Attention in children with obstructive sleep apnoea: An event-related potentials study

Abstract

Background:

Obstructive sleep apnoea (OSA) in children has been causally implicated in neurobehavioural and cognitive dysfunction. Consequently, the American Academy of Paediatrics highlighted the need to study pertinent functional cognitive outcomes before and after treatment. However, neurocognitive function has thus far only been assessed by caregiver-completed questionnaires, which can be labour intensive and time consuming, such that the need for complementary and more objective measures has emerged.

Objective:

The study aimed to identify electrophysiological correlates of neurocognitive alterations in children with OSA and investigate utility as a predictive tool.

Patients and methods:

Twenty-eight children (14 OSA and 14 matched controls) underwent overnight sleep studies and neurocognitive testing, as well as the oddball attention task while event-related potentials (ERPs) were recorded. ERPs were analysed using temporal principal components and source analyses that provided dependent variables for the subsequent repeated measures analysis of variance (ANOVA) and multiple regressions. Time-locked waveforms formed spatial models that localised electrical activity in the brain. ERP differences between groups were then correlated to neurocognitive outcomes.

Results:

OSA children exhibited significantly altered ERP patterns of neural activation and impaired neurocognitive performance. Specific ERP variables exhibited accurate predictive ability regarding performance on neurobehavioural measures.

Conclusion:

Specific ERP events during a single attention task can reliably identify the presence of OSA-associated cognitive dysfunction. Electrophysiological approaches during specific cognitive tasks may serve as simple, complementary, and reliable reporters of cognitive dysfunction associated with OSA in children.

1. Introduction

Obstructive sleep apnoea (OSA) is a frequent condition, affecting 1–3% of the children [1–6], which can lead to neurocognitive impairments [7,1], such as restlessness, aggression, excessive daytime sleepiness, and poor school performance [2,3]. Furthermore, associations between OSA and attention deficit–hyperactivity disorder (ADHD) have been proposed [4,5]. Increased brain vulnerability to OSA during childhood and risk for persistent damage if treatment is delayed have been recognised [1,6,7]. However, not every child with moderate to severe OSA will develop evidence of cognitive dysfunction, and vice versa [8–17].

‘Attention’ refers to a variety of phenomena, including concentration, selectivity, automaticity, conscious monitoring, and capacity limitations [9]. Current studies of attention distinguish between selectively attending to perceptual information and higher-order centralised, executive function concerned with planning and organising complex actions [10]. Growing evidence suggests causal association between OSA, inattention [11–14], and hyperactivity [15,16]. However, most children with ADHD do not have OSA, and most children with OSA do not have ADHD [17].

Measures of neurocognitive function are based on administration of neurocognitive standardised batteries, which can be labour intensive for children, or on the administration of caregiver-completed questionnaires, providing a second-hand, subjective report. A more objective measure of neurocognitive function is necessary to enable risk and vulnerability stratification during initial evaluation of snoring and to assess cognitive outcomes following treatment. Ideally, this measure would be non-invasive and time efficient relative to polysomnography (NPSG) and provide more objective measures of impairment than those obtained from parental reports, such as an electrophysiological measure [18].

Event-related potential (ERP) recordings are non-invasive and are used across the life span to investigate brain-based correlates of cognition [19]. They provide an ‘exquisitely sensitive’ measure of cognitive processing during wakefulness which is free of constraints of motor activity or the need for the participant to consciously attend to the task at hand [20], making them ideal for use with children. High-density net arrays are well tolerated by young children, providing temporal resolution of electrical activity at the millisecond level [21–24], while providing excellent spatial resolution of dipoles [25]. ERP is a portion of the continuous electroencephalogram (EEG) that is time-locked to stimulus onset. Multiple presentations of stimuli are averaged, removing random and non-stimulus-related activity. Amplitude of certain peaks, or components, in the brain wave and latency of the wave provide measures of brain activity [26]. For an overview of signal averaging theory in ERP research and its applications in sleep research, see Colrain’s and Campbell’s review [20].

