Cyclic alternating pattern in children with obstructive sleep apnea and its relationship with adenotonsillectomy, behavior, cognition, and quality of life

Abstract

Study Objectives

To determine in children with obstructive sleep apnea (OSA) the effect of adenotonsillectomy (AT) on the cyclic alternating pattern (CAP) and the relationship between CAP and behavioral, cognitive, and quality-of-life measures.

Methods

CAP parameters were analyzed in 365 overnight polysomnographic recordings of children with mild-to-moderate OSA enrolled in the Childhood Adenotonsillectomy Trial (CHAT), randomized to either early AT (eAT) or watchful waiting with supportive care (WWSC). We also analyzed CAP in a subgroup of 72 children with moderate OSA (apnea–hypopnea index > 10) that were part of the CHAT sample. Causal mediation analysis was performed to determine the independent effect of changes in CAP on selected outcome measures.

Results

At baseline, a higher number of A1 phases per hour of sleep was significantly associated with worse behavioral functioning (caregiver Behavior Rating Inventory of Executive Function (BRIEF) Global Executive Composite (GEC): ρ = 0.24, p = 0.042; caregiver Conners’ Rating Scale Global Index: ρ = 0.25, p = 0.036) and lower quality of life (OSA-18: ρ = 0.27, p = 0.022; PedsQL: ρ = −0.29, p = 0.015) in the subgroup of children with moderate OSA, but not across the entire sample. At 7-months follow-up, changes in CAP parameters were comparable between the eAT and WWSC arms. CAP changes did not account for significant proportions of variations in behavioral, cognitive, and quality-of-life performance measures at follow-up.

Conclusions

We show a significant association between the frequency of slow, high-amplitude waves with behavioral functioning, as well as the quality of life in children with moderate OSA. Early AT in children with mild-to-moderate OSA does not alter the microstructure of nonrapid eye movement sleep compared with watchful waiting after an approximately 7-month period of follow-up.

Clinical Trial

The study “A Randomized Controlled Study of Adenotonsillectomy for Children With Obstructive Sleep Apnea Syndrome” was registered at Clinicaltrials.gov (#NCT00560859).

Statement of Significance.

Obstructive sleep apnea (OSA) in children severely affects their behavior and most likely increases the risk of developing cardiovascular disease. The adverse effect of OSA on the sleep macrostructure of children has been extensively investigated, but its impact on nonrapid eye movement (NREM) sleep microstructure remains unclear. To ascertain the relationship between cyclic alternating pattern (CAP), characterizing sleep microstructure, and adenotonsillectomy (AT), behavior, cognition, and quality of life, we investigated 365 overnight polysomnographic recordings of children with mild-to-moderate OSA enrolled in the Childhood Adenotonsillectomy Trial (CHAT). Children with moderate OSA who experience a high frequency of slow, high-amplitude waves (A1 phases) display worse behavioral functioning and score lower in caregiver-rated quality of life. At 7-months follow-up, children who underwent AT showed no differences in CAP measures compared with children in the watchful waiting group, indicating that early AT does not yield an additional benefit in terms of NREM sleep instability; in both treatment arms, CAP rate, especially the A1 index, was increased at follow-up.

Introduction

Obstructive sleep apnea (OSA) is the most severe form of upper airway disease during sleep and found among 1% to 4% of children [1]. Compared with unaffected children, those with OSA are more likely to present with behavioral problems [2] and are at increased risk for developing cardiovascular and metabolic risk factors such as systemic hypertension [3] and higher levels of C-reactive protein (CRP) [4]. First-line treatment for childhood OSA is commonly adenotonsillectomy (AT) as enlarged tonsils and adenoids often result in narrowing the upper airway structure or pharyngeal collapse, leading to snoring and periods of apnea and hypopnea. Typically, apneas and hypopneas affect the quality and quantity of restorative sleep due to subsequent sleep stage transitions to lighter sleep, arousals, or periods of wakefulness.

