Sleep Disordered Breathing in Children is Associated with Impairment of Sleep Stage Specific Shift of Cardiac Autonomic Modulation

Abstract

We examined the effects of sleep stages and sleep disordered breathing (SDB) on autonomic modulation in 700 children. Apnea Hypopnea Index (AHI) during one 9-hour nighttime polysomnography was used to define SDB. Sleep stage specific autonomic modulation was measured by heart rate variability (HRV) analysis of the first available 5-minute RR intervals from each sleep stage. The mean (SD) age was 112 (21) months (49% male and 25% non-Caucasian). The average AHI was 0.79 (SD=1.03)/hour, while 73.0%, 25.8%, and 1.2% of children had AHI < 1 (No-SDB), 1–5 (Mild-SDB), and ≥ 5 (Moderate-SDB), respectively. In no-SDB group, the HF and RMSSD significantly increased from wake to stage 2, and slow-wave sleep (SWS), and then decreased dramatically when shifting into REM sleep. In moderate-SDB group, the pattern of HRV shift is similar to that of no-SDB. However, the decreases in HF and RMSSD from SWS to REM were more pronounced in moderate-SDB children [between group differences in HF (−24% in moderate-SDB vs. −10% in no-SDB) and RMSSD (−27% vs. −12%) were significant (p < 0.05)]. The REM stage HF is significantly lower in moderate-SDB group compared to no-SDB group [mean (SE): 4.49 (0.43) vs. 5.80 (0.05) ms², respectively, p < 0.05].

Conclusions: autonomic modulation significantly shifts towards higher parasympathetic modulation from wake to non-REM sleep, and reverses to a less parasympathetic modulation during REM sleep. However, the autonomic modulation is impaired among children with moderate-SDB in the directions of more reduction in parasympathetic modulation from SWS to REM sleep and significantly weaker parasympathetic modulation in REM sleep, which may lead to higher arrhythmia vulnerability, especially during REM sleep.

Introduction

In the cardiac disease literature, analysis of beat-to-beat HRV has emerged as a clinically and epidemiologically useful non-invasive method to quantitatively assess cardiac autonomic activity (Liao et al. 1997, Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology. 1996, Tsuji et al. 1996, Vaishnav et al. 1994,). Most importantly, lower HRV has been consistently associated with the risk of incident cardiovascular disease (Dekker et al. 2000, Liao et al. 1997, 1999, 2002, Tsuji et al. 1996), It is well-known that an episode of apnea is accompanied by a typical pattern of heart rate fluctuations – bradycardia during the apnoeic phase and tachycardia at the restoration of breathing. Several studies in adults have shown that sleep-disordered breathing (SDB) is significantly associated with lower heart rate variability (HRV), indicating impaired cardiac autonomic modulation among persons with apnea (Guilleminault et al. 1984, Penzel et al. 2003, Roche et al. 2003, Clifford & Tarassenko 2004,). Also, in adults, several studies have shown a normal healthy shift of the sympathetic and parasympathetic modulation of heart rhythm in the direction of reduced sympathetic outflow and increased sympathetic modulation from wake to NREM sleep, a pattern which is reversed in REM sleep (Brandenberger et al. 2003, Clifford & Tarassenko 2004, Mendez et al. 2006, Viola et al. 2002, Virtanen et al. 2007). Baharav et al reported a significant association between SDB and sleep stage specific shift of HRV in 10 SDB and 10 normal control children (Baharav et al. 1999). However, little is known about the quantitative relationship between sleep stages (REM, nonREM, and wake stages) and cardiac autonomic modulation in population-based sample of young children, and whether SDB has a negative impact on such relationship.

This study was designed to investigate the pattern of shifting of cardiac autonomic modulation across sleep stages (Wake, non-REM, and REM stages) in a population-based sample of young children and to investigate whether SDB has an adverse effect on the sleep stage specific autonomic modulation.