ERP patterns strongly correlate with learning, reading and school performance – areas of daytime functioning known to be impaired in apnoeic children- and has proven highly successful in ADHD [27,28]. ERP components such as the P300 have been consistently shown to be reduced in amplitude and of an increasing latency with impending sleep and in conditions of total sleep deprivation. Unfortunately, the ERP findings from studies of adults with sleep fragmentation and OSA have been less straightforward [20].

A 2006 study of 12 adult patients with OSA and 12 age-, sex-, and education level-matched controls revealed a delayed and sustained P300 (300–700 ms) in the OSA group relative to the controls [29]. Stimuli consisted of pure tones in a classic oddball paradigm (90% standard tones and 10% deviant). Subjects with OSA did have longer reaction times than their non-OSA counterparts, although this difference was not significant. In general, the control subjects exhibited an earlier and ‘sharper’ P300 compared with those in the OSA group. The OSA group had a sustained P300 with a large amplitude, suggesting that, although the attention resources allocated were sufficient to generate a correct and timely response, a greater amount of attention resources were needed for a longer amount of time compared with controls. There was also a significant between-groups difference in the P3a, which was a component obtained by subtracting responses to standard stimuli from that of deviant stimuli. On this measure, control participants exhibited a larger amplitude of the P3a, which occurred earlier than that in the OSA patients. These results are consistent with previous reports of a delayed and sustained P300 in apnoeic adults, suggesting that the volitional attention resources allocated to the processing of novel stimuli are sufficient in subjects with OSA, but more effortful and delayed relative to controls. The decreased amplitude of the P3a could be the result of a decrease in overall attention resources available for the task at hand. This is in line with the theory that OSA decreases the cognitive reserve available for such tasks. Gosselin further posits that this increased latency in the OSA group could reflect an alteration in P3a generators, possibly switching from the typically strong prefrontal generators to other brain areas such as the cingulate, auditory, and parietal cortices [29]. This is consistent with the literature discussed above, suggesting that the prefrontal cortex is especially susceptible to sleep fragmentation and hypoxia associated with OSA. It is important to note that ERP differences could distinguish between OSA patients and controls, while behaviourally observable phenomena could not.

In another 2006 study from Gosselin’s group, 13 adults with OSA and 13 controls participated in a passive auditory oddball task while reading [30]. They found that the individuals with OSA exhibited a decreased percentage of rapid-eye-movement (REM) sleep; increased body mass index (BMI); increased sleepiness as assessed by the Epworth sleepiness scale [31]; an increased number of arousals and sleep transitions; and decreased P3a amplitudes compared with controls. The authors concluded that this abnormal P3a amplitude reflected ‘involuntary attention switching’ but relatively preserved automatic processing. The blunted amplitude probably represents the reduced cognitive resources available in patients with OSA for the attention task. Increased hypoxaemia and decreased sleep efficiency were correlated to decreased amplitude of the P3a in frontocentral, central, and parietal electrode sites. However, citing the few ERP studies of OSA in existence, the authors are quick to point out that further research is needed [30]. Furthermore, this study did not use high-density arrays of electrodes. These differences in reports regarding ERPs in patients with OSA could be the result of differences in task demands or protocol or in the components of the EEG that are being isolated for study. Nonetheless, ERP recording is a method that can be used in evaluating the neurocognitive changes associated with OSA and the efficacy of adenotonsillectomy (AT) [32].

We hypothesised that assessment of ERPs using a high-density array during an oddball attention task provides objective and complementary evidence of impaired attention in apnoeic children aged 4–8 years. The theoretical model under study is that children experiencing insufficient or disrupted sleep (as in OSA) are less effective in organising cognitions. Therefore, the activation profiles observed in such children would be more diffuse and poorly localised compared with control counterparts. We hypothesised the performance impacted by OSA can be indexed by increased latencies and decreased amplitudes of specific ERP components and linked with measures of attention, inhibition, and executive functioning. We chose to test the predictive power of identified ERP patterns in apnoeic and control children using the attention and executive functioning and Memory and Learning subscales of the A Developmental NEuroPSYchological Assessment (NEPSY), as these are areas which have been of high interest in the field of paediatric sleep-disordered breathing (SDB) and impairments in attention and attentional switching are most prominent in ERP studies of adults with OSA [20].