By convention, sleep is scored by the rules of Rechtschaffen et al. [5] adopted and modified by the American Academy of Sleep Medicine (AASM) scoring manual [6]. In the AASM scoring, electroencephalography (EEG) arousals are defined as abrupt changes in EEG of a minimum duration of 3 seconds and do not consider 1 to 2-second activities observed in children [7], which consequently may limit the clinical utility of the conventional assessment of macrostructure in children. Cyclic alternating pattern (CAP) analysis is a method to assess the microstructure of sleep. It captures dynamic changes in EEG amplitude and frequency that recur periodically in nonrapid eye movement (NREM) stages [8]. Such sequences of recurring activation phases represent periods of high neural excitability with an intermittent background period in between [9]. As they coincide with physiological and pathological events, CAP provides insights into the fragmentation of NREM sleep not possible with tranditional sleep staging. Previous studies on the relationship between CAP and pathologies such as sleep disordered breathing (SDB), narcolepsy, and neuropsychological disabilities in pediatric populations concluded that children with these pathologies almost always show less synchronized slow-wave activity compared with healthy children [10].

The aim of this study was to investigate the relationship between NREM sleep microarchitecture and behavioral, cognitive, and quality-of-life measures in children with mild-to-moderate OSA and the effect of AT on these measures. We report data from the randomized controlled Childhood Adenotonsillectomy Trial (CHAT; ClinicalTrials.gov identifier: NCT00560859). Previous analysis on the CHAT study has shown a larger decrease in arousals, a larger decrease in the percentage of N1 sleep, and a greater increase in the percentage of N2 sleep but no change in N3 or rapid eye movement (REM) sleep following surgery compared with watchful waiting and supportive care [11]. It is currently unclear whether AT improves the sleep microstructure in children with OSA. To comprehensively probe NREM microstructure, we determined CAP in this sample. We also analyzed a subgroup of children enrolled in CHAT who presented with moderate OSA (apnea–hypopnea index [AHI] > 10) to test whether treatment results in a greater improvement in polysomnographic (PSG) findings and CAP parameters.

Methods

Definition and detection of the CAP

In agreement with the atlas and rules for scoring CAP published by Terzano et al. [8], we defined CAP as sequences of at least two consecutive CAP cycles. A CAP cycle consists of one activation phase (A phase) that represents transient, phasic events and an intervening background phase (B phase) that separates two successive A phases. Stereotypical A-phase patterns are delta bursts, vertex sharp transients, K-complex sequences, K-alpha, polyphasic bursts, intermittent alpha, and arousals [8]. We subdivided A phases into periods of slow high-amplitude waves (A1), fast low-amplitude EEG rhythms (A3), or a mixture of both (A2). A1 phases portray synchronized EEG patterns, while A3 phases represent desynchrony [9]. We defined A phases or B phases to last 2–60 seconds and restricted their occurrence to only NREM periods. Thus, REM periods and the periods between two CAP sequences were considered non-CAP. In this study, each CAP sequence was terminated by an A phase that was assigned to the following non-CAP period.

We analyzed CAP in overnight EEG recordings utilizing our previously developed, highly precise automated system for CAP analysis [12]. Our CAP detection system is a deep learning recurrent neural network that was trained with manually scored recordings from children. In the first step, the EEG channel is filtered and processed to remove powerline interference, noise, and cardiac field artifacts. Time and spectral features (Hjorth activity, Shannon entropy, Teager energy operator, band power descriptor, and differential EEG variance) were extracted from the processed signal and passed into the classifier as input values. The extracted features were sampled at 1 Hz, yielding a classification output that indicates whether the current second is part of an A phase or not and if so, what kind of A phase. We selected the F_β score as loss function for training to deal with the imbalanced dataset and increase the sensitivity and precision of the classification. Per the previously defined rules for CAP sequences, we post-processed the output of the A-phases detection system, i.e. isolated A phases and B phases, less than two CAP cycles, and the terminating A phases were removed.

We used 19 recordings of healthy children, 15 recordings of healthy adults, and 24 recordings of adults with sleep disorders as training set to cope with the inhomogeneous EEG characteristics of children and adults. Our system has previously demonstrated a second-by-second A-phase inter-rater reliability, quantified by the Cohen’s kappa coefficient, of 0.53 on a set of 16 healthy participants and 0.56 on a set of 30 participants with nocturnal frontal lobe epilepsy [13]. On the contrary, the event-based inter-rater reliability between human scorers ranges between 0.42 and 0.75 [14].

Here, we used the left and the right central EEG channels, re-referenced to the mastoid channels, in each PSG recording only counting overlapping A phases to increase the sensitivity of the classification. We defined the CAP rate as the percentage of NREM sleep that is covered by CAP sequences. Subtype indices represent the number of A1 and A2 + A3, respectively, per hour of NREM sleep. Subtypes A2 and A3 were merged into a single parameter due to their congruent nature. Additionally, the number, duration, and percentage of A1 and A2 + A3 phases were measured as well as the duration of B phases, CAP cycles, and CAP sequences.

