10. Figure 16–10A shows typical HR patterns for severe obstructive sleep apnea. Note the sharp increase in HR at the beginning of the CVHR and the relative abrupt decrease in HR at the termination of the event. Although the examples shown in Fig. 16–10 have regular cycles of CVHR, regularity is not always present.18 Of special clinical interest is the identification of central sleep apnea (lack of airflow due to failure to breathe) and Cheyne-Stokes respiration (a combination of central hypopnea or apnea and then periodic respiration) from HR tachograms. CVHR due to central sleep apnea is typically of lower amplitude and flatter shape than that due to apneas or hypopneas, because the resumption of respiration is not associated with a sharp HR arousal. Patients with central sleep apnea tend to have relatively low HRV on the tachogram; however, respiratory sinus arrhythmia is usually visible, and therefore, the lack of respiratory effort or the periodic respiration can usually be identified. Figure 16–10B shows a tachogram with a central sleep apnea HR pattern. Figure 16–10C shows clearly identifiable Cheyne-Stokes respiration HR patterns.

HR Tachograms to Identify Erratic Sinus Rhythm

Beat-to-beat HR tachograms can also reveal arrhythmias that are not apparent on ordinary Holter scanning. Respiratory sinus arrhythmia, as has been seen in the previous examples, is seen as a saw-tooth pattern on the HR tachogram and is especially visible during supine rest. However, there is another form of sinus arrhythmia that is highly irregular even though there is no visible difference in the P waves or PR intervals on Holter scanning. As previously mentioned, the presence of this erratic sinus rhythm can be identified from both FFT plots and from Poincaré plots. However, HR tachograms provide the most direct evidence for its presence.5 Figures 16–11A (from the subject originally shown in Figs. 16–2B, 16–4B, and 16–6B) and 16–11B (from a different subject) show examples of HR tachograms with erratic HR patterns. As mentioned, time and frequency domain HRV measures are confounded by erratic rhythm. However, many of the newer, nonlinear measures (eg, the short-term fractal scaling exponent) actually capture this increased randomness in HR patterns, which is likely the reason they perform so well as predictors of cardiovascular outcomes.18-20

Other Patterns Potentially Visible on HR Tachograms

Extremely abnormal HR tachograms can suggest severe scanning error, and investigation of the original recording is advised before any conclusions are drawn. Additional patterns are also seen on HR tachograms, although their clinical significance is unknown. Tachograms reveal episodes of "sinus bigeminy," an alternating pattern of slightly faster and slower HRs that is often not noticed on Holter scanning and that is remarkably common. Tachograms also reveal various oscillating HR patterns, including one with an oscillation at approximately 6 cycles/min that we have observed in a small number of high-risk patients and others where the HR cycles up and down for long periods of time, as if the system cannot find a stable state. Although generation of numbers for statistical comparison is the goal of HRV research, it cannot be emphasized enough that looking at the primary data, the actual beat-to-beat HR patterns, is essential for results to be meaningful.

Graphical analysis of HR patterns can potentially permit more accurate measurement of HRV. To apply this, it is important to qualitatively understand the scope of different HRV measures. Certain measures primarily reflect total HRV and therefore mostly circadian rhythms. These measures include SDNN, the standard deviation of the averages of N-N intervals for every 5 minutes (SDANN), total power, and ULF power. Because the largest contribution to HRV is circadian rhythm, these measures are less affected by erratic rhythm, sleep-disordered HR patterns, or even scanning error unless it is really severe.5 Of the circadian HRV measures, SDANN and ULF power, which are roughly equivalent,21 are the least affected and can, arguably, even be used to measure HRV in cases of atrial fibrillation.22 Thus, results of a 24-hour recording can generally categorize subjects as having low or high HRV even in the presence of very abnormal rhythms. Average HR is also unaffected by these abnormal rhythms, although it could be elevated by the arousals of severe sleep apnea. However, a high SDNN due to a monotonically increasing HR through the entire recording (which we have seen) is clearly not comparable to an SDNN that reflects circadian rhythm. This supports the importance of examining plots of HR patterns or, at the very least, examining hourly average HRs.

The second group of HRV measures is "intermediate" and reflects HRV on a scale of approximately 5 minutes or less. These measures include the average standard deviation of 5-minute N-N intervals (SDNNIDX), VLF power (which reflects variations in HR at cycles of 20 seconds to 5 minutes), and LF power (which reflects variations in HR at 3-9 cycles/min). VLF power is especially unaffected by erratic rhythm, sinus bigeminy, and subtle scanning error because it does not include cycles below every 20 seconds, but it is exaggerated by sleep apnea (although only for part of the 24-hour period) and by oscillating HR patterns. LF power reflects underlying oscillations of HR at about 3 to 9 cycles/min and is exaggerated by abnormal HR patterns.