2. Patients and methods

2.1. Participants

A total of 16 children were originally included in the suspected OSA group; however, two were excluded from the final analyses because they had too many eye-movement artefacts in their electrophysiological data. It should be noted that no families in the Sleep Centre declined to participate secondary to use of the high-density electrode array, and the ERP method was tolerated by all participants. The final sample included 14 children (nine females), 4–8 years of age, evaluated for suspected OSA in the Paediatric Sleep Medicine Centre and their 14 matched controls, for a total of 28 children. A control population of 14 age- and sex-matched children was recruited from the Jefferson County Public School system as part of two National Institutes of Health (NIH)-supported Projects (DC005994, HL070911). The enrolled sample was representative of the clinical population to which the study is intended to generalise (Paediatric OSA in Louisville, KY, USA) for socioeconomic status (SES), sex, and ethnicity. The sleep clinic served the population from the same Jefferson County Public Schools. Parental informed consent and child assent were obtained, and the study was approved by the University’s Institutional Review Board.

Inclusion criteria for the OSA group were suspected OSA at physician office visit and on Sleep Behaviour Questionnaire (SBQ); confirmation of OSA by NPSG; and age between 4 and 8 years. Inclusion criteria for the age-matched control group required: absence of OSA and snoring determined by SBQ and NPSG. Exclusion criteria for both groups included: pre-existing medical, behavioural, or learning disorders indicated by screening questionnaires; medications within 3 days of protocol; BMI greater than 95th percentile; failure to pass hearing and vision screening; failure to pass IQ screening (using Peabody Picture Vocabulary Test-III (PPVT-III) as proxy); or significant ‘at-risk’ scores on Childhood Symptom Inventory-4 (CSI-4) or Child Behaviour Checklist (CBCL). The first ERP session was performed the morning after the initial NPSG, whenever possible. However, when scheduling did not permit, the ERP session was performed in the morning as soon as possible after the initial NPSG. The second ERP session was performed 6 months after the first one.

2.2. PSG procedures

Overnight sleep studies were performed and scored, as previously reported [33]. The severity of OSAS was scored as none (0); none with mild findings (1) – apnoeic episodes less than 1 per hour associated with O₂ desaturation, and/or mild hypoventilation, less than 10 percent total sleep time (TST) with PetCO₂ greater than 50 mmHg – mild (2) apnoea hypopnea index (AHI) greater than 2 but less than 5 per hour TST; moderate (3) AHI between 5 and 10 per hour; or severe (4) AHI greater than 10 per hour TST.

2.3. Behavioural procedures

Behavioural questionnaires (SBQ and Neuropsychological Screening Form) were used to obtain demographic information and medical and family history. SES index was calculated using parental income, education level, and occupation with a version of Hollingshead’s index. Proxy for Intelligence Quotient (IQ) was obtained using the PPVT-III [34]. Edinburgh Handedness Inventory assessed hand preference [35]. Parental reports and neuropsychological assessments (CSI-4, CBCL, and NEPSY) screened for behavioural disorders and characterised the neurocognitive status of both groups. The CBCL served as a parental report of behaviour while the NEPSY provided a standardised tool that has been used in several studies of paediatric OSA.

2.3.1. Sleep Behaviour Questionnaire

Parents of all patients completed a questionnaire about sleep problems and practices, including bedtime routines and sleep disturbances. In previous studies from our centre, the SBQ had high sensitivity and specificity for snoring and non-snoring children in this age range [36].

2.3.2. CSI-4 [37]

The CSI-4 parent checklist screened for behavioural, emotional, and cognitive disorders based on parent-reported symptoms outlined in the Diagnostic and Statistical Manual of Mental Disorders-4 (DSM-IV). Internal consistency of the CSI-4 was high, with Chronbach’s α values from 0.94 to 0.74. The CSI-4 was used for children 5–8 years of age. Younger children were given the age-appropriate Early Childhood Inventory-4 (ECI-4).