Childhood Adenotonsillectomy Trial

We used overnight PSG from the CHAT, a multi-center, single-blind, randomized, controlled trial designed to analyze the efficacy of early AT (eAT) on children with mild-to-moderate OSA. The study tested whether children with mild-to-moderate OSA, randomized to eAT, demonstrate greater improvement in cognitive, behavioral, quality-of-life, and sleep measures at 7-months follow-up than children who were randomly assigned to watchful waiting with supportive care (WWSC) [11]. Children between 5.0 and 9.99 years of age diagnosed with OSA (OAI, number of obstructive apneas per hour of sleep, ≥1, or AHI ≥ 2), tonsillar hypertrophy ≥ 1, and cleared for surgery by an otolaryngologist were recruited in six clinical sites in the United States [15].

We removed recordings with fewer than 3 hours of good EEG quality and those from one specific clinical site (clusterid 13) due to equipment-related issues (n = 23). In consequence, 365 overnight recordings were evaluated (179 eAT and 186 WWSC). Also, we evaluated a subgroup of 72 children (38 eAT and 34 WWSC) with moderate SDB (AHI > 10) out of the 365 recordings as those children are likely to show a greater improvement in PSG findings [11] and CAP parameters. The dataset is available at the National Sleep Research Resource (NSRR) (available online at the National Sleep Research Resource; sleepdata.org) [16].

Outcome measures

The primary outcome measure of CHAT was the Attention/Executive (A/E) Functioning Domain Index from the Developmental Neuropsychological Assessment (NEPSY). Secondary outcome measures include indices of behavior, sleep symptoms, generic and disease-specific quality of life, PSG measures of sleep apnea, anthropometric measures, and blood pressure.

As the relationship between CAP and the child’s behavior and cognitive performance is of great interest [17, 18], we analyzed the causal mediation of changes in the A/E Functioning Domain Index and secondary cognitive and behavioral outcomes (Behavior Rating Inventory of Executive Function [BRIEF] Global Executive Composite [GEC] T score, and Conners’ Rating Scale Global Index T score) by variations in CAP parameters [15]. We also included the key quality-of-life measure in CHAT, the caregiver-rated total score from the Pediatric Quality of Life Inventory (PedsQL), as well as the disease-specific quality-of-life total score on the 18-item OSA-18 assessment tool. To assess symptoms of the OSA syndrome, we added the Pediatric Sleep Questionnaire sleep-related breathing disorder scale (PSQ-SRBD) to the list of outcome measures. Higher scores on the BRIEF GEC T score, Conners’ Rating Scale Global Index T score, OSA-18 score, and PSQ-SRBD scale indicate worse functioning, worse quality of life, or greater severity, respectively. On the contrary, higher scores on NEPSY A/E Functioning score and PedsQL caregiver-rated score represent better functioning and better quality of life, respectively. Finally, we included in the causal mediation analysis the change in AHI defined as the number of all obstructive and mixed apneas, plus hypopneas associated with either a ≥3% desaturation or electroencephalographic arousal, per hour of sleep, and the periodic limb movement sleep index (PLMSI) defined as the number of periodic limb movement (PLM) per hour of NREM.

Statistical methods

Statistical analysis was conducted using nonparametric tests assuming that CAP rate and subtype indices do not follow a normal distribution. The relationship between baseline measurements of CAP parameters and age, arousal index during NREM (AI-NREM), AHI, PLMSI, behavioral, cognitive, as well as quality-of-life performance measures was examined using the Spearman correlation coefficient. For each statistical test, the significance level was p < 0.05.

For the evaluation of treatment-specific changes in CAP parameters, including CAP rate, subtype indices, total number of subtypes, subtype percentages, mean duration of subtype phases, the average duration of B phases, the average duration of CAP cycles, and the average duration of CAP sequences, we applied two-factor repeated-measures analysis of variance (ANOVA) with time and treatment as factors and adjusting for the stratification factors of age, race, weight status, and study site.

We used the causal mediation analysis described by Imai et al. [19], to identify the independent effect of CAP changes on outcome measures (Figure 1). The model describes the total effect of treatment as a sum of the mediated effect using an independent mediator and the direct effect. Here, we analyzed three mediation models with CAP rate, A1 index, and A2 + A3 index, respectively, as mediators. We used a linear regression model where the pretreatment covariates were identical to those used in the repeated-measures ANOVA. We log-transformed the change of AHI due to its nonnormal distribution. We applied the mediation package for R with 1,000 nonparametric bootstrap resamples [20]. Results include the estimates of average causal mediation effects, the average direct effects, the total effects, and the proportion of the mediated effects with 95% confidence intervals and the associated p values.

