Heart rate variability, sleep and sleep disorders

Sleep Medicine Reviews 16 (2012) 47e66

Contents lists available at ScienceDirect

Sleep Medicine Reviews journal homepage: www.elsevier.com/locate/smrv

CLINICAL REVIEW

Heart rate variability, sleep and sleep disorders Phyllis K. Stein a, *, Yachuan Pu b, c a b

Washington University, School of Medicine HRV Laboratory, 4625 Lindell Boulevard, Suite 402, Saint Louis, MO 63108, USA CardioNet Inc., 1010 Second Avenue, Suite 700, San Diego, CA 92677, USA

a r t i c l e i n f o

s u m m a r y

Article history: Received 28 December 2010 Received in revised form 24 February 2011 Accepted 25 February 2011 Available online 11 June 2011

Heart rate (HR) is modulated by the combined effects of the sympathetic and parasympathetic nervous systems. Therefore, measurement of changes in HR over time (heart rate variability or HRV) provides information about autonomic functioning. HRV has been used to identify high risk people, understand the autonomic components of different disorders and to evaluate the effect of different interventions, etc. Since the signal required to measure HRV is already being collected on the electrocardiogram (ECG) channel of the polysomnogram (PSG), collecting data for research on HRV and sleep is straightforward, but applications have been limited. As reviewed here, HRV has been applied to understand autonomic changes during different sleep stages. It has also been applied to understand the effect of sleep-disordered breathing, periodic limb movements and insomnia both during sleep and during the daytime. HRV has been successfully used to screen people for possible referral to a Sleep Lab. It has also been used to monitor the effects of continuous positive airway pressure (CPAP). A novel HRV measure, cardiopulmonary coupling (CPC) has been proposed for sleep quality. Evidence also suggests that HRV collected during a PSG can be used in risk stratification models, at least for older adults. Caveats for accurate interpretation of HRV are also presented. Ó 2011 Elsevier Ltd. All rights reserved.

Keywords: Autonomic function Cardiopulmonary coupling Cyclic variation of heart rate Erratic rhythm Heart rate variability Insomnia Respiratory sinus arrhythmia Sleep-disordered breathing

Introduction Heart rate (HR) is modulated on a beat-to-beat basis by the combined effects of the sympathetic (SNS) and parasympathetic (PNS) nervous systems on the sino-atrial node. Therefore, analysis of changes in HR over time (heart rate variability or HRV) provides information about autonomic functioning. In clinical conditions associated with autonomic dysfunction, (e.g., congestive heart failure, diabetes, end-stage renal disease, etc.), abnormal, usually decreased, HRV is generally found. Moreover, abnormal HRV is an independent risk factor for mortality both in clinical and population studies.1 It should be noted that HRV cannot reflect autonomic “tone” which can only be measured using pharmacological blockade. Moreover, HRV is a black box with HR as the output. Hence the cause of decreased HRV, whether a lack of central signaling, lack of reflex feedback to the central nervous system or lack of responsiveness of the heart itself, cannot be determined. HRV is generally derived from mathematical analyses of intervals between normal heart beats (NN intervals) and requires

* Corresponding author. Tel.: þ1 314 286 1350; fax: þ1 314 286 1394. E-mail addresses: [email protected] (P.K. Stein), [email protected] (Y. Pu). c Tel.: þ1 949 610 3181. 1087-0792/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.smrv.2011.02.005

patients to be in sinus rhythm for most measures to be meaningful from HRV alone. However, one HRV measure, HR turbulence, is based on the NN interval response to ventricular ectopic beats. Although millions of continuous polysomnogram (PSG) electronic electrocardiogram (ECG) signals from which HRV could potentially be calculated have been stored, the potential for obtaining additional, clinically relevant information from them has scarcely been tapped. At the same time, there are a huge number of 24-h Holter recordings, and increasingly, multi-day telemetry recordings from which not only HRV, but clinically relevant information about sleep could be derived, yet these important data are generally ignored. Thus, in the current review, we will focus on a basic understanding of HRV and then on potential sleep-related clinical applications of HRV from both PSG ECGs and 24-h ambulatory recordings. For each section, a table will be provided that describes the population and methods for citations. Measurement of HRV The starting point for HRV analysis is a list (a “beat file”) of the intervals in milliseconds between heart beats on the ECG recording that includes the morphology of each heart beat so that normal, ectopic, paced beats and artifact can immediately be identified. Although beat files (or RR interval files) can be exported from many

48

P.K. Stein, Y. Pu / Sleep Medicine Reviews 16 (2012) 47e66

Table 1 Time-domain HRV measures. AVNN (ms)

Average of NN intervals for period of interest

SDNN (ms)

Standard deviation of NN intervals for period of interest Standard deviation of AVNN for 5-min intervals for period of interest Average of 5-min standard deviations of NeN intervals for period of interest Percent of NN intervals > 50 ms different from previous (NN) for period of interest Percent of NN intervals different from previous by 6.25% or more of local AVNN (NN) for period of interest Root mean square of successive differences of NN intervals for period of interest Average coefficient of variance (SD/Mean) for 5-min intervals for period of interest

SDANN (ms) SDNNIDX (ms) pNN50 (%) pNN625 (%) rMSSD (ms) CV

Can convert to average HR of NN intervals (HR ¼ 60,000/AVNN) Reflects total HRV Reflects primarily circadian HRV Reflects average short-term HRV and combined SNS and PNS influences With normal sinus rhythm reflects vagal activity With normal sinus rhythm reflects vagal activity normalized by HR With normal sinus rhythm reflects vagal activity Reflects average short-term HRV normalized by HR

HR: heart rate. HRV: heart rate variability PNS: parasympathetic nervous system SNS: sympathetic nervous system

PSGs, they usually do not have morphology annotations. Thus generating an accurate beat file generally requires some form of Holter scanning, just as PSG software determinations of sleep parameters need to be over read to provide an accurate sleep study. The ECG channel can easily be exported to a Holter scanner for accurate beat detection and annotation. However, because certain HRV measures are relatively insensitive to scanning error when calculated over a longer period of time, the clinical utility of HRV parameters derived from PSG RR interval files, perhaps with appropriate filtering for clearly ectopic beats, bears investigation. Traditionally, HRV is measured in the “time domain” and the “frequency domain.” Time-domain HRV (Table 1) is a set of statistical measures derived from the beat file. The most commonly used ones are described below. The most global HRV measure is SDNN (the standard deviation of all NN intervals). SDNN is usually measured on the entire recording, but it can be measured on any segment (e.g., every 5 min) or during specific sleep stages. It captures total HRV and is one of the measures that are relatively insensitive to small errors in scanning. In the late 1980s, 24-h SDNN < 50 ms was shown to identify post-myocardial infarction (MI) patients at an adjusted risk of 5.3 for mortality over a mean of 31 months follow-up compared to patients with SDNN > 100 ms.2 Adjustment for covariates did not explain this association, although relative risk declined to 2.8. However, in the era of modern therapy, the utility of SDNN per se to predict survival has been attenuated because of the markedly reduced prevalence of very low SDNN, although other HRV measures continue to risk stratify post-MI patients.3 A similar measure to SDNN is SDANN, which is the standard deviation of the 5-min averages of interbeat intervals. It is insensitive to scanning error. However, because low global values for these measures are primarily driven by a lack of circadian rhythm, it is unclear how useful they would be as global measures of HRV during PSGs. Other time-domain HRV measures capture more short-term variations in HR. SDNN index is the average of the 5-min standard deviations of NN intervals. The degree to which HRV changes on a beat-to-beat basis is reflected in two common time-domain measures, pNN50 (the % of beats where the change from one beat to the next is > 50 ms) and rMSSD (the average change in interbeat interval between beats). These are measured over the entire recording, but can easily be measured over specific time periods (e.g., for each 5 min) or by sleep stage. When patients are in truly normal sinus rhythm, both rMSSD and pNN50 reflect PNS control of HR, and changes in these

parameters, e.g., between sleep stages, serve as markers of changing PNS activity. This is because changes in PNS nerve traffic mediate HR through changes in acetylcholine binding, which is instantaneous, whereas changes in SNS nerve traffic are mediated by a complex signaling cascade that is initiated by the binding of norepinephrine, resulting in a time delay before they can be affected. Thus, changes in HR at (faster) respiratory frequencies (respiratory sinus arrhythmia or RSA) are due to PNS signaling, although when breathing gets slower than w9 breaths/min, HR changes will reflect both SNS and PNS nerve traffic. Sleep is an excellent time to measure PNS function because HR is primarily under PNS control during supine rest and RSA becomes prominent. As will be described in more detail in a later section, however, rMSSD and pNN50 measures are problematic, because they cannot differentiate between increased HRV due to RSA and HRV due to scanning error (uneven beat detection, missed or misclassified beats) or from irregular HR patterns (erratic rhythm) that are not reflective of better PNS functioning.4 This phenomenon is illustrated in Fig. 1 which shows the HR tachogram and the HR tachogram integrated with PSG respiratory signals during a period of normal sinus rhythm and one of erratic rhythm in the same subject. As can be seen, HR fluctuations correlate poorly with respiration during erratic rhythm and closely track them during true sinus rhythm. On the other hand, markedly decreased values for PNS HRV measures, whether from a PSG RR interval file or from Holter scanning, would reliably reflect severely blunted autonomic control of HR. Frequency domain HRV (Table 2) parses out the variance in beat-to-beat HR into its underlying components at different frequencies using fast-Fourier transforms (FFTs) or equivalent techniques. It should be emphasized that calculation of most frequency domain HRV requires a condition called “stationarity,” i.e., that the mean and variance of the signal do not change significantly at different points in the recording. In order to roughly meet this requirement, some frequency domain HRV measures are calculated over shorter intervals, e,g., 5 min or less, and averaged as needed but this also means that some frequency domain measures are less useful when the HR is changing rapidly. Frequency domain HRV measures have their counterparts in the time domain but often perform better in risk models. Values are generally skewed and most analyses are performed on natural logtransformed values. Total power, when measured over 24 h, should be very similar to SDNN squared, since the variance is mathematically the square of the standard deviation. Ultra low frequency power (ULF) is mathematically similar to SDANN.5 It reflects

P.K. Stein, Y. Pu / Sleep Medicine Reviews 16 (2012) 47e66

49

Fig. 1. (a) 10-min plot of instantaneous HR vs. time (HR tachogram) showing a period of erratic rhythm during sleep. Y-axis is 40e100 bpm. X-axis is time in minutes (starting from 0). (b) MatLab plot of PSG channels corresponding to the shaded area in (a) shows the lack of relationship of HR changes and respiration. (c) 10-min HR tachogram showing normal sinus rhythm in sleep of the same subject as in (a) and (b). (d) MatLab plot of the PSG channel corresponding to the shaded area in (c) showing a close coupling between sinus arrhythmia and respiration. HR: heart rate PSG: polysomnogram

underlying periodicities in the HR signal on a scale of every 5 min to every 24 h (if the recording is that long). Again, this is primarily a measure of circadian rhythm and relatively insensitive to scanning error. Whether PSG-derived ULF power would be clinically meaningful is unknown. Decreased very low frequency power (VLF), on the other hand, has been associated with clinical outcomes. VLF reflects underlying periodicities in HR at

frequencies of every 25 s to every 5 min (0.0033e0.04 Hz). As described in more detail below, VLF encompasses the underlying frequency of most sleep-disordered breathing and periodic limb movements. Limited data suggest that VLF is modulated by a combination of the renineangiotensin system (since it is reduced by angiotensin converting enzyme (ACE) inhibition) and the PNS (since it is blocked by atropine), but not by the SNS since it is not

50

P.K. Stein, Y. Pu / Sleep Medicine Reviews 16 (2012) 47e66

Table 2 Frequency domain HRV measures. TP (ms2) ULF (ms2) VLF (ms2)

LF (ms2) 2

HF (ms )

LF/HF LFnu (%) HFnu (%)

Total power over measured period Ultra low frequency power measures rhythms greater than every 5 min Very low frequency power measures rhythms between every 25 sec and every 5 min, i.e., 0.0033e0.04 Hz) LF measures HR rhythms from 2.5 to 9 cycles/min, i.e., 0.04e0.15 Hz). Averaged over 5-min or less. HF captures variations in HR due to respiratory sinus arrhythmia at 9e24 cycles/min, i.e., 0.15e0.4 Hz. Averaged over 5-min periods or less. LF/HF average over 5-min periods or less [LF/(TP  VLF)] for the measured period (5-min or less). Some calculate LF/(LF þ HF) [HF/(TP  VLF)] for the measured period (5-min or less). Some calculate HF/(LF þ HF)