2.3.3. PPVT-III [34]

The PPVT-III was used as an individually administered IQ-screening tool in which children selected one picture of four choices that best corresponded to the word the tester read out. The assessment has high construct validity. Correlations between PPVT-III standard scores and Wechsler Intelligence Scale for Children-Third Edition range from 0.82 to 0.92. Administration lasted approximately 15 min.

2.3.4. NEPSY [38]

The NEPSY is a battery with five subtests, providing assessment of neuropsychological functioning in children 3–12 years of age. Core assessments for children aged 4–8 years were administered for attention/executive function and Memory and Learning Domains. Internal consistency of the NEPSY was high with r values ranging from 0.71 to 0.91 for all domains across the age range included in this study. Administration lasted 60 min.

2.3.5. Child Behaviour Checklist [39]

The CBCL screened for clinical behaviour disorders by parental report of DSM-IV symptoms using two forms: one for children aged 6–18 years and one for preschoolers. Higher scores indicate more problems.

2.4. ERP apparatus

The Electrical Geodesics Inc. (Eugene, OR, USA) high-density system recorded ERP activity. The electrode net contained 128 Ag/AgCl electrodes inside sponges. NetStation software controlled impedances (under 40 kohms), baseline correction analogue/digital sampling rate, artefact rejection, and averaging. A second computer presented stimuli using E-prime 2.0, integrating this information with ERP recording software. Interstimulus intervals varied randomly (1.5 to 2.5 s).

2.4.1. Testing methods

Children were tested individually (Fig. 1) in sound-dampened rooms. Screening and behavioural tests were administered. The child’s head was measured and marked for proper net placement. Net application and impedance adjustments required less than 5 min. ERP signals were digitised at 4-ms intervals for a 1.5-s period. Filter settings remained at 0.1 Hz for high pass and 30 Hz for low pass, with a gain of 10 K.

Fig. 1.

Photograph of 5-year-old participant in 128-electrode net (with permission from legal caretaker).

2.4.2. Electrophysiological task

The oddball or P300 task is a classic attention task in the ERP field [40]. The amplitude of this positive peak occurring about 300 ms after stimulus onset is larger in response to infrequent events. In children, it typically occurs just after 400 ms [19,41,42]. The P300 (subtype P3b) is the most widely studied ERP component. We investigated how ERP components generated during the task correlate with attention domains of neurobehavioural assessments. Two single frequency tones between 1000 and 4000 Hz served as stimuli. Tones presented differed by 500 Hz. Rise and decay times were controlled. Stimulus duration was 300 ms. Auditory stimuli matched in loudness levels at 75 dB sound pressure level (SPL) (A) and presented through a speaker 1 m over the child’s vertex. Tones occurred in random order. One occurred in 70% of trials while the other (‘oddball’) occurred in 30%. The test included 120 tones over a 5-min period. Testers asked participants to attend to the target and press a button on hearing it.

2.5. Analyses

Independent samples t-tests were used to compare group differences. Variables on which the two groups significantly differed were related to ERP data using stepwise multiple regression to determine if ERP differences could predict behaviour. The null hypothesis that μ_OSA = μ_CONTROL for demographic, behavioural, and sleep variables was tested using multiple un-paired t-tests and Tukey’s Honestly Significant Difference test with Tukey’s studentised q used to accurately handle family-wise error correction for multiple comparisons.