Figure 1.

Causal mediation model to identify the independent effect of CAP changes on CHAT outcome measures. Causal mediation diagram illustrating the direct effect of treatment on the CHAT outcome (c′), the effect of treatment on ∆CAP as mediator (a), and the effect of the ∆CAP as mediator on the CHAT outcome (b). The product of the paths (a) and (b) equals the indirect effect on the CHAT outcome.

Results

Table 1 details the baseline characteristics for each outcome measure and their respective Spearman correlation value with CAP rate, A1 index, and A2 + A3 index for children with mild-to-moderate OSA (Table 1a) and the subgroup of children with moderate OSA (Table 1b).

Table 1.

Distributions and Spearman’s correlation between CAP parameters and age, PSG sleep disturbance indices, and behavioral, cognitive, and quality-of-life measures at baseline in (a) children with mild-to-moderate OSA and (b) children with moderate OSA (AHI > 10)

(a) Children with mild-to-moderate OSA (n = 365, 190 boys and 175 girls)
	Spearman correlation
	Baseline	CAP rate		A1 index		A2 + A3 index
	Median ± IQR	ρ	P	ρ	P	ρ	P
Age, years	6.0 ± 3.0	0.09	0.08	0.10	0.059	−0.02	0.69
BMI, kg/m2	17.2 ± 6.5	0.06	0.26	0.10	0.06	−0.14	<0.01*
AI-NREM, no./h	7.8 ± 4.4	0.00	0.99	−0.07	0.20	0.33	<0.001*
AHI, no./h	4.6 ± 6.2	0.03	0.54	0.01	0.79	0.04	0.42
PLMSI, no./h	0.6 ± 2.4	−0.09	0.082	−0.14	0.009*	0.15	0.003*
NEPSY Attention/Executive Functioning Scaled Score	102.0 ± 3.0	−0.06	0.27	−0.09	0.099	0.05	0.33
BRIEF Global Executive Composite Total T Score
Caregiver rating	49.0 ± 2.5	−0.04	0.39	−0.02	0.69	−0.05	0.31
Teacher rating	56.0 ± 17.0	−0.03	0.60	−0.02	0.70	−0.01	0.82
Conners’ Rating Scale Global Index Total T Score
Caregiver rating	50.0 ± 14.0	−0.04	0.49	−0.03	0.59	−0.03	0.61
Teacher rating	51.0 ± 17.0	−0.07	0.25	−0.04	0.53	−0.11	0.077
PedsQL caregiver-rated total score	81.7 ± 23.8	0.01	0.92	−0.01	0.86	0.06	0.22
PSQ-SRBD score	0.5 ± 0.26	−0.05	0.35	−0.03	0.54	−0.05	0.37
OSA-18 total score	51.0 ± 24.5	−0.01	0.82	−0.01	0.90	−0.01	0.78
(b) Children with moderate OSA (AHI >10) (n = 72, 40 girls and 32 boys)
		Spearman correlation
	Baseline	CAP rate		A1 index		A2+A3 index
	Median ± IQR	ρ	P	ρ	P	ρ	P
Age, years	6.0 ± 2.0	0.24	0.047*	0.28	0.018*	−0.29	0.014*
BMI, kg/m2	17.6 ± 6.9	0.22	0.06	0.26	0.03*	−0.18	0.14
AI-NREM, no./h	9.6 ± 5.6	0.04	0.74	0.00	0.97	0.35	<0.01*
AHI, no./h	15.0 ± 7.8	0.22	0.065	0.24	0.040*	−0.18	0.13
PLMSI, no./h	0.8 ± 2.9	−0.22	0.064	−0.27	0.021*	0.14	0.25
NEPSY Attention/Executive Functioning Scaled Score	98.0 ± 22.8	−0.17	0.14	−0.20	0.093	0.10	0.39
BRIEF Global Executive Composite Total T Score
Caregiver rating	47.0 ± 12.5	0.21	0.071	0.24	0.042*	−0.06	0.59
Teacher rating	57.5 ± 22.8	0.00	0.99	0.00	0.99	−0.10	0.47
Conners’ Rating Scale Global Index Total T Score
Caregiver rating	49.0 ± 16.0	0.23	0.051	0.25	0.036*	0.00	0.99
Teacher rating	52.0 ± 21.0	0.09	0.53	0.12	0.39	−0.15	0.28
PedsQL caregiver-rated total score	82.4 ± 29.4	−0.25	0.036*	−0.29	0.015*	0.11	0.34
PSQ-SRBD score	0.5 ± 0.3	0.07	0.56	0.12	0.32	−0.11	0.35
OSA-18 total score	56.0 ± 28.5	0.22	0.06	0.27	0.022*	−0.10	0.38