Methods

Population

The study population was from the Penn State Child Cohort (PSCC). The PSCC is a population-based study of the prevalence and correlates of SDB in prepubertal children. The study sample was selected through a two-phase sampling process, and during the first phase we collected general information from the parents about their child’s sleep and behavioral patterns, while during the second phase we collected more detailed data from a one night sleep study in our General Clinical Research Center (GCRC). The study was reviewed and approved by our Institutional Review Board as well as the GCRC review board. Briefly, in the first phase, a questionnaire was sent home to the parents of all elementary school (grade K-5) children in three school districts in Dauphin county, central Pennsylvania. In the first phase, we sent home 7,312 questionnaires and 5,740 were returned, with a response rate of 79%. In the second phase of sampling, each year we randomly selected 200 children from those who returned our questionnaires in phase I to participate in our GCRC sleep study. We stratified this sampling based on the severity of risk for SDB as reported by parents in the Phase I questionnaire and by grade and gender. A total of 700 children enrolled in our Phase II study, with a response rate of 70%.

Methods

During the Phase II, a detailed history and physical examination and a 9-hour polysomnogram (PSG) recording were performed in our GCRC. The physical examination, completed in the evening prior to the PSG recording, included height, weight, hip and waist circumferences, neck measurement, blood pressure, and a visual evaluation of the nose and throat by an ENT specialist and an evaluation of the respiratory function by a pediatric pulmonologist. All measurements were based on standardized procedures and protocols, and were subject to established quality control procedures.

For the sleep evaluation, all subjects, in the presence of a parent, spent one night in sound-attenuated, temperature- and light-controlled rooms in our GCRC. During this time the child’s sleep was continuously monitored for nine hours (24 analog channel and 10 dc channel TS amplifier using Gamma software, Grass-Telefactor Inc) with a four-channel electroencephalogram, a two-channel electro-oculogram, and a single-channel electromyogram. A single-channel electrocardiograph (ECG) was also recorded at a sampling rate of 100 Hz. The sleep records were subsequently scored independently according to standardized criteria (American Academy of Pediatrics 2002). All records were double scored. Respiration was monitored throughout the night by use of thermocouple at the nose and mouth (model TCT R, Grass-Telefactor, Inc), nasal pressure (MP 45–871 ± 2 cm H2O, Validyne Engineering Corp) and thoracic and abdominal strain gauges (model 1312 Sleepmate Technologies Midlothian, VA). A subjective estimate of snoring was obtained from parental report. In addition, we obtained an objective estimate of snoring during the PSG by monitoring breathing sounds with a microphone attached to the throat (model 1250 Sleepmate Technologies Midlothian, VA) ) as well as a separate room microphone. All night hemoglobin oxygen saturation (SaO₂) was obtained from the finger (model 8800, Noonin Medical).

SDB definition

We defined sleep apnea and hypopnea based on criteria currently used in clinical practice (American Academy of Pediatrics 2002, American Thoracic Society 1996). An apnea was defined as a cessation of airflow with a minimum duration of five seconds and an out of phase strain gauge movement. A hypopnea was defined as a reduction of airflow of approximately 50% with an associated decrease in oxygen saturation (SaO₂) of at least 3% or an associated arousal. We then combined the total number of apnea and hypopnea episodes and divided it by the total duration of sleep to form the Apnea Hypopnea Index (AHI) [episodes / hour of sleep]. We further defined children as “without SDB” if AHI was < 1, as “mild SDB” if AHI ≥ 1 but < 5, and as “moderate SDB” if AHI ≥ 5.

Sleep Stage Identification

From the 9-hour PSG, five stages were identified (wake stage, stage 2, stage 3, stage 4, and stage REM) using standardized criteria (Rechtschaffen & Kales 1968). An indicator of the clock time was also available from the PSG recording. For the purpose of this investigation, we used the cardiac autonomic modulation data calculated for wake stage (in the beginning of PSG recording), the non-REM stage (represented by stage 2 sleep and slow wave sleep), and REM sleep.

HRV variability indices

From the 1-channel ECG voltage data, we first converted the voltage file form into a wave form file. Then we identified the peak of each QRS complex as the R wave point. We then calculated beat-to-beat RR interval data from the entire night of ECG recording. A potential artifacts identification and remove algorithm was applied to the entire RR interval data. This algorithm identifies as artifacts (and removes them) any RR interval < 375 ms, any RR interval > 1200 ms, or RR interval ratio from two adjacent RR intervals < 0.8 or > 1.2.