Reflects total HRV Reflects circadian HRV Reflects vagal and renin-angiotensin system effects on HR. Exaggerated by SDB Reflect combination of SNS and PNS influences. Captures baroreflex rhythms Under normal circumstances reflects vagal activity

Purported to reflect SNS/PNS balance Purported to reflect SNS activity Purported to reflect PNS activity

HR: heart rate HRV: heart rate variability PNS: parasympathetic nervous system SDB: sleep-disordered breathing SNS: sympathetic nervous system

affected by beta-blockade.6 VLF should be relatively insensitive to small scanning errors and plausibly could be derived from the RR interval file of a PSG. Faster underlying periodicities in HR patterns are captured by low frequency (LF) power (variations from every 6.7 s to every 25 s, 0.04e0.15 Hz) and high frequency (HF) power (variations at normal respiratory frequencies, 9e24 times/min, 0.15e0.4 Hz). LF power is modulated by signals from both SNS and PNS inputs to the sino-atrial node and is believed to reflect baroreflex function.1 HF power is modulated by PNS inputs only but, as previously mentioned, erratic rhythm can exaggerate HF as well. Relatively accurate scanning is needed to measure LF and HF, more so for HF. However, markedly decreased values for LF or HF from PSG RR interval files or other automatically scanned signals would indicate autonomic dysfunction. Both LF and HF are measured over shorter periods of time (usually 2e5 min periods but as little as 1 min) and then averaged. Analysis over shorter time periods can be used to track trends over time. A family of ratio measures of HRV has been derived from LF and HF power.1 Since LF is modulated by both SNS and PNS activity and HF is modulated by PNS activity, it is attractive to assume that adjustment of LF by HF will somehow provide a measure of the SNS component of HRV or of “sympathovagal balance.” Proposed measures include the LF/HF ratio and normalized LF or HF power, i.e., LF or HF divided by (total power minus VLF power) averaged over, say, 5 min windows. Often, the LF/HF ratio is used to assess changes in autonomic function between sleep stages. Although there are many situations, e.g., tilt table testing and the transition from non-rapid eye movement(NREM) (non-rapid eye movement) to rapid eye movement (REM) sleep, where LF/HF generally increases with increased SNS activity, there are others where it does not, e.g, exercise and heart failure. Also, beta-blockade, which inhibits sympathetic activity, should markedly decrease the LF/HF ratio, but that is not the case. Furthermore, since both measures, and especially HF, can be confounded by scanning error and erratic rhythm, creating a ratio further muddies the situation. The assumption that SNS and PNS activity have a see-saw relationship, that an increase in one implies a decrease in the other, is also simplistic. Thus, although commonly reported and potentially meaningful, these measures need to be interpreted with caution. While both time and frequency domain HRV reflect the amount of HRV, another family of measures reflects the underlying organization of HRV. These measures, called non-linear measures, characterize how random or correlated the HR patterns are. Thus, two patients could have the same values for SDNN, but one could have

a very normal, organized HR pattern and the other could have a highly random and disorganized HR pattern. Non-linear HRV measures remain in the research domain and because of space limitations will not be reviewed here. A fourth, very different form of HRV is HR turbulence (HRT) which characterizes the HR response to ventricular premature beats and may reflect baroreflex function.7 Although there is no space to review HRT in detail here, it has been a powerful predictor of outcome in both clinical and population studies,7 and has shown promise in the Sleep Heart Health Study.8 HRV during different sleep stages (Table 3) One sleep-related application of HRV explores changing autonomic function during different stages of sleep. Zemaityte et al. studied HR and HRV changes by sleep stage in healthy young adult males.9 HR decreased in association with decreased variability in sleep stages 1, 2, 3 and 4, whereas HR increased, with increased variability, in REM sleep. HR during REM sleep was higher than during wakefulness. The high frequency peak was pronounced during slow-wave sleep, and was abolished with intravenous injection of atropine (i.e., blocking of the PNS) but not affected by intravenous injection of propranolol (a non-selective beta blocker), compared with baseline. Findings from Bonnet et al. and also Otzenberger et al. were consistent with those of Zemaityte et al.10,11 Scholz et al. demonstrated a decreased LF/HF ratio associated with synchronized sleep (reflecting the high values for HF during these stages) and a significantly increased LF/HF during REM sleep (possibly reflecting increased SNS activity during REM sleep).12 Crasset et al. compared HRV between young and older healthy subjects during wakefulness and different sleep stages, and found older subjects had faster HRs and lower HF during non-REM sleep than younger ones, indicating that aging affects cardiac vagal activity during nighttime.13 Others have suggested that the relationship of HRV to sleep stage could be altered by underlying clinical disorders. Vanoli et al. showed that post-MI patients had significantly higher LF/HF ratio than normal subjects did during both non-REM and REM sleep. Also, post-MI patients had an increase in LF/HF ratio from wake to non-REM sleep, instead of the decrease seen among normal subjects.14 Shinar et al. reported no differences in changes in frequency domain HRV parameters during different sleep stages between normal, obstructive sleep apnea (OSA) and other sleep disordered breathing (SDB) groups, although the LF/HF ratio was higher before and after sleep onset in both SDB groups.15 Among

P.K. Stein, Y. Pu / Sleep Medicine Reviews 16 (2012) 47e66

51

Table 3 Studies of HRV by sleep stage. First author/year

Subjects

Methods

HRV measure

Zemaityte et al., 198410

N ¼ 20 healthy males, (age: 25.6  5.1)

PSG

Mean and SDNN

Bonnet and Arand, 1997

11

Otzenberger et al., 1998

12

Scholz et al., 199713

Crasset et al., 200114

Vanoli et al., 199515

Shinar et al., 200616

Stein et al., 200817 Lanfranchi et al., 200718

Data collection/analysis

Selected 200 consecutive heart beats/stage, Poincare plot, histogram, and power spectral density used for illustration N ¼ 12 normal (age: 22e48) PSG Normalized LF, normalized HF 5-min block within same sleep stage (wake, stage 2, REM) sleep, w100 blocks/night, BMDP IT spectral analysis program N ¼ 15 healthy males, PSG Pearson coefficient Consecutive 5-min (23:00 and 07:00), (age: 22e30) (Poincare plot), Poincaré Plot, Hanning-Windowed SDNN, rMSSD, LF, HF, LF/HF Nonparametric FFT N ¼ 20 healthy volunteers PSG, Vegetative SDNN, pNN50 VLF, LF, HF, Stationary (with HR variance determined tests in the and LF/HF ratio by vegetative tests) segments  100 beats morning between 70 and 130 EEG transients selected, preceding PSG HRV from autoregressive parametric model based spectral analysis, HRV averaged over same stage or condition N ¼ 8 normal (age: 22.5  3.3, PSG LF, HF, total power, LF and HF Stationary segments from normal subjects all male), N ¼ 8 normal normalized by total power averaged for wake (565), non-REM sleep (1026), (age: 55.0  7.3, 2F/6M), N ¼ 8 heart and REM sleep (265), FFT based transplant (age: 56.5  6.4, 4F/4M) spectral analysis 8 pairs of age-matched normal PSG TP, LF, HF, %LF ¼ Consecutive 5-min segments free of arousals (1F/1M) and recent MI (2F/6M) LF/TP, %HF ¼ HF/TP, LF/HF for awake, stable non-REM sleep, and REM sleep, commercial software N ¼ 12 normal (age: 28  16, 6F/6M), PSG VLF, LF, HF, LF/HF ratio, Min-by-min segments 9 min before and N ¼ 11 OSA (age: 40  9, 3F/8M), mean, SDNN 9 min after sleep onset defined by EEG, N ¼ 11 sleep disorders continuous wavelet analysis for frequency (age: 37  18, 3F/8M) parameters N ¼ 116 volunteers from Sleep Heart PSG Mean, rMSSD, SDNN, VLF Every 2-min calculated and averaged Health Study (age: 77  4) within same stage 10 Pairs of age- and sex- matched PSG Mean, SDNN, pNN50, LF, HF, 5-min stationary segment (no arousals, normal and idiopathic REM sleep normalized LF and HF PLMs, apneas, etc.) selected for stage 2 behavior disorder: (age 63  6, and REM sleep respectively, CardioLab Software 2F/8M)

BMDP IT: This is the name of a statistical package FFT: fast Fourier transform HF: high frequency power HRV: heart rate variability LF: low frequency power pNN50: percent of NN intervals> 50 ms different from previous for period of interest OSA: obstructive sleep apnea PSG: polysomnogram REM: rapid eye movement SDB: sleep-disordered breathing SDNN: standard deviation of NN intervals for period of interest rMSSD: root mean square of successive differences of NN intervals for period of interest TP: total power over measured period VLF: very low frequency power

elderly participants in the Sleep Heart Health Study, significant changes in both HR and HRV were seen for every sleep stage.16 In this study, consistent with others, both HR and time and frequency domain HRV were lowest in slow-wave sleep. However, in this elderly cohort, HR during REM sleep was lower than during wakefulness, and LF/HF ratio was not highest in REM sleep. Also, when Lanfranchi et al. studied HRV changes during sleep in REM Sleep Behavior Disorder, they found that RR interval and HF decreased while LF and LF/HF increased from non-REM to REM sleep in age-matched controls but did not change in patients.17

Practice points 1) HR and HRV progressively decrease during non-REM sleep and increase during REM sleep. 2) Results consistent with increased vagal control of HR in non-REM sleep and increased SNS control during REM sleep.

Research agenda 1) Larger studies on HRV patterns during sleep. 2) Establish expected sleep HRV patterns in different populations. 3) Investigate clinical applications of HRV by sleep stage in understanding neuroautonomic integration in different clinical disorders (e.g., dementia). 4) Studies of HRV by sleep stage at different parts of the night and of the evolution of HRV within sleep stages. 5) Investigate whether the effect on HRV by sleep stage can be a useful measure of the effects of drug treatment. 6) Investigate whether application of non-linear HRV measures can provide additional insights.

52

P.K. Stein, Y. Pu / Sleep Medicine Reviews 16 (2012) 47e66

Acute HR and HRV effect of Sleep Apnea (Table 4) With preserved cardiac autonomic function, sleep-disordered breathing (SDB) induces dramatic changes in HR and in hemodynamics.18,19 These result in an HR pattern termed cyclic variation of HR (CVHR), clearly seen when instantaneous HR is graphed on an appropriate scale.20 CVHR peaks are due to abrupt increases in HR, sometimes as much as 30 bpm, during the arousal phase that terminates SDB events. Fig. 2a shows CVHR on an HR tachogram for a patient with severe OSA and Fig. 2b shows the corresponding respiratory channels of the PSG. The presence of such CVHR peaks  30% of the time has been shown to be associated with a positive predictive value of > 90% in identifying severe OSA, defined as apnea hyponea index (AHI)  30).21 Fig. 2c and d shows CVHR for a patient with central sleep apnea with RSA disappearing when respiratory effort is absent. CVHR peaks are also seen in association with the arousal phase of hypopneas, upper airway resistance syndrome (UARS) and periodic leg movements (PLMs), and also in association with central respiratory events, and considerable suggestive information about underlying etiology can be gleaned from the magnitude, frequency and shape of the CVHR peaks and associated sinus arrhythmia. Moreover, CVHR does not have to be regular for it to be associated with significant SDB and the bradycardia assumed to always occur during the obstructive phase of all events is more often seen with severe desaturations. A detailed discussion of graphical HRV analysis during sleep is beyond the scope of the current review. Increases in HR, as seen with CVHR, reflect decreased PNS and/ or increased SNS control of HR. Direct measurement of PNS activity is not possible in humans, but SNS activity can be measured directly by peripheral muscle SNS nerve activity (MSNA).22 MSNA has demonstrated that increased SNS nerve traffic occurs during OSA events and that elevated SNS nerve discharge persists during the daytime in OSA patients.22 Measurement of frequency domain HRV per se during or immediately after OSA, however, is problematic. OSA can occur at an average frequency of about one/min. During that minute, there is obstruction, arousal and recovery. RSA usually persists, driven by continuing respiratory effort, during the obstructive phase,

which lasts, by definition, for >10 s (e.g., Fig. 2b). The arousal tachycardia often lasts w20e30 s during which HR is increasing, but the amplitude of the RSA may be increasing during the first few seconds because of the hyperpnea often seen after a respiratory event. Then the system returns to baseline, only to begin anew with the collapse of the airway. Measurement of HRV during the specific segments of this cycle is challenging, but so is comparing HRV in people who are and are not having events, because the events per se cause dramatic alterations in HRV, which makes the logic somewhat circular and violates the stationarity condition required for frequency domain HRV. Some investigators choose to measure HRV only during event-free periods, but this too is problematic because some patients do not have significant event-free periods during sleep and many are having subclinical events that are not scored (and therefore potentially “event free”) but still are associated with significant disruptions of autonomic functioning. Dingli et al. compared untreated OSA patients to healthy subjects and applied HRV analysis to 2-min windowed RR intervals centered on the end of SDB events.23 Events chosen for analysis had 2 min of apnea-free periods afterwards. Both normalized LF and HF were significantly higher (p < 0.001) at the end of such respiratory events during sleep, compared with baseline sleep, even though the RR interval duration showed no difference. The authors suggested that HRV results were consistent with SNS enhancement during sleep due to sleep apneas and may help explain increased cardiac risk. HRV after respiratory events was also studied by Guilleminault et al. in mild OSA and UARS patients. Both were symptomatic (Epworth sleepiness scale (ESS)  10) and free from medications, co-morbidities, morbid obesity, and any other sleep disorders.24 At the end of a respiratory event, a consistent pattern of increased HR was found, with greater changes in OSA subjects compared to UARS. A significant increase in HR was found, especially when EEG arousals were detected at the end of the respiratory event. During NREM, LF/HF after events was increased significantly for both UARS and OSA patients, but HF was decreased only for UARS. HRV showed no significant changes after respiratory events during REM sleep in either group. A limitation of such studies, as