AI-NREM, Non-rapid eye movement sleep (NREM) Arousal Index; AHI, Obstructive Apnea-Hypopnea Index (AHI) >= 3% - number of [obstructive apneas] and [hypopneas with >=3% oxygen desaturation or arousal] per hour of sleep; PLMSI, Number of Periodic limb movement (PLM) per hour of non-rapid eye movement sleep (NREM); NEPSY, Developmental Neuropsychological Assessment (NEPSY); BRIEF, Behavior Rating Inventory of Executive Function; PedsQL, Pediatric Quality of Life Inventory; PSQ-SRBD, Pediatric Sleep Questionnaire sleep-related breathing disorder scale, OSA-18, Obstructive Sleep Apnea-18 assessment tool. *significance level: p < 0.05.

In children with mild-to-moderate OSA at baseline, AI-NREM significantly correlated with the A2 + A3 index (ρ = 0.33, p = <0.001). PLMSI showed significant negative correlations with the A1 index and positive correlation with the A2 + A3 index (A1 index: ρ = −0.14, p = 0.009; A2 + A3 index: ρ = 0.15, p = 0.003). Other outcome measures did not show any significant correlations with CAP rate and subtype indices, respectively.

In the baseline subgroup of children with AHI > 10, age was significantly correlated with the CAP rate and the A1 index (CAP rate: ρ = 0.24, p = 0.047; A1 index: ρ = 0.28, p = 0.018) and significantly inversely correlated with the A2 + A3 index (ρ = −0.29, p = 0.014). AI-NREM showed significant correlations with the A2 + A3 index (ρ = 0.35, p = <0.001). On the contrary, A1 index demonstrated significant correlations with AHI (ρ = 0.24, p = 0.040) and significant inverse correlations with PLMSI (ρ = −0.27, p = 0.021). Regarding cognitive, behavioral, and quality-of-life measures, the A1 index demonstrated significant correlations with the caregiver BRIEF GEC T score (ρ = 0.24, p = 0.042), the caregiver Conners’ Rating Scale Global Index T score (ρ = 0.25, p = 0.036), and the OSA-18 score (ρ = 0.27, p = 0.022). Finally, the PedsQL caregiver-rated score was significantly inversely correlated with the CAP rate and the A1 index (CAP rate: ρ = −0.25, p = 0.036; A1 index: ρ = −0.29, p = 0.015).

The effect of AT on CAP

Table 2 details the change of CAP parameters from baseline to follow-up for children in the eAT and WWSC arms for the entire CHAT sample. The median CAP rate increased in both treatment groups by 1%–4%. The eAT group demonstrated a marginally higher increase but the interaction between treatment and time was not significant (p = 0.37). Similar outcomes were observed for both subtype indices. The A1 index increased by 3.8% from baseline to follow-up PSG for children undergoing eAT, whereas children in the WWSC arm displayed an increase of 1.1%. Both groups showed a similar trend for the A2 + A3 index with an overall lower increase (eAT: 1.3%, WWSC: 0.1%). However, repeated-measures ANOVA indicated no significant interaction between treatment and time for either subtype index (A1 index: p = 0.58, A2 + A3 index: p = 0.24). In line with the results for CAP rate and subtype indices, other CAP parameters did not show any significant interactions between treatment and time either.

Table 2.