We applied HRV analysis to the above described “artifact-free” RR interval files if the total length of the “artifact free” RR interval file was more than 6.5 hours (corresponding to approximately 75% of the total recording time). As a result, we performed HRV analysis for this study on 616 participants (88% of total sample size of 700). For the HRV analysis, we initially identified and output the first available 5-minute (300 seconds) continuous RR from each sleep stage for stage-specific HRV analysis, after excluded the segment where an episode of AHI occurred if there was any that were sampled. In general, because of the high sleep efficiency in this population-based sample of young children, we were able to obtain such first available 5-minute stage-specific RR data from the first two sleep cycles. As a result, the stage-specific RR data we analyzed were obtained from a very similar time frame. We then calculated, based on the sleep stage specific 5 minutes (300 seconds) RR interval window, the following frequency domain HRV indices for each of the sleep stages: Low Frequency Power (LF) – the power in the low frequency range (0.04 – 0.15 HZ), High Frequency Power (HF) – the power in the high frequency range (0.15 – 0.40 Hz), and LF/HF Ratio. We also calculated the following time domain HRV indices for this study: SDNN – the standard deviation of all RR intervals (ms), RMSSD – the square root of the mean of the sum of the squares of differences between adjacent RR intervals (ms), and Mean HR – Mean heart rate (BPM). For the HRV analysis, we used the HRV Analysis Software (Biomedical Signal Analysis Group, 2009). When performing frequency domain HRV analysis, we used Fast Fourier Transformation. We first detrended the 300 seconds RR series to remove second order linear trend, and we then interpolated the RR-series at 2 Hz interpolation rate to achieve data spatially stationary. In general, HF represents the parasympathetic modulation, LF represents a mixture of sympathetic and parasympathetic modulation. To some degree, LF/HF ratio represents the balance of sympathetic and parasympathetic modulation. Lower HF and higher LF/HF ratio are used as indicators of autonomic balance of less parasympathetic, and some time relatively more sympathetic modulation. The time domain measures are used to represent the overall variability of RR intervals, with SDNN be highly correlated with the total power of frequency domain analysis, and RMSSD highly correlated to HF from the frequency domain analysis.

Other covariables

The covariables we investigated included age, race, sex, BMI, percent REM sleep, snore status, and sleep efficiency. All covariables were collected during the Phase II physical examination or the PSG, and subject to standardized protocols. The BMI percentile was used, which was calculated as the percentile of BMI distribution for age and gender, based on CDC criteria (Center for Disease Control and Prevention 2008). Percent REM sleep was defined as the proportion of sleep time spent in REM stage. Sleep efficiency was defined as proportion of the recorded time that the participant was in a sleep stage.

Statistical Methods

Means and proportions of main variables were calculated for the entire study population, as well as stratified according to SDB status. The analysis of covariance (ANCOVAR) was used to calculate multivariable adjusted means and the standard errors (SE) of HRV indices at each level of SDB and each sleep stage. Following the convention of HRV analysis (Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology 1996), HF and LF values were log-transformed when analyzed (termed Log-HF and Log-LF respectively). We accounted for the sampling probability from Phase I to Phase II enrollments in all of the analyses in order to generate population level estimates and to make inference back to the population from which the Phase II study participants were selected from. Since this is the first opportunity to systematically investigate the sleep stage, SDB, and autonomic modulation in a large population-based sample of young children, we consider our analysis exploratory and hypothesis generating. Therefore, we did not adjust our analysis for multiple comparisons. SAS 9.1 was used for our analysis.

Results

The population characteristics are presented in Table 1, as mean levels for continuous variables and as proportions for categorical variables. The mean (SD) age of participants was 112 (21) months, with 49% male and 25% non-Caucasian. The average AHI (SD) was 0.79 (1.03) / hour during the entire sleep, while 73.0% had AHI < 1 / hour (No SDB group), 25.8% had 1–5 AHI / hour (mild SDB group), and 1.2% had ≥ 5 AHI / hour (Moderate SDB group). For 7 children with moderate SDB, the average AHI is 5.79, ranging from 5.02 to 7.47. None of the 7 moderate SDB children has any chronic conditions. In this population, both systolic and diastolic blood pressures were higher among children with higher levels of AHI (Bixler et al. 2008). Children who snored during the sleep period were also more likely to have higher levels of AHI. In general, children in the “moderate SDB” group had lower HRV indices, calculated from the entire night or 5 minutes wake stage ECG data.