Table 4 Acute HRV effect of Sleep Apnea. First author/year

Subjects

Methods

HRV measure

Data collection/analysis

Stein et al., 200321

N ¼ 46 OSA patients (age: 58  11, 12F/34M, mean BMI 33) N ¼ 80 patients referred to a tertiary sleep center to rule out OSA

PSG

RR interval

MCOT, ECG synced with PSG

RR interval

Tachogram display of beatebeat changes of RR, criteria for CVHR Tachogram display of beatebeat changes of RR, criteria for CVHR, automatic detection of CVHR 2-min centered at the end of 7e34 respiratory events/per patient, 2-min sliding window over entire record, Welch averaging of Hammingwindowed FFT (50% overlap) RR averaged every 4 s over 60 s centered at the end of each respiratory event (n ¼ 148), FFT on 120 s of stable RR intervals centered at the end of each respiratory event

22

Stein et al., 2009

Dingli et al., 200324

N ¼ 14 OSAHS and N ¼ 7 healthy controls, age- (51  9) and mean BMI-(29  2) matched

PSG

Mean, LF, HF, normalized LF and HF

Guilleminault et al., 200525

N ¼ 10 mild OSA (age: 46.5  10.0, mean BMI 29.6) and 10 UARS (age 39.7  8.6, mean BMI 29.3)

PSG

Mean, LF, HF, LF/HF ratio

BMI: body mass index CVHR: cyclic variation in heart rate FFT: fast Fourier transform HF: high frequency power LF: low frequency power MCOT: moblie cardiac outpatient telemetry OSA: obstructive sleep apnea OSASHS: obstructive sleep apnea hypopnea syndrome PSG: polysomnogram Synced: synchronized UARS: upper airway resistance syndrome

P.K. Stein, Y. Pu / Sleep Medicine Reviews 16 (2012) 47e66

53

Fig. 2. (a) 10-min HR tachogram showing severe OSA CVHR patterns (b) MatLab plot of the PSG channels corresponding to the shaded area in (a) showing the relationship of HR patterns to respiration. (c) 10-min HR tachogram showing CSA CVHR patterns. (d) MatLab plot of the PSG channels corresponding to the shaded area in (c) showing the relationship of HR and respiratory patterns. CSA: central sleep apnea CVHR: cyclic variation in heart rate OSA: obstructive sleep apnea

previously alluded to, is that respiratory events are followed by autonomic arousals and in the case of obstructive events by a large, brief increase in respiratory sinus arrhythmia due to compensatory hyperpnea. Thus, the period immediately after a respiratory event is one of non-stationary HR changes and this

lack of stationarity violates the assumptions under which frequency domain HRV can be calculated. From this point of view, it might make more sense to begin the calculation to compare post-event and baseline HRV after the arousal and hyperpnea phases are complete.

54

P.K. Stein, Y. Pu / Sleep Medicine Reviews 16 (2012) 47e66

Practice points Characterizing HR and HRV changes in association with sleep-disrupting events may potentially add information to PSGs by: 1) Focusing on autonomic disruption of underlying events rather than using fixed cutpoint to define significant events. 2) Identify patients with significant autonomic dysfunction who have a reduced or absent CVHR response to SDB.

Research agenda 1) Further investigate role of CVHR per se in adding to PSG diagnosis. 2) Study whether CVHR characteristics can reflect abnormalities of underlying autonomic function. 3) Calculate HRV during truly event-free periods to understand acute and chronic autonomic effects of different sleep disorders in different populations.

activity.29The effect of SDB on HRV during sleep was explored by Khoo et al. in the Sleep Heart Health Study.30 Using traditional HRV parameters and their respiration-adjusted counterparts, the authors found that mean RR interval negatively correlated with RDI in men for all sleepewake states (p < 0.001) after adjustment for age and BMI. In addition, HF and respiration-adjusted HF negatively correlated with RDI in men only during wakefulness. In women, LF and respiration-adjusted LF did not correlate with RDI during wakefulness but were positively correlated during NREM. Their findings concurred with previous literature on how SDB and sleep stage impact cardiac autonomic control respectively but suggest an intrinsic gender-dependent difference in this association. Finally, Jurysta et al. used a coherence analysis between normalized HF and normalized delta EEG power to study how cardiac vagal influence and sleep EEG were affected by OSAHS and its severity.31 Using 3 groups of uncomplicated subjects (normal control, moderate-to-severe apnea, severe apnea), they found that HFnu, consistent with prior work, was larger during NREM while LFnu predominated both REM sleep and wake stages. They also reported that coherence and gain between normalized HF and delta EEG decreased from normal control to severe apnea patients. Results supported an altered interaction (coherence correlating negatively with obstructive AHI) between cardiac autonomic control and delta sleep EEG in OSA patients.

Chronic HRV effect of sleep-disordered breathing (Table 5) The effect of OSA on daytime HRV was studied by Hilton et al. in uncomplicated patients vs. matched healthy controls.25 HRV was calculated on 25-min of data during wake to assess awake ANS function. HF% was significantly (p < 0.03) reduced in OSA compared to controls. In the previously cited study,26 where daytime peripheral SNS activity was significantly increased in uncomplicated moderate-to-severe OSA patients compared with unmatched control subjects, a significant daytime increase in LF/HF and LFnu, along with a decrease in HFnu and RR intervals were seen in the moderateesevere OSA patients. Ueno et al. studied the effect of OSA on HRV in stable congestive heart failure (CHF) outpatients with left ventricular ejection fraction (LVEF) <¼ 45%.27 All patients had overnight PSGs and were divided into sleep apnea and non-sleep apnea by AHI > 10. Respiration, HR and blood pressure were measured for 10 min while awake (22:00 h), during S2 sleep, awake (6:00 h) and awake (10:00 h). In addition, MSNA was measured at 10:00 h. The study demonstrated that patients with CHF and OSA had reduced cardiac autonomic modulation but increased SNS activity across the 24-h period. OSA appears to have an adverse effect on HRV in children as well. Liao et al. compared HRV measured during PSG in a population-based cohort of 700 children enhanced by a clinical sample of children with severe SDB. Among the children with at least moderate SDB (AHI  5), there were marked reductions in HF power and a resultant marked increase in the LF/HF ratio consistent with decreased parasympathetic control of HR. Although the authors suggest that LF/HF results are consistent with increased sympathetic activity, these results are entirely driven by the decrease in HF because LF was not different between groups.28 Kwok et al. investigated supine HRV during a 1-h daytime ECG in snoring children who had undergone overnight PSGs. PNN50, an HRV measure of PNS function was decreased among those children who had been diagnosed with OSA (AHI > 1.5/h) but no other significant differences were found in the time or frequency domain measures, including in other measures of parasympathetic

Practice points OSA is associated with abnormal cardiac autonomic control that continues during the daytime.

Research agenda Study whether impact of SDB on HRV is associated with individual risk of adverse cardiac outcomes.

HRV effect of periodic leg movements (Table 6) Several investigators have examined HRV in association with periodic limb movements (PLMs). Sforza et al. analyzed the effects of PLMs on HRV during non-REM sleep in patients with PLMs but without other major sleep disorders, neuromuscular or cardiac diseases.32 Standard HRV was calculated during selected 10-min periods. Results were interpreted as showing that the occurrence of PLMS was associated with an increase of SNS activity without significant changes in PNS activity. Winkelman found a significant rise (10 beats after onset vs. before) in HR following the onset of leg movements during sleep, in comparison with waking leg movements.33 Such rise in HR was 10e40% higher (not statistically significant) when leg movements were associated with American Academy of Sleep Medicine (AASM) defined arousals than without. Guggisberg et al. compared the impact on HRV and EEG spectra of PLMs, isolated leg movements and respiratory-related leg movements during sleep.34 SNS activation, measured by increased LF power, was reported to be greater for PLMs than other types of leg movements, suggesting “a primary role of the SNS in the generation of PLMS”.

Table 5 Chronic HRV effect of sleep-disordered breathing. Subjects

Methods

HRV measure

Data collection/analysis

Hilton et al., 200126

N ¼ 15 M SAHS (age: 47.2  1.8) and N ¼ 14 M controls (age: 43.8  2.5)

LF, HF, total power, %LF and %HF (over total power), LF/HF ratio

Narkiewicz et al., 199827

N ¼ 16 normal, 12M/4F, (age: 41  9, mean BMI 33), N ¼ 18 mild OSA (14M/4F, age 45  10, mean BMI 32), N ¼ 15 moderateesevere OSA 12(M/3F, age 40  9, mean BMI 36) N ¼ 25 Congestive Heart Failure patients divided into with SA (n ¼ 17, age 57  2, 13F/4M, mean BMI 26) and without SA (n ¼ 8, age 58  2, 2F/6M, mean BMI 26) N ¼ 700, mean (SD) age 112 (21) months, 343M/357F, 73% no SDB, 25.8% AHI 1-5, 1.2% AHI > 5. 43 clinically diagnosed SDB N ¼ 91, 74M/17F snoring children

30-min ECG supine and standing in the morning after PSG 10-min ECG daytime awake supine

25-min windowed into 5-min segments, Hanning-windowed Welch spectral averaging (2.5 min with 50% overlap) Autoregressive parametric spectral analysis

PSG and ECG in sleep and next morning

RR Variance, LF, HF, LFnu and HFnu normalized by LF þ HF, LF/HF ratio

9 h PSG to define SDB. 1st available 5 min RR intervals Continuous 1-h ECG daytime awake PSG records

SDNN, rMSSD, HR, LF and HF power, LF/HF ratio SDNN, SDANN, rMSSD, pNN50, VLF, LF, HF, ‘LFnu, HFnu, LF/HF MeanLF, HF, LFra and HFra were ‘respiration adjusted’ LF and HF respectively Mean LF, HF, LFnu and HFnu

Ueno et al., 200928

Liao et al., 201029 Kwok et al., 201130 Wang et al., 200831

N ¼ 288 subgroup of SHHS study (age: 62.7  6.7, 221M/67F, mean BMI 30.3, RDI: 0.3e85)

Jurysta et al., 200632

N ¼ 12 M controls (age: 43  6, mean BMI 26), N ¼ 12M moderateesevere SAHS (age: 44  4, mean BMI 27), N ¼ 12M severe SAHS (age: 44  6, mean BMI 32)

AHI: apnea hypopnea index BMI: body mass index HF: high frequency power HR: heart rate HFnu: [HF/(TP - VLF)] for the measured period (5-min or less) HFra: respiration-adjusted HF power OSA: obstructive sleep apnea pNN50: percent of NN intervals> 50 ms different from previous for period of interest PSG: polysomnogram RDI: respiratory disturbance index rMSSD: root mean square of successive differences of NN intervals for period of interest SD: standard deviation SDB: sleep-disordered breathing SAHS: sleep apnea hypopnea syndrome SDNN: standard deviation of NN intervals for period of interest SDANN: standard deviation of AVNN for 5-min intervals for period of interest VLF: very low frequency power