Comparison of CAP parameters between both randomized arms (eAT and WWS) at follow-up for children with mild-to-moderate OSA

CAP measures	Early AT (n = 179)		Watchful waiting (n = 186)		P †
	Baseline	Change from baseline to 7 months	Baseline	Change from baseline to 7 months	Treatment‡	Time§	Treatment * Time||
CAP rate, %	38.8 (±18.1)	3.5 (±19.7)	37.8 (±22.3)	1.2 (±16.9)	0.12	<0.01*	0.37
A1, no.	243.0 (±143.5)	21.0 (±148.5)	237.0 (±197.5)	11.0 (±130.8)	0.31	<0.01*	0.53
A2 + A3, no.	41.0 (±36.0)	7.0 (±35.5)	33.0 (±36.0)	0.5 (±30.5)	0.045*	0.068	0.28
A1, %	84.5 (±11.1)	−0.7 (±9.2)	86.2 (±12.4)	0.7 (±11.8)	0.22	0.69	0.31
A2 + A3, %	15.5 (±11.1)	0.0 (±0.1)	13.8 (±12.4)	0.0 (±0.1)	0.22	0.69	0.31
A1 mean duration, s	4.0 (±0.4)	0.0 (±0.5)	4.1 (±0.5)	0.0 (±0.5)	0.84	0.39	0.60
A2 + A3 mean duration, s	12.5 (±2.5)	0.2 (±3.1)	12.4 (±2.7)	0.1 (±3.3)	0.78	0.018*	0.52
A1 index, no./h	38.7 (±22.7)	3.8 (±23.0)	37.7 (±30.3)	1.1 (±19.8)	0.35	<0.01*	0.58
A2 + A3 index, no./h	6.4 (±5.6)	1.3 (±5.6)	5.3 (±6.2)	0.1 (±5.4)	0.036*	0.093	0.24
B duration, s	27.4 (±2.5)	−0.5 (±3.4)	27.3 (±3.4)	−0.5 (±3.4)	0.41	<0.01*	0.85
CAP cycle duration, s	30.8 (±2.4)	−0.4 (±3.3)	30.8 (±3.2)	−0.5 (±3.0)	0.40	<0.01*	0.79
CAP sequence duration, s	184.6 (±59.5)	13.9 (±73.5)	181.9 (±79.2)	3.8 (±66.8)	0.25	0.044*	0.19

^†Repeated measures analysis of variance (ANOVA) adjusting for the stratification factors of age (5–7 vs. 8–9 years old), race (African American vs. other), weight status (overweight/obese vs. non-overweight), and study site.

^‡Effect of treatment (eAT vs. watchful waiting) on CAP measures after controlling for the effect of time (baseline vs. follow-up).

^§Effect of time (baseline vs. follow-up) on CAP measures after controlling for the effect of treatment (eAT vs. watchful waiting).

^||Effect of the treatment * time interaction on CAP measures.

*Significance level: p < 0.05.

Table 3 lists the change of CAP parameters from baseline to follow-up PSG for children in the eAT and WWSC arms in the subgroup of children with AHI > 10. CAP rate was increased in both treatment groups by 1%–2% with no significant interaction between treatment and time (p = 0.84). The A1 index showed only for the WWSC group an increase of 1%. On the contrary, the A2 + A3 index was only in the eAT group decreased by 1%. Repeated-measures ANOVA of both subtype indices demonstrated no significant interaction between treatment and time (A1 index: p = 0.94, A2 + A3 index: p = 0.31). In line with the results for CAP rate and subtype indices, other CAP parameters did not show any significant interaction between treatment and time either.

Table 3.

Comparison of CAP parameters between both randomized arms (eAT and WWS) at follow-up for children with moderate OSA (AHI > 10)

CAP measures	Early AT (n = 38)		Watchful Waiting (n = 34)		P †
	Baseline	Change from baseline to 7 months	Baseline	Change from baseline to 7 months	Treatment‡	Time§	Treatment * Time||
CAP rate, %	38.7 (±17.0)	1.3 (±17.8)	37.9 (±15.3)	1.8 (±22.5)	0.77	0.30	0.84
A1, no.	262.0 (±155.0)	−2.0 (±180.8)	240.5 (±152.0)	5.0 (±179.3)	0.79	0.25	0.95
A2 + A3, no.	44.5 (±41.3)	−4.0 (±43.0)	37.0 (±23.8)	−0.5 (±25.5)	0.96	0.99	0.29
A1, %	83.9 (±13.0)	0.2 (±10.9)	85.2 (±15.7)	1.8 (±12.4)	0.86	0.27	0.94
A2 + A3, %	16.1 (±13.0)	−0.2 (±10.9)	14.8 (±15.7)	−1.8 (±12.4)	0.86	0.27	0.94
A1 mean duration, s	4.1 (±0.3)	0.1 (±0.4)	4.1 (±0.6)	0.0 (±0.4)	0.73	0.48	0.67
A2 + A3 mean duration, s	12.5 (±3.0)	0.6 (±3.2)	12.4 (±2.4)	0.5 (±2.4)	0.083	0.068	0.48
A1 index, no./h	40.7 (±25.5)	−0.1 (±26.1)	36.8 (±23.4)	0.8 (±23.9)	0.83	0.23	0.94
A2 + A3 index, no./h	7.0 (±6.3)	−0.7 (±6.7)	5.9 (±4.1)	−0.2 (±4.5)	0.95	0.87	0.31
B duration, s	26.7 (±2.2)	−0.1 (±2.7)	27.7 (±3.1)	−0.5 (±3.9)	0.88	0.51	0.37
CAP cycle duration, s	30.3 (±2.0)	−0.1 (±2.9)	30.8 (±3.0)	−0.5 (±3.5)	0.81	0.49	0.33
CAP sequence duration, s	195.4 (±64.4)	0.3 (±70.7)	180.4 (±69.4)	8.3 (±57.5)	0.62	0.50	0.72