Table 1.

Characteristics of the Study Population

	All	SDB Status
	(N=616)	No SDB AHI < 1 (N=450)	Mild SDB AHI 1 – 5 (N=159)	Moderate SDB AHI ≥ 5 (N=7)	p-value
Age (Months)	112 (21)	110 (21)	112 (19)	117 (17)	NS
Male (%)	49	48	52	15	< 0.05
Race (% non-Caucasian)	25	20	36	0	< 0.05
BMI Percentile	61 (29)	59 (30)	66 (28)	59 (41)	< 0.05
REM Sleep (%)	20 (5.6)	20 (5.9)	20 (4.9)	18 (6.8)	NS
Sleep Efficiency (%)	86 (8.5)	85 (8.6)	87 (7.9)	80 (7.8)	NS
Snore (%)	24	20	34	65	< 0.05
AHI (# / hr sleep)	0.79 (1.03)	0.29 (0.27)	1.98 (0.80)	5.79 (0.93)	< 0.01
Systolic BP (mmHg)	111 (12)	110 (12)	113 (11)	128 (11)	< 0.01
Diastolic BP (mmHg)	65 (8)	65 (8)	65 (7)	72 (5.3)	NS
One night (9 hours) HRV
Log-HF (ms2)	6.70 (0.84)	6.67 (0.86)	6.83 (0.77)	5.91 (0.73)	< 0.01
Log-LF (ms2)	6.57 (0.66)	6.55 (0.66)	6.65 (0.64)	6.37 (0.71)	NS
LF/HF Ratio	0.98 (0.53)	1.00 (0.55)	0.91 (0.42)	1.69 (0.62)	< 0.01
SDNN (ms)	94 (27.4)	93 (27.1)	98 (28.2)	85 (23.1)	NS
RMSSD (ms)	73 (34.0)	72 (34.2)	78 (33.3)	45 (15.5)	< 0.01
Heart Rate (bpm)	77 (8.2)	77 (8.2)	76 (8.4)	79 (6.8)	NS

NS Not significant (p > 0.05)

Log-LF log transformed low frequency power

Log-HF log transformed high frequency power

LF/HF the ratio of LF and HF

SDNN the standard deviation of all RR intervals (ms)

RMSSD the square root of the mean of the sum of the squares of differences between adjacent RR intervals (ms)