The third class of HRV measures reflects beat-to-beat changes. This group includes the percentage of N-Ns >50 ms different from the prior one (pNN50), root mean square of the average difference in N-N interval between successive pairs of beats (rMSSD), HF power, and various ratio measures such as the LF/HF ratio, normalized LF power, and normalized HF power. It is often assumed that these measures reflect parasympathetic control of HR or, in the case of the ratio measures, some sort of balance between the two arms of the autonomic nervous system.1 These measures, however, are totally confounded by erratic rhythm, scanning error, and sinus bigeminy, which significantly exaggerate them.5 Putting them into the denominator, as is done with the ratio measures, simply adds to the problem.

Thus, examination of graphical HR patterns is essential before deciding to include subjects in a standard HRV analysis. When graphical plots seem normal, HRV can meaningfully be calculated. When they are clearly abnormal, it is likely that only the circadian measures and VLF power will be meaningful. Short-term HRV measures are meaningless when erratic rhythm is present, and erratic rhythm is common among both cardiac patients and older adults in whom HRV may be useful for risk stratification.

Scanning of all of the Holter recordings shown in the figures was carefully verified. Revisiting the series of plots shown for the different subjects, it becomes clear that some subjects have normally configured plots by every test. In those subjects, traditional time or frequency domain HRV, whether high or low, can be calculated and should reflect cardiac autonomic function. Although, with experience, direct visualization of the HR tachogram provides an excellent starting point, a set of 24 hourly Poincaré plots and FFTs, each printed on a single page, provide a rapid way to assess each continuous electrocardiogram recording, and those that are normally organized by FFT and Poincaré plot criteria can generally be accepted for HRV analysis. The question of what to do with abnormal plots is complex and is the subject of ongoing investigations. At a minimum, the Holter scanning should be verified to ensure that the irregular rhythm is not due to undetected ectopy or nonuniform detection of beat onsets. If, however, this rhythm is truly present, as mentioned, circadian HRV measures, calculation of VLF power, and nonlinear methods can still be applied to characterize HRV for research studies. Short-term measures are meaningless in these cases, expect that when they appear to be unusually high relative to total HRV or to the population being studied, they are a red flag for erratic rhythm.

Clinically, each of these plots, including hourly HR tachograms plotted on a 10-minute scale, is available from at least one commercial Holter system, but no system, to our knowledge, provides all of them. Some clinical scanners attempt to detect CVHR that indicates sleep-disordered breathing. Measures like the short-term fractal scaling exponent and other novel nonlinear HRV measures are not yet commercially available. These require high-quality Holter scanning and further validation of their potential clinical utility.

The relationship of autonomic assessments from imaging and HRV analysis has been explored in numerous prior studies. For example, HRV has been studied in relation to cardiac iodine 123 (123I)-meta-iodobenzylguanidine (MIBG) uptake, a direct measure of sympathetic nerve innervation of the ventricles of the heart. Not surprisingly, although sympathetic innervation of the ventricles and sympathetic innervation of the sinoatrial node are not the same thing, post–myocardial infarction patients with globally impaired cardiac 123I-MIBG were also found to have impaired HRV.23 The relationship between HRV and cardiac 123I-MIBG parameters was confirmed in a group of implantable cardioverter-defibrillator (ICD) recipients by Koutelou et al24 and among diabetic patients by Freeman et al.25 Furthermore, baroreceptor sensitivity, LF power, rMSSD, and MIBG washout were independent predictors of fast ventricular arrhythmia episodes over an average of 32 months of follow-up among the ICD recipients.24 The relationship of HRV to ventricular function during exercise as assessed by radionuclide imaging has also been explored. For example, Nagaoka et al26 reported a relationship between depressed ventricular response to exercise and decreased HRV during different time periods in patients with idiopathic cardiomyopathy. However, these investigations have been limited by the assumption that low HRV is the only marker for abnormalities in cardiac autonomic function and high values for HRV reflect greater health. Moreover, the nature of the underlying HR patterns was never taken into account. As has been illustrated earlier, low HRV is only one form of abnormal HRV. Thus, future studies incorporating both numeric and graphical methods for analysis of HRV promise to enhance the combined power imaging and HRV for risk stratification and understanding of the underlying physiology of different cardiac pathologies.

Use of graphical plots of HR patterns provides additional clinically relevant information that would be missed by the numeric measurements; in addition, visualization of HR patterns by some graphical method is absolutely essential before many of the numerical measurements can be said to have any true meaning. With current computer technology, generation of graphical plots of HR patterns is neither difficult nor time consuming. Potentially, therefore, the future of ambulatory ECG analysis, whether from Holter recordings or some other form of inpatient or outpatient telemetry, will incorporate such plots, thereby markedly adding to the utility of HRV-type analysis for both research and clinical practice.