^†Repeated measures analysis of variance (ANOVA) adjusting for the stratification factors of age (5–7 vs. 8–9 years old), race (African American vs. other), weight status (overweight/obese vs. non-overweight), and study site.

^‡Effect of treatment (eAT vs. watchful waiting) on CAP measures after controlling for the effect of time (baseline vs. follow-up).

^§Effect of time (baseline vs. follow-up) on CAP measures after controlling for the effect of treatment (eAT vs. watchful waiting).

^||Effect of the treatment * time interaction on CAP measures.

The effect of CAP changes on behavior, cognitive performance, and quality of life

Supplementary Table S1 displays the results of the mediation analysis with CAP rate as the mediator in the entire study sample. The total effects of treatment were significant for Conners’ Rating Scale Global Index scores, the caregiver BRIEF GEC score, the PedsQL caregiver-rated total score, the PSQ-SRBD scale, the OSA-18 score, and the AHI. No significant average mediation effects were identified for CAP rate. Similarly, neither A1 index nor A2 + A3 index showed a significant average mediation effect (Supplementary Tables S2 and S3). We obtained identical outcomes in the subgroup of children with moderate OSA. The results of the mediation analysis with CAP rate and subtype indices as the mediator for children with AHI > 10 are presented in Supplementary Tables S4–S6.

Discussion

We show that children with moderate OSA (AHI > 10) demonstrate a significant association between a higher frequency of slow high-amplitude rhythms, the so-called A1 phases, and worse behavioral functioning and lower quality of life rated by their caregivers at baseline. However, CAP rate and both subtype indices did not account for significant proportions of changes in behavioral, cognitive, and quality-of-life performance measures plus changes in AHI and PLMSI at follow-up. This outcome may be explained by the negligible changes in CAP parameters at follow-up within children with moderate OSA. Considering the entire CHAT sample, we demonstrate that children with mild-to-moderate OSA display elevated CAP rates, specifically more frequent A1 phases, at 7-months follow-up. However, this increase is independent of the treatment as no significant interaction between treatment and time was found.

At baseline, children with moderate OSA demonstrated a positive correlation between A1 index and AHI. Previous studies in children with SDB have shown opposing findings, i.e. a decrease of A1 subtypes, mainly during N3 sleep compared with children without SDB [21, 22]. Our results may reflect the fact that children often do not respond to apneas with a noticeable EEG arousal [23] but instead with a short 1 to 3-second burst primarily in the theta band [24]. This is in contrast to adults who terminate obstructive apneas regularly with an arousal stimulus [25]. In a previous study, children with SDB demonstrated a positive correlation between the intelligence quotient and the percentage of A2, and the A2 index, respectively, suggesting that respiratory events in children with SDB elicit arousal-like protective responses to preserve neurodevelopment to the detriment of NREM sleep instability [26]. The higher age of children in that study (9.1 ± 2.3 years) may explain the shift from an increase in A1 phases to more arousal-like A2 phases. The authors also reported a positive correlation between the A1 index during N3 and mean overnight oxygen saturation, indicating an increased occurrence of A1 phases in response to respiratory events. In summary, children tend to react to internal or external disturbing stimuli with a protective mechanism that minimizes the negative effects on neurodevelopment [27], which may potentially explain the increase in slow high-amplitude rhythms defined as A1 phases in our study.