The age, race, sex, BMI percentile, percent REM sleep, snore status, and sleep efficiency adjusted mean levels of 5-minute HRV indices and their standard errors according to SDB categories and sleep stages are presented in Table 2. Data in Table 2 are assessed in two ways: (A) within each of the three SDB groups comparisons to highlight the sleep stage related shift of HRV (p-values are indicated in “p-1” column) and (B) between SDB group comparisons of the same stage HRV (p-values are indicated in “p-2” column). There are several patterns that are most worthy highlighting: First, in children without SDB, there is a significant increase in HF, but a decrease in LF/HF ratio from wake to non-REM sleep, and then a decrease in HF and an increase in LF/HF ratio during the REM stage. RMSSD showed a similar pattern as HF. The LF, HR, and to a lesser degree SDNN, showed a similar pattern as LF/HF ratio. The statistical comparisons of the means at each sleep stage to that of wake are presented in column “p-1” in Table 2. These consistent stage specific changes of HRV indices are indicative of a shift toward a more parasympathetic modulation from wake to stage 2 and SWS, and a shift back to a less parasympathetic modulation during REM sleep. This pattern of changing HRV indices across sleep stages suggests a normal healthy shift of the cardiac autonomic modulation of heart rhythm in the direction of increased parasympathetic modulation from wake to stage 2 and SWS, followed by a reversal to reduced parasympathetic modulation in REM sleep. Secondly, among children with moderate SDB, the pattern of HRV indices shift across sleep stages is similar to that of no-SDB group, except for LF/HF ratio, which did not decrease until SWS stage (it actually significantly increased in stage 2 (1.46, SE=0.15) compared to wake (1.04, SE=0.18), p=0.05). Furthermore, we calculated and presented in Figure 1 percentage differences in HRV indices contrasting SW and REM, in the moderate SDB and no-SDB groups. It is evident from this figure that in both groups, HF, RMSSD, and SDNN decreased, and LF increased, from SW sleep to REM sleep. The SW-REM changes were more pronounced in moderate SDB children, and the between group differences for HF (−24% [SE=7] in moderate SDB vs. −10% [SE=1] in no-SDB) and RMSSD (−27% [SE=5] vs. −12% [SE=2]) were significant (p < 0.05). The LF/HF ratio increased by 298 % and 410% from SW to REM, respectively for the no-SDB and moderate SDB group, although the between group difference was not statistically significant (data not shown in Figure 1). These data suggest more decrease in parasympathetic modulation from SW to REM sleep in the moderate SDB group. Thirdly, in children with moderate SDB, the mean levels of stage-specific HRV indices are generally lower than that in the no-SDB group, with the statistical comparisons shown in column “p-2” in Table 2. For instance, the REM stage HF and LF are significantly lower in moderate SDB group compared to no-SDB group (mean (SE) of HF: 4.49 (0.43) vs. 5.80 (0.05) ms², respectively, p < 0.05, mean (SE) of LF: 4.74 (0.36) vs. 5.36 (0.05), respectively, p=0.05). However, some of the statistical comparisons (column labeled “p-2” in Table 2) did not reach the traditional p ≤ 0.05 statistical significance level, because of small sample size and a population-based young children sample was analyzed, hence, a sample (N=7) of relatively less severe and shorter duration of SDB as compared to clinical samples. The lower HRV values in the moderate SDB group imply that the impaired stage-specific 5-minute HRV indices in this group are indicative of autonomic balance of less parasympathetic modulation. Lastly, the HRV indices shift across sleep stages among mild SDB group are in a similar pattern and direction as that of the no-SDB group, and unexpectedly, as shown in Table 2, the mild SDB group had a statistically better HRV profiles than the no-SDB group in all HRV indices during REM sleep, but significant differences in HRV indices between the mild-SDB and the no-SDB group were sparse or absent in other stages. Specifically, mild-SDB group had higher REM stage HF, LF, SDNN, RMSSD values, and lower LF/HF ratio and HR values, than that of the REM stage values in the no-SDB group, indicated by p-values in “p-2” column. None of the SW stage HRV comparisons between mild-SDB and Non-SDB groups was significant. Similarly, Stage 2 between group comparisons was only significant for HF and Wake Stage between group comparisons were only significant for HF, RMSSD, and HR. Our speculations on such unexpected finding are presented in the discussion section.

Table 2.

Multivariable Adjusted Means (SE) of HRV Indices calculated from the first available 5-minute continuous RR interval data from each sleep stage, according to SDB categories