PSG

Mean LF, HF, LFnu and HFnu

Selected 10-min of stable breathing for each stage or condition, autoregressive algorithm-based spectral analysis Measured from full night of recording 1 h during daytime Consecutive 5-min segment over record, Welch method with Hanning window Spectral analysis using sliding window of 120 s with update every 20 s to correlate EEG power

P.K. Stein, Y. Pu / Sleep Medicine Reviews 16 (2012) 47e66

First author/year

55

56

P.K. Stein, Y. Pu / Sleep Medicine Reviews 16 (2012) 47e66

Table 6 HRV effect of PLMs or RLS (Restless Legs Syndrome). First author/year

Subjects

Methods

HRV measure

Data collection/analysis

Sforza et al., 200533

N ¼ 14 insomnia with RLS suspected (age: 50  3, 5F/9M, mean BMI 25)

Holter ECG and PSG

NN50, pNN50, SDANN, SDNN, SDNNIDX, rMSSD, TIN, NVLF, LF, HF, LFnu, HFnu

Winkelman, 199934

N ¼ 8 patients with PLMs (age: 49  19, 2F/6M)

PSG

Instantaneous HR

Guggisberg et al., 200735

N ¼ 8 PLMs (age: 51  5, 3F/5M, BMI 25), N ¼ 7 OSA (age: 57  4, 0F/7M, BMI 35), and N ¼ 9 normal (age: 40  5, 5F/4M, BMI 26) N ¼ 20 children referred for PSG, 10 with and 10 without PLMs (age: 7e12)

PSG

LF, HF. LF/HF ratio

Selected 10-min segment for each stage, FFT-based and Wavelet Transform-based power spectral analysis were used to obtain frequency domain HRV parameters Window of 20 heart beats centered on onset of each of 796 (out of 1885) leg movements Wavelet transform over 10 s to 10 s centered around each leg movement

Overnight PSG

LF/HF ratio

Walter, et al., 200936

PLM þ compared with PLM- periods

BMI: body mass index ECG: electrocardiogram FFT: fast Fourier transform HF: high frequency power HFnu: [HF/(TP - VLF)] for the measured period (5-min or less) LF: low frequency power LFnu: [LF/(TP-VLF)] for the measured period (5-min or less) rMSSD: root mean square of successive NN differences NN50: number of NN intervals different by >50 ms from prior NN OSA: obstructive sleep apnea PSG: polysomnogram RLS: restless legs syndrome PLM: periodic leg movement pNN50: percent of NN intervals> 50 ms different from previous for period of interest SDNN: standard deviation of NN intervals for period of interest SDNNIDX: average of 5-min standard deviations of NN intervals for period of interest TINN: triangular index of NN intervals VLF: very low frequency power

All of these results are based on the fact that during PLMs CVHR is observed in either the LF or VLF range, depending on the frequency of the leg movements themselves. Again, as described in the discussion of HRV during SDB, the application of this to power spectral HRVderived autonomic function is a bit circular. Changes in HR mean that there is a change in autonomic activity; either decreased PNS or increased SNS activity, usually both. The fact that HR is increasing or decreasing tells us that without making any calculations. Increased VLF or LF power mathematically captures the presence of these underlying leg-movement generated “HR arousals” over specific ranges of frequencies. If they are at a frequency of every 25 s, they will increase VLF power, but VLF power, technically. is not modulated by the SNS since it is unaffected by beta-blockade, so the hypothesis of increased sympathetic activity would not be confirmed.6 If mixed, they will show up as increased power in both bands. However, the relationship of LF and VLF to underlying autonomic activity applies to intrinsic autonomic heart rhythms, whereas the underlying frequency of the HR changes caused by PLMs is event based. Measurement of HF during a period of unstable HRs is also problematic. So caution needs to be applied in interpreting the above results. Walter et al. in a study of PLMs in children took this problem into account.35 They did not attempt to interpret LF power “due to the likelihood that leg movements had a direct effect on the lower frequencies” and concluded that the rapid cardiac acceleration seen in association with PLMs was mediated by vagal withdrawal.

Practice points PLMS are associated with CVHR, although generally of higher frequency and shorter duration than SDB.

Research agenda Validate whether the specific HR patterns associated with PLMS provide complementary information to standard PSG scoring.

Using HRV to screen for sleep-disordered breathing (Table 7) Guilleminault et al. first coined the term cyclic variation of the heart rate and demonstrated the feasibility of using it to screen for moderate and severe OSA.36 In the same study, they concluded that CVHR was regulated mainly by vagal instead of SNS activity. This conclusion was based on the observations that CVHR is abolished by IV injection of atropine (vagal pressor) but not by propranolol (non-selective beta blocker). This discovery was not followed up until more recently when advances in computational power made screening of ECGs for CVHR more feasible. As previously discussed, CVHR reflecting tachycardia at the termination of the respiratory event with a return to baseline afterwards characterizes sleep-disordered breathing and also PLMs and arousals.20,36e38 When severe OSA is present with preserved autonomic functioning, there is a characteristic severe and continuous CVHR that is pathomnemonic for this disorder (Fig. 2). It can be reliably detected by almost any method. At the same time, in the presence of severe autonomic dysfunction, CVHR will be virtually absent whether severe SDB is present or not. Anecdotally, the high likelihood of SDB among these patients might suggest the need for PSG even though no CVHR is seen. What is more challenging is the spectrum in between; patients with mildly depressed autonomic function in whom lower level SDB-associated CVHR might resemble that of patients with preserved autonomic function

Table 7 Using HRV to screen for sleep-disordered breathing. Subjects

Methods

HRV measure

Data collection/analysis

Guilleminault et al., 198437

N ¼ 400 sleep apnea patients (16F, age 25e68, 384M, age 19e71)

24-h Holter synched with PSG

RR interval

Tachogram display of beatebeat changes in RR

N ¼ 50 moderateesevere OSA (age: 42  8, mean BMI 36), N ¼ 15 mild OSA (age: 43  8, mean BMI 31), N ¼ 27 normal (age: 42  6, mean BMI 31) N ¼ 30 patients with RDI > 15 (age: 43  11, 4F/26M) and N ¼ 30 with RDI < 15 (age: 35  12, 11F/19M) PhysioNet public Sleep Apnea Test Database (N ¼ 30)

PSG

AHI estimated by a matrix of HRV parameters and ECG-derived respiration

1-min analysis window over the record

Selected 5-h of ECG during sleep

Mean, SDNN, rMSSD, LF and HF, Sub-band LF1 (0.02e0.04 Hz), Sub-band LF2 (0.04e0.06 Hz), LF/HF ratio, LFnu Mean and SD of HR amplitude and frequency, fraction of time that amplitude and frequency within their thresholds

1024 point FFT over HR data resampled at 4Hz

Vazir et al., 200642

N ¼ 33M with CHF (age: 62  12, mean BMI 29)

PSG

%VLFI

Tateishi et al., 200343

N ¼ 13 CHF with SDB (age: 68  4, 5F/8M)

24-h Holter

Shalaby et al., 200744

N ¼ 14 pacemaker patients with sinus node disease (age: 66  12, 1F/13M) and SDB N ¼ 28 consecutive patients (age: 58  12, 5F/23M) referred to sleep lab as Group I derivation set and N ¼ 35 (age: 52  11, 4F/31M) as Group II validation set N ¼ 147 OSA suspects (age: 54  11, 46F/101M)

PSG

Mean, correlation variance of NN, SDNN, HF, LF, VLF., %VLF, %LF, and %HF by total power. Apnea band, power ratio of apnea band by power over 0.03e0.04 Hz RR interval

Holter synched with PSG

VLF (0.01e0.05 beat-1), %VLFI as % of VLF over total power (0.01e0.5 beat-1)

Holter ECG synched with PSG

Sum of squares of the Wavelet coefficients at 3 groups of levels that approximate to VLF, LF, and HF by FFT spectral analysis Mean, SDNN, SDANN, SDNNIDX, RDSSD

Penzel et al., 200238 Heneghan et al., 200839

Chang, 200940

Mietus et al., 200041

Roche et al., 2004

46

Roche et al., 200347

Roche et al., 199948

Hayano et al., 201050

N ¼ 91 OSA suspects (55  11, F20/M71) as derivation set and N ¼ 52 (age: 53  12, 18/34M) as validation set Optimized with 63 sleep studies in a training set. Confirmed with 70 sleep studies from Physionet Apnea-ECG database. Applied to PSGs in 862 consecutive subjects referred for diagnostic sleep study.

PSG

PSG and 24-h Holter PSG

Autocorrelated wave detection with adaptive threshold (ACAT).

Hilbert Transform on processed (moving-average, bandpass filtering) RR to obtain instantaneous amplitude and frequency, 5-min analysis window with 1-min increment User-selected 16-min throughout record, FFT-based spectral analysis (commercial software) Commercial software without disclosure of analysis details Tachogram display of beatebeat changes of RR, criteria for CVHR Analysis software to obtain individual power spectral and then smoothed by 50 consecutive values Wavelet analysis on sliding window of 512 RRs 5-min consecutive segments over record, StrataScan 563 (DelMar) HRV module

P.K. Stein, Y. Pu / Sleep Medicine Reviews 16 (2012) 47e66

First author/year

AHI: apnea hypopnea index FFT: fast Fourier transform HF: high frequency power HR: heart rate HRV: heart rate variability LF: low frequency power NN: interval between adjacent normal heart beats PSG: polysomnogram rMSSD: root mean square of successive differences of NNs SD: standard deviation SDANN: standard deviation of AVNN for 5-min intervals for period of interest SDNNIDX: average of 5-min standard deviations of NN intervals for period of interest Synced: synchronized VLF: very low frequency power

57

58

P.K. Stein, Y. Pu / Sleep Medicine Reviews 16 (2012) 47e66

who do not have OSA. Also challenging is the presence of intermittent events, perhaps specific to certain sleep stages (e.g., REMonly apnea) or a mixture of both clinical and subclinical events. Thus, for example, a hypopnea that fulfills the 50% loss of airflow criterion is just as likely to result in CVHR as one that has only a 40% reduction in airflow, but one would be scored and the other would not. Significant events during periods scored as wake are also often seen but not counted. Another way of looking at this, however, would be to consider the possibility that quantification of CVHR could add complementary information to current PSG scoring. There are several reports of successful utilization of continuous ECGs to screen for OSA. Specifically, OSA-induced CVHR has been defined with a characteristic frequency ranging from 0.01 to 0.06 Hz and amplitudes from 6 beats/min to >30 beats/min.36,39,40 Although obstructive hypopneas tend to have a higher frequency and lower amplitude of HR changes, there is considerable overlap with CVHR from obstructive apneas. Similarly, CVHR from PLMs tends to have the highest frequency but there is considerable overlap with CVHR from respiratory events. Others have used CVHR to screen for central sleep apnea (CSA) or combined OSA and CSA. CSA is also associated with CVHR, although the frequency of oscillations can be slightly lower and the HR arousal that terminates each episode is often slower and of lower amplitude. Sometimes CSA can be identified by the combination of CVHR and the complete absence of respiratory sinus arrhythmia HR patterns in the period before the HR arousal. When CheyneeStokes respiration (CSR) is present, the waxing and waning of amplitude or frequency can often be seen by examining RSA patterns on an HR tachogram. Vazir et al. applied a VLF power spectral density index (VLFI) to identify SDB, including OSA and CSR/CSA, based on PSG records from 33 CHF patients.41 Software automatically calculated the percentage of VLF power based on 20-min segments from 0:00h to 6:00h to identify the SDB-associated VLF rhythm. Defining “true patients” with AHI > 20 and “positive detect” with VLFI > 2.23%, this method yielded a sensitivity of 85%, specificity of 65%, positive predictive value of 61% and negative predictive value of 87% for SDB. Tateishi et al. obtained 24-h ambulatory recordings, including ECG, respiration, SpO2 and body position, from CHF patients.42 AHI was determined based on the ambulatory record and AASM task force guidelines. A power ratio (%) of apnea was calculated as the power in an apnea band (0.01e0.03 Hz) over that of 0.03e0.04 Hz. They recommended that power  80% was highly associated with presence of SDB. It must be noted however, that CVHR is a marker for SDB, not the cause. Thus, when Shalaby et al. observed CVHR among pacemaker patients, they discovered that with an increase in atrial overdrive pacing rate, the amplitude of the CVHR (mean peak to trough of the cyclic variation) was reduced or even eliminated.43 However, the reduction in CVHR did not correlate with AHI improvement in patients. In 2000, PhysioNet and Computers in Cardiology initiated an international challenge to stimulate algorithm research on screening for OSA using ECG alone.44 A database of 70 ECG records and their concurrent clinical sleep diagnoses, selected from patients with uncomplicated OSA and subjects without SDB, was provided. Twelve participants achieved detection accuracies of 90% using various algorithms, including HR-only methods and a combination of HR methods and ECG morphologies.37 Several research groups have studied the feasibility of such ECGbased SDB screening methods in clinical settings, including different sleep apnea types or different patient populations. 20,38,45e47 Roche et al. used the VLF index (%VLFI e percentage of VLF power over the total power spectral density) from HRV to quantify the abnormal power spectral distribution of HRV due to