Children with moderate OSA also demonstrated a significant association between the A1 index and the caregiver BRIEF GEC T score, the caregiver Conners’ Rating Scale Global Index T score, the OSA-18 score, and the caregiver-rated PedsQL. The correlation between the number of A1 phases per hour of sleep and the disease-specific quality-of-life OSA-18 score is most likely a result of the significant association between A1 index and AHI. Bruni et al. [28] have previously reported that a high frequency of A1 phases is significantly associated with poor behavior in children with Asperger syndrome. Our results corroborate these findings by revealing a significant association between poorer behavioral functioning, i.e. higher scores in the caregiver BRIEF GEC T score and the caregiver Conners’ Rating Scale Global Index T score and more frequent A1 phases. Furthermore, we demonstrate a significant association between lower children’s quality of life rated by their caregivers and a higher A1 index. Potentially, these results may reflect an indirect effect of the significant correlations between the A1 index and AHI and PLMSI, respectively, in children with moderate OSA.

The CAP rate results obtained with our automated system are in line with previously reported values scored manually in children with SDB. Both treatment arms demonstrated a median CAP rate of 38%–39% during the baseline visit, congruent to the values of children with disordered sleep breathing of the same age range reported by Kheirandish-Gozal et al. [21]. The increase in CAP rate and A1 phases observed in the CHAT sample at 7-months follow-up agrees with the previously described pattern of A-phase occurrence during childhood development. The CAP rate increases from preschool age (3–6 years) to school age (6–10 years), peaking during peripubertal age at a previously reported value of around 60% and subsequently decreases during young adulthood [29–32]. The subgroup of children with moderate OSA did not show such an increase in CAP rate, especially the number of A1 phases did not change. One can speculate that an increase in CAP sequences due to maturation compensated the expected decrease in the number of CAP sequences due to AHI normalization.

The comparison between treatment arms did not show any significant difference in CAP rate between children undergoing surgery and watchful waiting. A previous study on the treatment of OSA in children with rapid maxillary expansion indicated an increase in CAP rate during slow-wave sleep that was associated with a rise in A1 phases per hour of sleep [33]. However, follow-up PSG were only recorded for children in the treatment group, preventing the assessment of the time effect and developmental changes in CAP in children without intervention. Hence, those results may reflect the overall increase of CAP sequences and A1 phases with age rather than the effect of treatment, which is in line with our findings. In CHAT, normalization in PSG findings (defined as AHI < 2 or an obstructive apnea index score of <1 event per hour) was found not only in 79% of the children after AT but also in 46% of the children randomized to WWSC [11]. Moreover, the CHAT study reported no significant improvement in cognitive functioning for children after AT compared with children after WWSC. This result is in agreement with other studies that reported no significant improvement in cognitive functioning after AT [34, 35] as neurocognition in children is strongly linked to other biomarkers such as the level of CRP [36] or urinary neurotransmitters [37]. The marginal difference in AHI normalization between treatment arms and the lack of improvement in cognitive functioning most likely explains the lack of improvement in CAP parameters after AT compared with watchful waiting.

We also investigated the contribution of changes in CAP to changes in neurocognitive, behavioral, and quality-of-life measures. Although we observed a significant association between the A1 index and behavioral and quality-of-life measures in children with moderate OSA, mediation analysis does not suggest that the changes in cognitive, behavioral, and quality-of-life performance scores due to treatment are attributable to changes in CAP. This observation is supported by the negligible changes in CAP parameters between baseline and follow-up in children with moderate OSA. When considering the entire CHAT sample, mediation analysis yielded an identical outcome, probably due to the predominance of mild cases.

Our study is limited by the original CHAT study design that constraints enrollment to children with an OAI ≥ 1 or AHI ≥ 2. A control group of normal children would enable comparative analyses and ranking of CAP parameters. Furthermore, the time span between baseline and follow-up is relatively short potentially limiting the effect of surgery on CAP parameters. Another limitation is the accuracy of our developed automated detection system. Although it has already shown strong performance [12] and high reproducibility [13], the performance for subtype detection is limited by the low number of manually scored pediatric PSG available for machine learning.

In conclusion, we show a significant association between the frequency of slow, high-amplitude waves and the behavioral functioning, as well as the quality of life in children with moderate OSA. Early AT in children with mild-to-moderate OSA does not alter the microstructure of NREM sleep compared with watchful waiting.

Funding

The CHAT study was supported by grants (HL083075, HL083129, UL1-RR-024134, and UL1 RR024989) from the National Institutes of Health. S.R. was partially supported by NIH R35 HL135818. Data access was supported by the National Sleep Research Resource (R24 HL114473) and contract NHLBI (75N92019C00011).

Conflict of interest statement. None declared.