SDB	Stage	HF (ms2)	p-1	p-2	LF (ms2)	p-1	p-2	LF/HF	p-1	p-2	SDNN (ms)	p-1	p-2	RMSSD (ms)	p-1	p-2	HR (bpm)	p-1	p-2
No SDB AHI < 1 (N=450)	Wake	5.80 (0.05)			5.84 (0.04)			1.28 (0.04)			64.0 (1.14)			55.3 (1.29)			85.7 (0.49)
	Stage 2	6.04 (0.05)	<0.01		5.62 (0.05)	<0.01		0.91 (0.04)	<0.01		61.9 (1.16)	NS		60.7 (1.39)	<0.01		78.0 (0.44)	<0.01
	SWS	6.03 (0.05)	<0.01		5.04 (0.05)	<0.01		0.53 (0.03)	<0.01		54.0 (1.04)	<0.01		60.4 (1.40)	<0.01		77.1 (0.45)	<0.01
	REM	5.40 (0.05)	<0.01		5.36 (0.05)	<0.01		1.42 (0.07)	0.05		55.1 (1.09)	<0.01		49.0 (1.24)	<0.01		79.9 (0.41)	<0.01
Mild SDB AHI 1 – 5 (N=159)	Wake	5.98 (0.08)		<0.05	5.89 (0.07)		NS	1.13 (0.08)		NS	65.7 (1.97)		NS	60.7 (2.23)		<0.05	83.5 (0.85)		<0.05
	Stage 2	6.24 (0.08)	<0.05	<0.05	5.67 (0.08)	<0.05	NS	0.82 (0.06)	<0.01	NS	64.8 (1.93)	NS	NS	64.5 (2.32)	NS	NS	77.5 (0.74)	<0.01	NS
	SWS	6.13 (0.08)	NS	NS	5.10 (0.09)	<0.01	NS	0.51 (0.05)	<0.01	NS	56.2 (1.78)	<0.01	NS	63.4 (2.38)	NS	NS	76.0 (0.77)	<0.01	NS
	REM	5.73 (0.09)	<0.01	<0.01	5.53 (0.08)	<0.01	<0.05	1.10 (0.12)	NS	<0.05	59.9 (1.81)	<0.05	<0.05	56.0 (2.06)	<0.05	<0.01	78.4 (0.67)	<0.01	<0.05
Moderate SDB AHI ≥ 5 (N=7)	Wake	5.41 (0.34)		NS	5.25 (0.32)		<0.05	1.04 (0.18)		NS	54.1 (8.52)		NS	45.5 (9.61)		NS	85.6 (3.67)		NS
	Stage 2	5.68 (0.37)	NS	NS	5.73 (0.39)	NS	NS	1.46 (0.15)	0.05	<0.01	58.2 (9.18)	NS	NS	48.6 (11.0)	NS	NS	77.4 (3.51)	<0.05	NS
	SWS	5.90 (0.36)	NS	NS	4.91 (0.38)	NS	NS	0.49 (0.20)	NS	NS	52.0 (7.75)	NS	NS	56.3 (10.4)	NS	NS	75.6 (3.37)	<0.01	NS
	REM	4.49 (0.43)	<0.01	<0.05	4.74 (0.36)	NS	0.05	1.63 (0.29)	0.05	NS	41.2 (8.62)	NS	NS	34.2 (9.80)	0.05	NS	81.9 (3.21)	NS	NS

^•Adjusted for age, race, sex, body mass index, percentage of REM sleep, snore status, and sleep efficiency.

^p-1p-value of within SDB group statistical comparisons of mean HRV: other stages vs. wake.

^p-2p-value of between SDB group statistical comparisons of mean stage-specific HRV: between no-SDB and mild-SDB or between no-SDB and moderate SDB.

Figure 1.

The percentage differences and Standard Errors in HRV indices contrasting SWS and REM, according to SDB status.

Footnote: (a) In both non-SDB and moderate SDB groups, HF, RMSSD, and SDNN decreased, and LF increased, from SWS into REM sleep.

(b) The SWS-REM changes were more pronounced in moderate SDB children, and the between group differences for HF (−24% in moderate SDB vs. −10% in no-SDB) and RMSSD (−27% vs. −12%) were significant at p < 0.05.

We also tested interactions between SDB and several covariables, such as gender, race, snore status, percent REM sleep, and sleep efficiency, and none of them were statistically significant at p < 0.10 level.

Discussion

In this population-based sample of young children, we observed several important findings. First, in all stages of sleep, there is in general a lower HRV among young children with even moderate SDB, as compared to children without SDB, with REM-stage HF, LF, and RMSSD comparisons reaching statistical significance. This finding is consistent with our a priori hypothesis that moderate SDB is significantly associated with impaired HRV profile in the direction of reduced parasympathetic modulation. It is also consistent with our findings from the analysis of the HRV indices calculated from the entire night ECG data (Liao et al, 2009). Second, in children without SDB, there is a healthy and normal shift of cardiac autonomic modulation across sleep stages, in the direction of more parasympathetic modulation from wake to non-REM sleep, and a reversal to a balance of less parasympathetic modulation entering REM sleep. However, such normal shift of cardiac autonomic modulation is impaired among children with moderate SDB, in the pattern of much more pronounced decline in parasympathetic modulation in REM stage. Our data (Figure 1) suggest that the difference between SWS and REM were more pronounced in moderate SDB vs. no-SDB children for LF/HF ratio, HF and RMSSD. These data suggest more parasympathetic modulation reduction during REM stage in the moderate SDB group. It is worth mentioning that the above summarized associations remained statistically significant after multivariable adjustment of several factors that were associated with SDB and HRV, such as age, race, sex, BMI percentile, percent REM sleep, snore status, and sleep efficiency. These findings are similar to that reported by Baharav et al in 10 healthy children (Baharav et al, 1995) and Villa et al in 17 children (Villa et al, 2000). To our knowledge this is the first study to systematically examine and establish these associations in a large population-based sample of children. These findings are especially relevant considering that the study participants were randomly selected from school-aged children and not from a clinic population that could have included large proportions of very ill children. Hence, these findings represent the burden of moderate SDB on the cardiac autonomic modulation in the population of healthy young children. We hypothesize that the impairment of autonomic modulation and its shift among children with moderate SDB may also lead to higher cardiac arrhythmia vulnerability, especially during REM sleep.