CVHR.45e47 Stein et al. adopted the HR tachogram concept to display instantaneous HR changes during sleep, relying on human visual detection of the presence of CVHR.20 Combining HRV analysis and ECG-derived respiration technology, Heneghan et al. developed an automated algorithm, which derives a matrix of features from a single lead ECG and uses classifier models for detection and achieved a satisfactory clinical performance through the training database.38 DelMar Reynolds Medical obtained FDA 510(k) clearance for intended use of ECG-based algorithm for screening adults for the probability that they suffer from OSA, mixed apnea or hypopnea, in order to evaluate the necessity for a PSG examination.48 Most recently, Hayano et al. conducted a large-scaled clinical study to further evaluate the efficiency of CVHR-based automated screening of OSA.49 An algorithm of autocorrelated wave detection with adaptive threshold (ACAT) was applied to the interbeat intervals from ECGs extracted from PSG in 862 consecutive apnea suspects and achieved 83% sensitivity and 88% specificity when “true” OSA was defined as AHI  15/h. An interesting finding of this study was that detection performance was not affected by older age or cardiac autonomic dysfunction. Use of the ECG alone to screen for sleep apnea has the potential to identify a large proportion of at-risk patients. Some sort of screening algorithm should be applied to all clinically obtained Holter recordings, and this is an area of active research for many companies. Use of additional channels that are affected by OSA, e.g., oximetry, finger pulse volume, or surrogates for respiration should increase diagnostic sensitivity and specificity, but that is beyond the scope of this review. At present, there is not one agreed-upon method for using the ECG alone to screen for risk of having sleep apnea, in part because it depends on the patient population and the threshold of AHI that is being detected. It also depends on the resources available to the clinician. As previously mentioned, there is a large set of OSA patients with clear, one might even say, spectacular, CVHR. They can be readily identified by any method including presence of a VLF peak on Holter recordings, although when available, the beat-to-beat heart rate tachogram provides the clearest information. There is another set of patients with autonomic dysfunction or atrial fibrillation where heart rate is unaffected by underlying OSA and they cannot be identified using ECG analysis. This too can be seen on the HR tachogram as a lack of HR changes, consistent with severe autonomic dysfunction. These patients probably should be screened for SDB. Finally, there is the smaller subset, possibly with some degree of autonomic dysfunction, where HR changes are more equivocal and who probably account for the differences in the specificity and sensitivity of the different algorithms. Tachogram analysis is useful in these cases as well, but the verdict is not yet in on the ideal screening algorithm.

Practice points Solid clinical evidence has demonstrated that: 1) A sufficient degree of overnight CVHR from ambulatory monitoring indicates high risk for moderateesevere OSA suggesting that ECG-based screening for referral to PSG is feasible. 2) When screening is negative, attention must be paid to underline clinical conditions that may fundamentally alter autonomic and/or sinus node function hence HR response to respiratory events in order to rule out a false negative.

P.K. Stein, Y. Pu / Sleep Medicine Reviews 16 (2012) 47e66

Research agenda 1) Need to estimate sleep period for ambulatory recordings (especially multi-day telemetry). 2) Continuing validation against PSG. 3) Further evaluate impact of co-morbidities or conditions, especially those that fundamentally alter autonomic function or cardiac rhythm, hence reduce HR response to respiratory disorders, on detection sensitivity. 4) Further evaluate impact of confounding factors such as CSR or PLMs. 5) Evaluate impact of different ECG devices or different ECG lead configurations when incorporating ECGderived respiration technology for screening. 6) Further evaluate clinical performance when additional sensor(s) are incorporated, e.g., oximetry to enhance power of detecting SDB.

HRV effects of insomnia and sleep deprivation (Table 8) Aside from their sleep complaints, few physiologic measures consistently differentiate between patients with insomnia and those with normal sleep.50 The concept that hyperarousal processes “play a key role in the pathophysiology of primary insomnia” makes it intuitively appealing that HRV might capture some differences in underlying autonomic state during insomnia.51 HRV has been compared between insomniacs and good sleepers in a small number of studies. HRV was calculated from a 5-min segment selected from 30-min before lights out, each sleep stage, and 10-min standing after wake-up the next morning and compared in age-, sex-, and weight-matched insomniac patients and controls by Bonnet et al.52 A reduced HRV in insomniacs compared with control was seen across all sleep stages. Fang et al., however, found no difference in resting HRV (5min ECG collected during afternoon supine resting with respiration controlled at 12e15 breaths per min) in insomniacs compared with normal sleepers.53 In another study by Jurysta et al., male chronic primary insomnia patients were matched with healthy men. Although HRV analyzed for the first 3 NREMeREM cycles did not show any difference, the coherence between normalized HF power and delta EEG was decreased significantly in insomniacs (p < 0.5) in comparison to controls.54 Short sleep duration has been shown to be a risk factor for cardiovascular mortality in population studies but it has not yet been convincingly demonstrated that the autonomic effects of short sleep mediate this relationship.55 Barnett et al. assessed individuals with no symptoms of clinical sleep disorders and showed that people with <6 h sleep had higher HRs (during awake, eye-closed and open resting condition) compared with people with >6 h of sleep.56 Studies of people who are not getting enough sleep provide insights into potential mechanisms for the deleterious effect of short sleep duration. Healthy college students were assessed before (baseline) and after chronically sleep deprivation for 4 weeks during finals by Takase et al.57 Results showed a decrease in HRV and erythrocyte-Mg associated with an increase in norepinephrine via intracellular measurements. In another study from the same group, this time of sleep-deprived drivers by Michail et al., LF/HF and LF were lower in subjects with driving errors (related to sleepiness) than in those with no driving errors.58 Also, in a small study by Chung et al., long-term nighttime nurses were found to have higher LF and LF/HF during

59

non-REM sleep compared with regular shift nurses, supporting the hypothesis that disturbed sleep affects cardiac SNS regulation.59 Compared with healthy subjects, Nielsen et al. found that patients with frequent nightmares showed significantly higher LF and LF/HF, and lower HF through analysis over 3-min noise free periods selected from REM sleep, Stage 2 NREM sleep and presleep wakefulness.60 There have been a number of controlled laboratory-based investigations of the autonomic effect of acute sleep deprivation. Results have been inconsistent. One study by Sauvet et al. found that after 32-h of acute sleep deprivation, healthy males had significantly increased systolic blood pressure, HR and normalized LF, along with an earlier (after 29-h) occurrence of vascular dysfunction indicated by biomarkers.61 Another study by Sayk et al. showed that, although one night slow-wave sleep deprivation significantly affected the magnitude of blood pressure dipping during sleep, it did not impact vascular, cardiac or blood pressure function during the day time.62 Consistent with this, Pagani et al. reported that among 24 healthy subjects one night of sleep deprivation, although it induced tiredness and reduction in performance, did not affect hemodynamic parameters or HRV significantly.63 However, Zhong et al. studied 36-h sleep deprivation effects on cardiovascular autonomic modulation in normal subjects.64 LF and LF/HF increased while HF decreased as did baroreflex sensitivity which is a measure of the ability of the HR to respond to changes in blood pressure. Mellman et al. found LF/HF higher in REM sleep among patients with positive symptoms for posttraumatic stress disorder (whose sleep is presumably more disturbed) than among patients without.65 Finally, Lombardi et al. studied whether specific cardiac autonomic modulation during SDB could be related to excessive daytime sleepiness (EDS).66 Subjects underwent nocturnal PSG with a multiple sleep latency test administered the following morning. Spectral HRV was averaged over the whole recording as well as over each sleep phase. Compared with patients without EDS, patients with EDS demonstrated higher LF/HF and lower baroreflex sensitivity throughout different sleep stages, suggesting EDS accompanied by a deranged cardiac autonomic control during sleep.

Practice points 1) Mixed evidence suggests that poor sleep and sleep deprivation can adversely affect HRV. 2) No test yet to use HRV to identify insomniacs.

Research agenda 1) Continuing investigation of the possible bi-directional relationship between sleep quality and ANS function. 2) Further understanding of whether relationships between sleep quality and HRV and HR can measure the autonomic effect of poor sleep. 3) Understand whether the autonomic effect of poor sleep has prognostic value in identifying patients with vulnerability to adverse outcome.

60

Table 8 HRV effect of insomnia and sleep deprivation. Subjects

Methods

HRV measure

Data collection/analysis

Bonnet et al., 199853

12 pairs age-, sex- and weight-matched normal and insomniacs N ¼ 18 insomniacs (age: 34  15, 12F/6M) and N ¼ 21 normal (age: 28  9,14F/7M)

PSG

Mean and SDNN LFnu, HFnu.

5-min ECG in supine awake at controlled breathing rate of 12e15 breaths/min in the afternoon PSG

Natural logarithm of LF, HF, and total power. LF/HF ratio

Selected 5-min for each stage, BMDP T1 Spectral Analysis Program FFT-based frequency analysis (Nevrokard HRV analysis Software package)

Mean, LF, HF, LFnu and Hfnu

Spectral analysis of sliding 120 s window updated every 20 s over the first 3 2-REM cycles

3-min conscious rest ECG/EEG 1-h ambula-tory ECG between 16:00 h and 18:00 h

Mean, SDNN, rMSSD, NN50, VLF, LF, HF, LF/HF Mean, SDNN, SDANN, SDNN index, RMSSD, pNN50, LF, HF, total power

Niskanen Software FFT-based Welch Periodogram

Max of 1 h ambulatory EEG and ECG during driving Ambulatory PSG

Mean normalized LF, normalized HF, and LF/HF ratio LF/HF ratio during REM and NREM sleep Mean, SDNN, pNN50, VLF, LF, HF, and LFnu and HFnu, LF/HF ratio

Fang et al., 200854

Jurysta et al., 200955

14 pairs of male chronic primary insomnias (age: 42  12, mean BMI 26) and male healthy controls (age: 41  10, Mean BMI 24) N: 338 (188F/150M), Age: 41.4  13.2

Barnett et al., 200857 Takase et al., 2004

58

N ¼ 30 healthy male college students (age: 22  2)

Michail et al., 200859

N ¼ 21 drivers with Mean age 26.5 (1F/20M)

Chung et al., 200960

N ¼ 10 regular and N ¼ 10 night shift nurses.