Longitudinal studies (Liao et al, 1996, Schroeder, et al, 2003) in adults have demonstrated that individuals with low HRV at baseline were at higher risk of developing hypertension. In this young children population, SDB, especially moderate SDB, is associated with higher systolic and diastolic blood pressures (Bixler et al. 2008). This finding, together with the current finding of impaired parasympathetic modulation shift across sleep stages among children with moderate SDB, suggests complex inter-relationships between SDB, autonomic modulation, and blood pressure. It is plausible that childhood SDB affects cardiac autonomic modulation, and through the impact on cardiac autonomic modulation, SDB adversely affects blood pressure. Alternatively, it is also plausible that SDB adversely affects both autonomic modulation and blood pressure, and the impaired sympathetic-parasympathetic modulation enhances the impact of SDB on blood pressure. Studies are needed to test these complex inter-relationships.

In this population-based sample of young children, the autonomic modulation shift across sleep stages within the mild SDB group are in a similar pattern and direction as that of the no-SDB group. However, and contrary to our hypothesis, we found that the mild SDB group had a statistically higher HF variable and lower LF/HF ratio than the no-SDB group in several sleep stages, indicated by p-values in p-3 columns. One possible explanation is that the lack of effects in the mild SDB group may be due to the fact that this is a population-based sample of healthy young children, and as a result, the severity and the duration of SDB may not yet be sufficient enough to impact the autonomic modulation among the mild-SDB group. Or, it is also possible that the mild-SDB group in this young and healthy population is shifted to a stage of more efficient autonomic modulation in order to counter for the adverse impact of SDB – thus, their autonomic modulation is shifted toward a better profile. Alternatively, previous studies have identified that primary snoring is associated with elevated blood pressure and altered arterial dispensability in similarly aged children (Kwok et al., 2003). In this study, we did not include primary snoring in the definition of SDB. This may also have explained the findings that the mild SDB did not display similar trends as did the moderate group. However, in this population, primary snoring was not associated with clinically relevant outcome (BP) (Bixler, 2008). Thus, we decided not to force primary snoring into the definition of SDB. In addition when we adjusted for snoring, the association between SDB and HRV did not change. Finally, when we compared HRV between children who did not snore and those with primary snoring; we did not find significant difference.

Results from population-based follow-up studies also suggest that lower HRV is associated with the risk of developing coronary heart disease (Dekker et al. 2000, Liao et al. 1997, 1999, 2002, Tsuji et al. 1996), and other cardiovascular related co-morbidity and risk factors (Liao et al 1996, 1997,1998, 2002, Carnethon et al. 2002, 2003, Schroeder et al 2003, 2005). HRV has since been proposed as a marker of vulnerability to cardiac arrhythmia and acute cardiac events (Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology 1996). Also in the study of adult patients with SDB, several studies have reported significant associations between SDB and lower HRV, indicating impaired cardiac autonomic modulation among persons with apnea (Guilleminault et al 1984, Penzel et al 2003, Roche et al 2003, Clifford & Tarassenko 2004).

There are several limitations in this study. The ECG data were collected digitally at a sampling rate of 100 Hz from one-ECG channel. At such a sampling frequency, it is possible that some degree of random misclassification of the R wave, resulting in non-differential misclassification of RR intervals may exist. However, it is worth repeating that we have established the internal validity of 100 Hz sampling data in HRV analysis, especially when it is calculated from at least 2 minutes or longer beat to beat RR data in the population-based Atherosclerosis Risk in Communities (ARIC) study, which systematically established the consistent associations between lower HRV, especially lower HF, and various cardiovascular disease outcome and co-morbidity, such as metabolic syndrome, diabetes, hypertension, levels of fasting glucose and insulin among non-diabetes, using HRV indices calculated from 100 Hz sampling rate from 2-minute beat-to-beat ECG data. (Dekker et al. 2000, Liao et al. 1996,1997, 1998, 1999, 2002, 2004, Carnethon et al. 2002, 2003, Schroeder et al 2003, 2005). Our current use of 5 minutes of beat-to-beat data should have sufficient length to overcome the relatively low sampling frequency. Current recommendations (Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology, 1996) also indicated that for short-term HRV analysis, HF component requires at least 1 minute RR data, and LF requires at least 2.5 minute data. Since the ECG data were collected from only one channel, we cannot make judgments/adjudications from other ECG channels about the presence of artifacts. Instead, we applied a set of statistical criteria (RR interval < 375 ms, RR interval > 1200 ms, or RR interval ratio from two adjacent RR intervals < 0.8 or > 1.2) to identify and remove such RR intervals from the analysis of HRV. This approach was at the expense of shortening the duration of RR interval data for HRV analysis. However, this limitation is offset by the major strengths of this study, which are large population, and long-duration of continuous ECG. For example, we successfully performed HRV analysis on 88% of the original PSCC sample (616 out of 700) based on at least 6.5 hours of artifact free RR interval data. To avoid major bias introduced by different duration of RR data in the final analysis, we performed statistical analysis only in those participants whose total RR interval data after artifacts elimination were 6.5 hours or longer. Another limitation is that the ECG data were collected from the sleep study at night. It is well recognized that there is a circadian shift of the cardiac autonomic modulation. Thus, the HRV profile at evening/night while sleeping may not be representative of day-time HRV file. However, this is offset by a supine, quiet, and no-interference data collection environment. Similarly, since we obtained first available 5-minute continuous RR from each sleep stage for stage-specific HRV analysis this may result in circadian differences in both within stage and between stage RR time frame. However, because of the very high sleep efficiency in these population-based sample young children, we were able to obtain the first available 5-minute stage-specific RR data from the first two sleep cycles. As a result, the stage-specific RR data we analyzed were obtained from a very similar time frame, for both the SDB and non SDB children. In additional, after adjusting for the clock time from which the 5-minute stage specific HRV variables were calculated, the pattern of association remain unchanged. In addition, our observation of decreased HRV among moderate SDB group was based a small sample size (N=7). Thus, these results must be interpreted with caution. On the other hand, this is one of the largest population-based sample of young children, and we have demonstrated that in this sample, moderate SDB group was consistently associated with higher blood pressure, higher BMI, and lower HRV. Therefore, we think the HRV and SDB associations we reported in this paper would be enhanced if we compared to a clinical sample, which often includes more severe and longer duration SDB individuals.

In summary, in this healthy population-based sample of K-5 school-aged children, moderate SDB as defined by AHI ≥ 5 / hour is significantly associated with lower sleep stage specific and wake HRV, especially during REM sleep. Furthermore, among healthy children, there is a significant shift of the sympathetic and parasympathetic balance in the direction of an increased parasympathetic modulation from wake to non-REM sleep, and a significant reversal to less parasympathetic modulation during REM sleep. However, in children with moderate-SDB group, the decrease in HF and RMSSD from SWS to REM are more pronounced. If these results are confirmed by other studies, these associations may suggest an increased risk of acute cardiac event and arrhythmia in persons with SDB, especially during REM-sleep, even before they reach the traditional “high risk age.” However, it should be noted that the lower HRV and higher risk of acute cardiac events literature were all based on adults, and no data to date have been published relating childhood HRV and the future development of acute cardiac events. Our interpretation of the impact of SDB related impairment on cardiac autonomic modulation in children is based on the premise of adult literature. Thus, it is possible that lower HRV in children may or may not represent higher risk of future acute cardiac events, which requires further study.