Nielsen et al., 2010

61

N ¼ 16 patients (age: 26.1  8.7) with frequent nightmares and N ¼ 11 controls (age: 27.1  5.6) N ¼ 12 healthy M (age: 29  3, mean BMI 23 kg/m2) N ¼ 11 healthy subjects (age: 24.5  1.6, 6F/5M, mean BMI 22) N ¼ 24 young healthy subjects (age 27e45, 12F/12M, mean BMI 22)

Sauvet et al., 201062 Sayk et al., 201063 Pagani et al., 2009

64

Zhong et al., 200565

N ¼ 21 normal (age: 26.1  4.4, 2F/19M)

Mellman et al., 200466 Lombardi et al., 2008

67

N ¼ 19 suspects of posttraumatic stress disorder (age: 37  11, 6F/113M) N ¼ 53 patients with SDB divided into EDS (n ¼ 39, age: 49  11, mean BMI 28k) and non-EDS (n ¼ 14, age: 46  6, mean BMI 32)

PSG

10-min ECG in supine every 3-h between 0900 h and 1800h PSG 10-min ECG in supine after 10-min rest 10-min ECG in awake sitting and supine PSG Finapres continuous blood pressure monitoring along with PSG

BMI: body mass index SDB: sleep-disordered breathing PSG: polysomnogram ECG: electrocardiogram SDNN: standard deviation of NN intervals for period of interest LFnu: [LF/(TP-VLF)] for the measured period (5-min or less) HFnu: [HF/(TP - VLF)] for the measured period (5-min or less) LF: low frequency power HF: high frequency power rMSSD: root mean square of successive differences of NN intervals for period of interest VLF: very low frequency power SDANN: standard deviation of AVNN for 5-min intervals for period of interest NREM: non-rapid eye movement REM: rapid eye movement FFT: fast Fourier transform

5-min segments averaged for time-domain HRV, consecutive non-overlapping 2-min segment for FFT (commercial software) 5-min segment with 75% overlap, biosig software

3-min selected for each stage, Cardiolab software

Mean, LF, HF, LFnu and HFnu

5-min artifact-free segment

Mean and HRV parameters by task force standards Mean, Variance, LF (w0.1 Hz), HF (w0.25 Hz), normalized LF and HF, LF/HF ratio VLF, LF, HF, LFnu and HFnu, LF/HF ratio. Mean, LF, HF, LF/HF ratio

Selected 5-min for each condition

Mean, LF, HF, LF/HF ratio

Autoregressive based spectral analysis

Last 5-min artifact-free, spectral analysis 5-min selected from each sleep stage, Autoregression based spectral analysis (MatLab) HR spectrogram applied on each 110 s-segment and averaged for each sleep stage

P.K. Stein, Y. Pu / Sleep Medicine Reviews 16 (2012) 47e66

First author/year

P.K. Stein, Y. Pu / Sleep Medicine Reviews 16 (2012) 47e66

Using HRV to monitor the effect of CPAP (Table 9) Effective CPAP abolishes CVHR, while at the same time markedly reducing the excessive SNS activity due to respiratory events. Therefore, use of CPAP would be expected to have significant effects on both nighttime and daytime autonomic function. Direct measurement of MSNA has confirmed a reduction in SNS activity associated with successful CPAP.22 However, MSNA is not clinically useful for assessment of changes in autonomic function associated with OSA treatment. HRV provides a noninvasive measurement of cardiac autonomic function, so it is potentially usable to monitor the autonomic benefits, if any, of treatment for SDB.67 Roche et al. investigated the effect of 3-months of CPAP on ANS function among uncomplicated OSA patients free from diseases or medications known to impact HRV.68 Time-domain HRV was calculated on 24-h RR intervals, while frequency domain HRV was calculated on pre-defined nighttime and daytime periods. Results showed that effective CPAP treatment was associated with significant decrease in LF power, HF power and the LF/HF ratio. Decreases in the LF/HF ratio persisted into daytime suggesting that treatment of OSA by CPAP had autonomic effects that persisted after sleep. Most recently, Shiina et al. studied the impact of 3-month use of CPAP among N ¼ 50 consecutive eligible OSA patients. Using HRV based on 10-min conscious supine recording, they observed a significant decrease in the LF/HF ratio, HR and the brachial-ankle pulse wave velocity (baPWV), an indicator of arterial stiffness. Their analysis demonstrated “a significant correlation between the change in the LF/HF ratio and that in baPWV, independent of the changes in the mean blood pressure, plasma c-reactive protein (CRP) level and heart rate”. Results suggest that CPAP’s improvement of sympathovagal balance may be directly related to improvement of arterial stiffness.69 To determine the effect of CPAP on autonomic function in CHF patients, Gilman et al. randomized stable patients (LVEF  45% with ischemic or non-ischemic dilated cardiomyopathy) with obstructive apnea (AHI  20), to a control group and a CPAPtreatment group.70 Morning HRV was compared at baseline and after 1-month of no treatment for controls and 1-month of effective CPAP treatment in the CPAP group. Compared with the control group, the CPAP-treatment group had significantly reduced arousals during sleep (p < 0.001), significantly increased HF and HF% (p ¼ 0.002; p ¼ 0.021) and decreased LF/HF (p ¼ 0.045). CPAP also improved LVEF by a significant 8%. Results suggested CPAP improved PNS modulation of HR in CHF patients with OSA, improved LV function and reduced SNS activation. Their analysis of coherence between respiration and HF in HRV indicated an increased respiratory influence on HF power. Results concurred with prior reports on uncomplicated OSA patients that also suggest that CPAP improves the gain of the transfer function relating changes in RR intervals to changes in lung volume both in sleep and while awake.71,72 In a recent study, HRV was used to compare the cardiovascular improvement in OSA patients by auto-titrating positive airway pressure (APAP) vs. conventional CPAP.73 Newly diagnosed symptomatic OSA patients (AHI  15 with ESS score  10) were enrolled for a one-night baseline PSG and another night of manual CPAP titration. One week later, all patients received one night of APAP with PSG, and another week later CPAP with PSG. Time-domain and frequency-domain HRV were calculated for 5-min segments of qualifying ECG data from each PSG night. Results showed that the time-domain parameters (rMSSD or SDANN) were not different among the 3 study nights, while frequency domain parameters like normalized LF and HF and the LF/ratio were improved by CPAP but not by APAP. Improvement with CPAP was most pronounced during stage 2 sleep. Given previous reports that APAP was not as effective

61

as CPAP in reducing several cardiovascular risk factors, the authors suggested HRV differences observed in this study may provide an explanation74,75 A caveat here, and in other research applications, is the importance of making calculations of HRV from true sinus rather than erratic rhythm.

Practice points 1) Effective CPAP is consistently associated with changes in both nighttime and daytime HRV. 2) HRV may provide supplemental information on the effectiveness of OSA treatment.

Research agenda 1) Further study of HRV and CPAP in different comorbidities. 2) Understand if the observed different impact on HRV by CPAP vs. APAP is real, and whether any such differences are due to physiological causes or if APAP algorithm enhancement is needed. 3) Show whether CVHR and HR patterns during “adequate” OSA treatment, can indicate complex SDB.

Cardiopulmonary coupling: a novel application of HRV to sleep (Table 10) Thomas and his collaborators at Harvard proposed a new approach to measure HRV during sleep called the cardio-pulmonary coupling (CPC) algorithm.76 This algorithm uses shortwindowed analyses of the cross-spectrum of 2 ECG-derived signals: RR intervals and a surrogate respiration signal from R-peak amplitude changes or estimated cardiac vector axial changes, assuming both are modulated by respiration. The CPC cross-spectrum, the product of the Fourier transform amplitudes of the 2 derived signals over pre-specified frequencies, is normalized to the power over 0e1 Hz and becomes a function of time as the analysis window moves. The CPC algorithm determines coupling or nocoupling based on amount of integrated cross-spectral power over frequency bands designated as the VLF (0e0.01 Hz), LF (0.01e0.1 Hz) and HF (0.1e1 Hz) bands. Thomas et al. studied the utility of the CPC algorithm in a series of publications describing potential clinical applications including sleep stability assessment (EEG cyclic alternating pattern), and sleep apnea detection and classification.76e79 The CPC method has shown that clear and strong coupling in the high frequency range is concurrent with known normal respiration in non-REM sleep and stable sleep stages. In addition, coupling in the low frequency range is concurrent with known repeated apnea events, while the absence of or weak coupling suggests a lack of such concurrencies. Furthermore, CPC analysis demonstrated that the occurrence of an elevated low frequency coupling (cross-spectrum power concentrated over a narrowed frequency band) could differentiate central from obstructive sleep apneas, and the degree of such elevation could further classify central from mixed or complex apneas.

62

Table 9 HRV to monitor the effect of CPAP. Subjects

Methods

HRV measure

Data collection/analysis

Roche et al., 199969

N ¼ 14 OSAS suspects (age: 61  8, 2F/12M, mean BMI)

Holter ECG synced with PSG

Mean, SDNN, SDANN, SDNN Index, RMSSD, pNN50, VLF, LF, HF, LFnu and HFnu, LF/HF ratio

Shiina et al., 201070

N ¼ 50 consecutive OSA with CPAP treatment (age: 54  10, 5F/45M, mean BMI: 27  3) N ¼ 19 patients with heart failure and OSA, randomized into 12 CPAPs (age: 58, 11M/1F, mean BMI) and 7 non-CPAP (age: 57, 6M/1F, mean BMI 30) N ¼ 6 normal (age: 28  1, 1F/5M) and N ¼ 7 moderateesevere OSA (age: 53  2, 0F/7M) N ¼ 13 M OSA (AHI>20, age: 46  3, mean BMI 34) with CPAP prescribed

ECG

Mean, LF, HF, LF/HF

Consecutive 5-min segments for time-domain, 2-min segments for Hanning windowed FFT, then averaged hourly 10-min record in supine position

ECG in supine w2 h after awake in the morning

Log-transformed: LF, HF, total power. LF/HF ratio, LF% and HF% as percentage of total power respectively Mean, SDNN, VLF, LF, HF, HFnur, LF/HF ratio Mean, Variability, LF, HF, GRSA, MLHR as LF/HF ratio

Gilman et al., 200871

Khoo et al., 199972 Khoo et al., 2001

73

Karasulu et al., 201074

N ¼ 28 symptomatic apnea patients (age: 46  10, 6F/22M, mean BMI 29)

Dursunoglu et al., 200575

N ¼ 12 sleep apnea suspects (age: 53  4, 3F/9M, mean BMI 32)

Patruno et al., 2007

76

N ¼ 31 symptomatic OSA patients randomized into CPAP group (n ¼ 16, age: 47  11, 3F/13M, mean BMI 35k) and APAP group (n ¼ 15, age: 47  12, mean BMI 36)

OSA: obstructive sleep apnea CPAP: continuous positive airway pressure BMI: body mass index ECG: electrocardiogram PSG: polysomnogram SDNN: standard deviation of NN intervals for period of interest SDANN: standard deviation of AVNN for 5-min intervals for period of interest rMSSD: root mean square of successive differences of NN intervals for period of interest pNN50: percent of NN intervals> 50 ms different from previous for period of interest VLF: very low frequency power LF: low frequency power HF: high frequency power LFnu: [LF/(TP-VLF)] for the measured period (5-min or less) HFnu: [HF/(TP - VLF)] for the measured period (5-min or less)

ECG synced with PSG 5-min ECG in supine and standing at controlled respiration PSG

24-h blood pressure monitoring with PSG 3 consecutive 5-min of Aterial blood monitoring after PSG

SDANN, rMSSD, log-transformed VLF, LF, and HF, LF and HF normalized by LF þ HF, LF/HF ratio HR or pulse interval HR or pulse interval

14-min analysis window, FFT

Selected stationary 5-min for each stage, Hanning windowed Welch Spectral analysis Hanning-windowed Welch spectral analysis

5-min segment free of respiratory or other events was selected for 2 stage 2 Average over daytime, nighttime for diagnostic and treatment days Average of the 3 consecutive 5-min

P.K. Stein, Y. Pu / Sleep Medicine Reviews 16 (2012) 47e66

First author/year

Table 10 Cardiopulmonary coupling: a novel application of HRV to sleep. Year

Subjects

Methods

HRV measure

Data collection/analysis

Thomas RJ et al.76

2007

PhysioNet public Sleep Apnea Database (N ¼ 70)

PSG

8.5-min sliding window with 2.1-min increment, FFT applied to 3-overlapping sub-window (4.2-min)

Thomas RJ et al.77

2009

PSG

Same as Thomas et al.,77

Thomas RJ et al.78

2010

PSG

Same as Thomas et al.,77

Same as Thomas et al.,77

Yeh GY et al.79

2008

Database (N ¼ 5247) from Sleep Heart Health Study (SHHS) N ¼ 14 fibromyalgia patients and N ¼ 13 matched controls A subset of N ¼ 18 stable heart failures randomized into Tai Chi (n ¼ 8, age: 64.2  16.2) and control (n ¼ 10, age: 54.7  11.8) with gender match

Very low frequency coupling band (0e0.01 Hz), low-frequency coupling band (0.01e0.1 Hz), high-frequency coupling band (0.1e0.4 Hz), elevated low-frequency coupling band (0.006e0.1 Hz) Same as Thomas et al.,77

24-h Holter ECG at baseline and 12-week

Mean, SDANN, SDNN, RMSSD, pNN10, pNN20, pNN30, pNN40, pNN50, ULF, LF, HF, LF/HF ratio, Cardio-pulmonary-coupling based HRV described by Thomas et al.,77

5-min segments over record, spectral analysis by Lomb periodogram,, CPC method by Thomas et al., 2007

SHHS: Sleep Heart Health Study PSG: polysomnogram ECG: electrocardiogram SDNN: standard deviation of NN intervals for period of interest SDANN: standard deviation of AVNN for 5-min intervals for period of interest rMSSD: root mean square of successive differences of NN intervals for period of interest pNN50: percent of NN intervals> 50 ms different from previous for period of interest ULF: ultra low frequency power LF: low frequency power HF: high frequency power HRV: heart rate variability FFT: fast Fourier transform CPC: cardiopulmonary coupling

P.K. Stein, Y. Pu / Sleep Medicine Reviews 16 (2012) 47e66

First author

63

64

P.K. Stein, Y. Pu / Sleep Medicine Reviews 16 (2012) 47e66

Thomas et al. argue that traditional frequency domain HRV parameters (i.e., VLF, LF, HF, or LF/HF) do not indicate whether high VLF or LF power is due to signal non-stationarity (“noise”) or a cyclic variation pattern resulting from repeated obstructive sleep apneas (“signal”). By incorporating the respiration coupling concept into HRV analysis, CPC is able to “filter” out power spectra due to non-respiratory induced HR changes, enhancing “signal” over “noise” and hence, arguably the potential diagnostic utility in the above clinical applications.

Practice points CPC is a novel method of sleep scoring that is independent of conventional methods.

Research agenda Determine if CPC provides information complementary or superior to conventional PSG for both sleep scoring and prediction of outcomes.

Use of HRV during sleep for risk stratification (Table 11) Although there have been numerous studies of the predictive value for adverse outcomes of HRV assessed over periods ranging from 24 h to a few seconds, little has been done with the enormous number of ECG signals from stored PSGs that could potentially be used to develop criteria for risk stratification using overnight HRV or potentially even HRV changes with different sleep stages. While overnight PSG can be viewed as a short 24-h Holter, there is evidence that sleep per se is a period that may be worth focusing on. Verrier et al. thoroughly reviewed the

scientific and clinical evidence for the important impact of sleep on cardiac arrhythmogenesis that led to morbidity and mortality, especially among patients with pre-existing heart disease and/or sleep-disordered breathing.80 They emphasized the importance of developing technology for autonomic tone assessment, as well as incorporation of “simultaneous measurement of ventilation and oxygen saturation to advance our understanding of causal links”. Among the few studies specifically focusing on HRV during sleep, Eguchi et al. followed 457 patients with normotensive and hypertensive type-2 diabetes and hypertensive patients without type-2 diabetes for 9.7 years.81 HR and its variability (standard deviation, rMSSD) were derived from 1 min ambulatory blood pressure monitoring performed every 30-min for 24 h. They concluded that an increase in HRV during sleep predicts incidence of cardiovascular disease in type-2 diabetes, although lack of HRV in general was not a good sign for hypertensives without diabetes. Their findings suggest that “hemodynamic change at nighttime would be very important in preventing cardiovascular events”. However, the possibility that this increased HRV during the nighttime among patients with type-2 diabetes was a consequence of severe sleep-disordered breathing rather than a lack of hemodynamic stability per se, and that possibility that diabetes could interact with OSA to potentiate CVD in this population was not considered in this study. The ability of HRV measured during sleep was also tested in an elderly subcohort of the Sleep Heart Health Study.82 HRV was measured from overnight PSGs. Even after adjustment for age and gender, decreased values for multiple measures of HRV including SDNN index, rMSSD, ln VLF, ln LF and ln HF were significantly associated with mortality.

Practice points Measurement of HRV from PSG may potentially prove useful for risk stratification.

Table 11 Using HRV during sleep for risk stratification. First author/year

Subjects

Methods

HRV measure

Data collection/analysis

Eguchi et al., 201082

Normotensive or hypertensive T2DM (N ¼ 200,age 66  9, 47% male) vs. hypertensive non-T2DM (N ¼ 257, age 68  9, 31%M) followed 67  27 months

Ambulatory blood pressure monitoring (ABPM)

ABPM collected for 1 min/30 min for 24 h, parameters averaged over awake, sleep and 24-h.

Stein et al., 200783

N ¼ 272 from Sleep Heart Health Study (age: 76  4, 154F/118M)

PSG

HR derived from ABPM, SDHR, and RMSSD during awake, sleep and 24-h respectively, SBP and HR relation-ships derived for awake, sleep and 24-h SDNNDIX, rMSSD, pNN625, total power, VLF, LF, HF

ABPM: Ambulatory blood pressure monitoring PSG: polysomnogram HR: heart rate T2DM: type 2 diabetes mellitus SDHR: standard deviation of heart rate rMSSD: root mean square of successive differences of NN intervals for period of interest SBP: systolic blood pressure SDNNDIX: 5-min average of standard deviations of NN intervals pNN625: percent of NN intervals different from the prior one by 6.25% of local average NN VLF: very low frequency power LF: low frequency power HF: high frequency power

5-min window analysis over the record

P.K. Stein, Y. Pu / Sleep Medicine Reviews 16 (2012) 47e66

Research agenda 1) Larger studies of relationships of PSG HRV, PSG HRV by sleep stage or integrated PSG HRV and EEG signals and risk of cardiovascular outcomes in different populations. 2) Improved software to more easily obtain usable HRV from PSGs.

Conclusions and recommendations The many studies reviewed here clearly demonstrate the relevance of HRV analysis to clinical sleep medicine. At the same time, it is clear that clinical applications of HRV to sleep are in their infancy, and the naiveté with which HRV numbers are sometimes accepted as measures of autonomic function without examination of the underlying HR patterns must be emphasized. This is a general problem in HRV research, not just in sleep-related applications. Also, although not discussed in detail here, there is enormous potential for integrating the many PSG signals with the HR signal to gain further insight into mechanisms underlying health and disease. In the light of the enormous number of stored electronic ECG signals already available from PSGs and the ease with which they can be processed using commercially-available Holter analyzers, collaborations between investigators knowledgeable about HRV and sleep researchers would make important contributions to sleep medicine. References *1. Kleiger RE, Stein PK, Bigger Jr JT. Heart rate variability: measurement and clinical utility. Ann Noninvasive Electrocardiol 2005;10:1e14. 2. Kleiger RE, Miller JP, Bigger JT, Moss AJ, The Multicenter Post-infarction Research Group. Decreased heart rate variability and its association with increased mortality after myocardial infarction. Am J Cardiol 1987;59:256e62. 3. Jokinen V, Tapanainen JM, Seppänen T, Huikuri HV. Temporal changes and prognostic significance of measures of heart rate dynamics after acute myocardial infarction in the beta-blocking era. Am J Cardiol 2003;92:907e12. *4. Stein PK, Domitrovich PP, Hui N, Rautaharju P, Gottdiener J. Sometimes higher heart rate variability is not better heart rate variability: results of graphical and non-linear analyses. J Cardiovasc Electrophysiol 2005;16:954e9. 5. Bilge AR, Stein PK, Domitrovich PP, Gérard PL, Rottman JN, Kleiger RE, et al. Assessment of ultra low frequency band power of heart rate variability: validation of alternative methods. Int J Cardiol 1999;71:1e6. 6. Taylor JA, Carr DL, Myers CW, et al. Mechanisms underlying very-lowfrequency RR-interval oscillations in humans. Circulation 1998;98:547e55. 7. Bauer A, Malik M, Schmidt G, Barthel P, Bonnemeier H, Cygankiewicz I, et al. Heart rate turbulence: standards of measurement, physiological interpretation, and clinical use: International Society for Holter and Noninvasive Electrophysiology Consensus. J Am Coll Cardiol 2008;52:1353e65. 8. Stein PK, Redline S. Abnormal heart rate turbulence from PSGs is associated with cardiovascular mortality in the elderly: results from the Sleep Heart Health Study. Sleep 2010;33(A378). 9. Zemaityte D, Varoneckas G, Sokolov E. Heart rhythm control during sleep. Psychophysiology 1984;21:279e89. 10. Bonnet MH, Arand DL. Heart rate variability: sleep stage, time of night, and arousal influences. Electroencephalogr Clin Neurophysiol 1997;102:390e6. *11. Otzenberger H, Gronfier C, Simon C, Charloux A, Ehrhart J, Piquard F, et al. Dynamic heart rate variability: a tool for exploring sympathovagal balance continuously during sleep in men. Am J Physiol Heart Circ Physiol 1998;275(3):H946e50. 12. Scholz UJ, Bianchi AM, Cerutti S, Kubicki S. Vegetative background of sleep: spectral analysis of the heart rate variability. Physiol Behav 1997;62(5):1037e43. 13. Crasset V, Mezzetti S, Antoine M, Linkowski P, Degaute JP, van de Borne P. Effects of aging and cardiac denervation on heart rate variability during sleep. Circulation 2001;103(1):84e8. 14. Vanoli E, Adamson PB, Ba-Lin, Pinna GD, Lazzara R, Orr WC. Heart rate variability during specific sleep stages. A comparison of healthy subjects with patients after myocardial infarction. Circulation 1995;91(7):1918e22.

* The most important references are denoted by an asterisk.

65

15. Shinar Z, Akselrod S, Dagan Y, Baharav A. Autonomic changes during wakesleep transition: a heart rate variability based approach. Auton Neurosci 2006;130(1e2):17e27. 16. Stein PK, Domitrovich PP, Redline S. Cardiac autonomic function during different sleep stages in the elderly: results from the Sleep Heart Health Study. Sleep; June 10, 2008. Baltimore, MD. Abstract 0297. 17. Lanfranchi PA, Fradette L, Gagnon JF, Colombo R, Montplaisir J. Cardiac autonomic regulation during sleep in idiopathic REM sleep behavior disorder. Sleep 2007;30(8):1019e25. 18. Leung RST. Sleep-disordered breathing: autonomic mechanisms and arrhythmias. Prog Cardiovasc Dis 2009;51(4):324e38. 19. Somers VK, White DP, Amin R, Abraham WT, Costa F, Culebras A, et al. Sleep apnea and cardiovascular disease e an American Heart Association/American College of Cardiology Foundation Scientific Statement from the American Heart Association Council for High Blood Pressure Research Professional Education Committee, Council on Clinical Cardiology, Stroke Council, and Council on Cardiovascular Nursing. J Am Coll Cardiol 2008;52:686e717. *20. Stein PK, Duntley SP, Domitrovich, Nishith P, Carney RM. A simple method for detecting sleep-disordered breathing using holter monitoring. J Cardiovasc Electrophysiol 2003;14:467e73. 21. Stein PK, Pu Y, Goldmuntz A, MacNamara A, Gropper C, Elling R, et al. Feasibility of using mobile cardiac outpatient telemetry to identify severe sleep disorders. Heart Rhythm 2009;6(5):S259e60 [Abstract]. 22. Somers VK, Dyken ME, Clary MP, Abboud FM. Sympathetic neural mechanisms in obstructive sleep apnea. J Clin Invest 1995;96:1897e904. 23. Dingli K, Assimakopoulos T, Wraith PK, Fietzel I, Witt C, Douglas NJ. Spectral oscillations of RR intervals in sleep apnoea/hypopnea syndrome patients. Eur Respir J 2003;22(6):943e50. 24. Guilleminault C, Poyares D, Rosa A, Huang YS. Heart rate variability, sympathetic and vagal balance and EEG arousals in upper airway resistance and mild obstructive sleep apnea syndromes. Sleep Med 2005;6(5):451e7. 25. Hilton MF, Chappell MJ, Bartlett WA, Malhotra A, Beattie JM, Cayton RM. The sleep apnoea/hypopnea syndrome depresses waking vagal tone independent of sympathetic activation. Eur Respir J 2001;17:1258e66. *26. Narkiewicz K, Montano N, Cogliati C, van de Borne PJH, Dyken ME, Somers VK. Altered cardiovascular variability in obstructive sleep apnea. Circulation 1998;98:1071e7. 27. Ueno LM, Drager LF, Rodrigues ACT, Rondon MUPB, Mathias Jr W, Krieger EM, et al. Dayenight pattern of autonomic nervous system modulation in patients with heart failure with and without sleep apnea. Int J Cardiol; Nov 16, 2009 [Epub ahead of print]. 28. Liao D, Li X, Rodriguez-Colon SM, Liu J, Vgontzas AN, Calhoun S, et al. Sleepdisordered breathing and cardiac autonomic modulation in children. Sleep Med 2010;11:484e8. April 1, 2010 [Epub ahead of print]. 29. Kwok KL, Yung TC, Ng DK, Chan CH, Lau WF, Fu YM. Heart rate variability in childhood obstructive sleep apnea. Pediatr Pulmonol; Jan 18, 2011 [Epub ahead of print]. 30. Wang W, Tretriluxana S, Redline S, Surovec S, Gottlieb DJ, Khoo MCK. Association of cardiac autonomic function measures with severity of sleep-disordered breathing in a community-based sample. J Sleep Res 2008;17:251e62. 31. Jurysta F, Lanquart JP, van de Borne P, Migeotte PF, Dumont M, Degaute JP, et al. The link between cardiac autonomic activity and sleep delta power is altered in men with sleep apnea-hypopnea syndrome. Am J Physiol Regul Integr Comp Physiol 2006;291:R1165e71. 32. Sforza E, Pichot V, Barthelemy JC, Haba-Rubio J, Roche F. Cardiovascular variability during periodic leg movements: a spectral analysis approach. Clin Neurophysiol 2005;116(5):1096e104. 33. Winkelman JW. The evoked heart rate response to periodic leg movements of sleep. Sleep 1999;22:575e80. 34. Guggisberg AG, Hess CW, Mathis J. The significance of the sympathetic nervous system in the pathophysiology of periodic leg movements in sleep. Sleep 2007;30(6):755e66. 35. Walter LM, Foster AM, Patterson RR, Anderson V, Davey MJ, Nixon GM, et al. Cardiovascular variability during periodic leg movements in sleep in children. Sleep 2009;32(8):1093e9. 36. Guilleminault C, Connolly S, Windle R, Melvin K, Tilkian A. Cyclic variation of the heart rate in sleep apnea syndrome e mechanisms, and usefulness of 24 h electrocardiography as a screening technique. Lancet 1984;I:126e31. 37. Penzel T, McNames J, de Chazal P, Raymond B, Murray A, Moody G. Systematic comparison of different algorithms for apnea detection based on electrocardiogram recordings. Med Biol Eng Comput 2002;40:402e7. *38. Heneghan C, de Chazal P, Ryan S, Chua CP, Doherty L, Boyle P, et al. Electrocardiogram recording as a screening tool for sleep disordered breathing. J Clin Sleep Med 2008;4(3):223e8. 39. Chang KM. Portable obstructive sleep apnea screening system using overnight ECG and a PDA-based wireless transmission system. Telemed e-Health 2009;15(4):353e61. 40. Mietus JE, Peng CK, Ivanov PCh, Goldberger AL. Deteection of obstructive sleep apnea from cardiac interbeat interval time series. Comput Cardiol 2000;27:753e6. 41. Vazir A, Dayer M, Hastings PC, Mclntyre HF, Henein MY, Poole-Wilson PA, et al. Can heart rate variation rule out sleep disordered breathing in heart failure? Eur Respir J 2006;27:571e7. 42. Tateishi O, Shouda T, Sakai T, Honda Y, Mochizuki S, Machida K. Apnea-related heart rate variability in congestive heart failure patients. Clin Exp Hypertens 2003;25(3):183e9.

66

P.K. Stein, Y. Pu / Sleep Medicine Reviews 16 (2012) 47e66

43. Shalaby AA, Atwood CW, Hansen C, Konermann M, Freedman R, Fowler J, et al. Analysis of interaction of acute atrial overdrive pacing with sleep-related breathing disorder. Am J Cardiol 2007;99:573e8. 44. Available from: http://www.physionet.org/challenge/2000. 45. Roche F, Sforza E, Duverney D, Borderies J, Pichot V, Bigaignon O, et al. Heart rate increment: an electrocardiological approach for the early detection of obstructive sleep apnoea/hypopnoea syndrome. Clin Sci (Lond) 2004;107:105e10. 46. Roche F, Pichot V, Sforza E, Court-Fortune I, Duverney D, Costes F, et al. Predicting sleep apnoea syndrome from heart period: a timeefrequency wavelet analysis. Eur Respir J 2003;22:937e42. 47. Roche F, Gaspoz JM, Court-Fortune I, Minini P, Pichot V, Duverney D, et al. Screening of obstructive sleep apnea syndrome by heart rate variability analysis. Circulation 1999;100:1411e5. 48. Available from: http://www.accessdata.fda.gov/cdrh_docs/pdf4/K042745.pdf. 49. Hayano J, Watanabe E, Saito Y, Sasaki F, Fujimoto K, Nomiyama T, et al. Screening for obstructive sleep apnea by cyclic variation of heart rate. Cric Arrhythm Electrophysiol; November 12, 2010. 50. Parrino L, Ferrillo F, Smerieri A, Spaggiari MC, Palomba V, Rossi M, et al. Is insomnia a neurophysiological disorder? The role of sleep EEG microstructure. Brain Res Bull Jun 30, 2004;63:377e83 [Review]. 51. Riemann D, Spiegelhalder K, Feige B, Voderholzer U, Berger M, Perlis M, et al. The hyperarousal model of insomnia: a review. Sleep Med Rev 2010;14(1):19e31. *52. Bonnet MH, Arand DL. Heart rate variability in insomniacs and matched normal sleepers. Psychosom Med 1998;60(5):610e5. 53. Fang SC, Huang CJ, Yang TT, Tsai PS. Heart rate variability and daytime functioning in insomniacs and normal sleepers: preliminary results. J Psychosom Res 2008;65(1):23e30. 54. Jurysta F, Lanquart JP, Sputaels V, Dumont M, Migeotte PF, Leistedt S, et al. The impact of chronic primary insomnia on the heart rateeEEG variability link. Clin Neurophysiol 2009;120(6):1054e60. 55. Mullington JM, Haack M, Toth M, Serrador JM, Meier-Ewert HK. Cardiovascular, inflammatory, and metabolic consequences of sleep deprivation. Prog Cardiovasc Dis 2009;51(4):294e302. 56. Barnett KJ, Cooper NJ. The effects of a poor night sleep on mood, cognitive, autonomic and electrophysiological measures. J Integr Neurosci 2008;7(3):405e20. 57. Takase B, Akima T, Satomura K, Ohsuzu F, Mastui T, Ishihara M, et al. Effects of chronic sleep deprivation on autonomic activity by examining heart rate variability, plasma catecholamine, and intracellular magnesium levels. Biomed Pharmacother 2004;58(Suppl. 1):S35e9. 58. Michail E, Kokonozi A, Chouvarda I, Maglaveras N. EEG and HRV markers of sleepiness and loss of control during car driving. Conf Proc IEEE Eng Med Biol Soc; 2008:2566e9. 59. Chung MH, Kuo TB, Hsu N, Chu H, Chou KR, Yang CC. Sleep and autonomic nervous system changes e enhanced cardiac sympathetic modulations during sleep in permanent night shift nurses. Scand J Work Environ Health 2009;35(3):180e7. 60. Nielsen T, Paquette T, Solomonova E, Lara-Carrasco J, Colombo R, Lanfranchi P. Changes in cardiac variability after REM sleep deprivation in recurrent nightmares. Sleep 2010;33(1):113e22. 61. Sauvet F, Leftheriotis G, Gomez-Merino D, Langrume C, Drogou C, Van Beers P, et al. Effect of acute sleep deprivation on vascular function in healthy subjects. J Appl Physiol 2010;108(1):68e75. 62. Sayk F, Teckentrup C, Becker C, Heutling D, Wellhöner P, Lehnert H, et al. Effects of selective slow-wave sleep deprivation on nocturnal blood pressure dipping and daytime blood pressure regulation. Am J Physiol Regul Integr Comp Physiol 2010;298(1):R191e7. 63. Pagani M, Pizzinelli P, Traon AP, Ferreri C, Beltrami S, Bareille MP, et al. Hemodynamic, autonomic and baroreflex changes after one night sleep deprivation in healthy volunteers. Auton Neurosci 2009;145(1e2):76e80. 64. Zhong X, Hilton HJ, Gates GJ, Jelic S, Stern Y, Bartels MN, et al. Increased sympathetic and decreased parasympathetic cardiovascular modulation in

65.

66.

67.

68.

69.

70.

71. 72.

73.

74.

*75.

*76.

77.

78.

79.

80. 81.

82.

normal humans with acute sleep deprivation. J Appl Physiol 2005;98(6):2024e32. Mellman TA, Knorr BR, Pigeon WR, Leiter JC, Akay M. Heart rate variability during sleep and the early development of posttraumatic stress disorder. Biol Psychiatry 2004;55(9):953e6. Lombardi C, Parati G, Cortelli P, Provini F, Vetrugno R, Plazzi G, et al. Daytime sleepiness and neural cardiac modulation in sleep-related breathing disorders. J Sleep Res 2008;17(3):263e70. Heart rate variability: standards of measurement, physiological interpretation and clinical use. Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology. Circulation 1996;93(5):1043e65. Roche F, Court-Fortune I, Pichot V, et al. Reduced cardiac sympathetic autonomic tone after long-term nasal continuous positive airway pressure in obstructive sleep apnoea syndrome. Clin Physiol 1999;19:127e34. Shiina K, Tomiyama H, Takata Y, Yoshida M, Kato K, Saruhara H, et al. Effects of CPAP therapy on the sympathovagal balance and arterial stiffness in obstructive sleep apnea. Respir Med 2010;104(6):911e6. Gilman MP, Floras JS, Usui K, Kaneko Y, Leung RST, Bradley TD. Continuous positive airway pressure increases heart rate variability in heart failure patients with obstructive sleep apnoea. Clin Sci 2008;114:243e9. Khoo MCK, Kim TS, Berry RB. Spectral indices of cardiac autonomic function in obstructive sleep apnea. Sleep 1999;22:443e51. Khoo MCK, Belozeroff V, Berry RB, Sasson CSH. Cardiac autonomic control in obstructive sleep apnea: effects of long-term CPAP therapy. Am J Respir Crit Care Med 2001;164:807e12. Karasulu L, Epozturk PO, Sukucu SN, Dalar L, Altin S. Improving heart rate variability in sleep apnea patients: differences in treatment with auto-titrating positive airway pressure (APAP) versus conventional CPAP. Lung 2010;188(4):315e20. Dursunoglu N, Dursunoglu D, Cuhadaroglu C, Kilicaslan Z. Acute effects of automated continuous positive airway pressure on blood pressure in patients with sleep apnea and hypertension. Respiration 2005;72:150e5. Patruno V, Aiolfi S, Costantino G, Murgia R, Selmi C, Maliani A, et al. Fixed and autoadjusting continuous positive airway pressure treatments are not similar in reducing cardiovascular risk factors in patients with obstructive sleep apnea. Chest 2007;131:1393e9. Thomas RJ, Mietus JE, Peng CK, Gilmartin G, Daly RW, Goldberger AL, et al. Differentiating obstructive from central and complex sleep apnea using an automated electrocardiogram-based method. Sleep 2007;30(12): 1756e69. Thomas RJ, Weiss MD, Mietus JE, Peng CK, Goldberger AL, Gottlieb DJ. Prevalent hypertension and stroke in the Sleep Heart Health Study: association with an ECG-derived spectrographic marker of cardiopulmonary coupling. Sleep 2009;32(7):897e904. Thomas RJ, Mietus JE, Peng CK, Goldberger AL, Crofford LJ, Chervin RD. Impaired sleep quality in fibromyalgia: detection and quantification with ECG-based cardiopulmonary coupling spectrograms. Sleep Med 2010;11(5): 497e8. Yeh GY, Mietus JE, Peng CK, Phillips RS, Davis RB, Wayne PM, et al. Enhancement of sleep stability with Tai Chi exercise in chronic heart failure: preliminary findings using an ECG-based spectrogram method. Sleep Med 2008;9:527e36. Verrier RL, Josephson ME. Impact of sleep on arrhythmogenesis. Circ Arrhythm Electrophysiol 2009;2:450e9. Eguchi K, Schwartz JE, Pickering TG, Hoshide S, Ishikawa J, Shimada K, et al. Increased heart rate variability during sleep is a predictor for future cardiovascular events in patients with type 2 diabetes. Hypertens Res 2010; 33(7):737e42. Stein P, Domitrovich P, Lundequam E, Redline S. Decreased heart rate variability from polysomnography predicts mortality in the elderly (ABS). Sleep 2007;30:A110.