2.5.1. A priori power and effect size calculations

To increase statistical power while keeping the number of dependent variables reasonable, the 128 electrodes were grouped into five regions for each hemisphere based on anatomical boundaries. There were equal numbers of males and females in two groups (OSA/control), and 10 electrode regions (frontal, central, temporal, parietal, and occipital means for each hemisphere) and only two conditions (frequent/infrequent). The sleep variables (AHI, arousal index, lowest O₂, etc.) were analysed separately and correlated separately with ERP amplitudes and latencies. Cohen and Cohen [43] provided data on effect sizes for correlation coefficients. For α = 0.05, two-tailed t-test and a desired power of 0.80, 28 subjects were required if the average correlation among variables was 0.50. For any single correlation, a power of 0.83 with a set at 0.05 was achieved with 30 subjects (Table F.2 in Cohen and Cohen). Table 4.7 in Stevens showed that for a D² = 0.64 (measure of effect size derived from Hotelling T²), α = 0.05, seven dependent measures and 25 subjects per group were required for a power of 0.82 [44]. Data from our laboratory from a normal control population of 48 children performing similar tasks showed a power of almost 1.00 with an eta of 0.80 for peaks’ × amplitudes as dependent and group × electrode sites as independent variables. This suggested that the power analysis of large effect sizes was reasonable for 48 subjects. Investigators used an α level of 0.05 to determine significance for all statistical results across experiments.

2.5.2. ERP data initial processing

Single trial ERPs were screened for artefacts, excluding trials with eye channel differences exceeding 70 μV or more than 5% bad channels. Following artefact screening, EEGs were baseline-corrected using the 100 ms prestimulus average as baseline. ERPs were adjusted using 30 Hz digital low-pass filters and average reference. Trials were averaged for each subject separately [45,46]. Trials with incorrect responses were not omitted from analyses. The minimum amount of artefact-free trials permissible for inclusion in analyses was 65% of the total task.

Data were analysed by a mixed linear model in Statistical Package for Social Sciences (SPSS) 11 for Mac OSX (SPSS Inc., Chicago, IL, USA). Condition differences were retested for separate effects by t and F tests and a priori tests considering hypotheses. Principal components analysis (PCA)/Varimax procedure was blind to individual experimental conditions and generated the same solution regardless of data entry order. Once PCA identified main components, repeated measures analysis of variance (RMA) used factor scores for each component to identify sources of variance. PCA has less variance misallocation than ‘peak-picking’ with topography taken into account [47]. Factor loadings from PCA served as dependent variables in the mixed factorial ANOVA. Significant group differences were recorded. Interactions between group and within-subjects variables were investigated with post hoc tests, including Tukey’s Honestly Significant Difference test.

2.5.3. Source localisation

Scalp potential source estimates were modelled using Geo-Source software (Electrical Geodesics Inc., Eugene, OR, USA). Investigators used a minimum-norm least squares solution and finite difference model with local autoregressive average (LAURA) constraint [48] to derive sources based on Montreal Neurological Institute probabilistic magnetic resonance imaging (MRI). Settings included weighting per location, truncated singular value decomposition regularisation at 10⁻³, and radius of influence at 12.2 mm [49]. To compute source estimates, investigators focussed on time points of maximal amplitude peaks from temporal PCA.

As with any source localisation procedure, the proposed dipole solution is only one of a theoretically infinite number of solutions. Caution is used in the interpretation of PCA and source localisation, and statistical constraints such as Varimax rotation are artificial to a degree, because electrical activity recorded at the scalp is the result of multiple complex generators. However, in an attempt to isolate and measure components of such complex electrical patterns so that they can be related to observable behaviour and clinical isolates, we have employed principles and assumptions that are generally accepted in electrophysiological research in general and in the field of ERP recordings specifically, including confirming the robust nature of our results through confirmatory analyses. Furthermore, the results of the ANOVA support our assertion that this specific PCA solution maps onto specific experimental dimensions.

3. Results

3.1. Demographic and behavioural

There were no demographic differences between groups, including SES, BMI, or PPVT (Table 1). Significant differences existed for behavioural assessment scores (Table 2). Control children scored higher than apnoeic children on the NEPSY Visual Attention subtest (t = −2.125, p = 0.048) and Memory and Learning percen tile scores (t = −2.126, p = 0.043). Significant parent-reported group differences were found on many CBCL scales. The OSA group scored higher on all measures.

Table 1.

Demographic and descriptive variables for OSA and control children.

Variable	OSA groupa	Control groupa
Age (years)	6.21 ± 1.81	6.20 ± 1.31
Gender (number, %)
Male	5, 36%	5, 36%
Female	9, 64%	9, 64%
BMI (kg/m2)	19.36 ± 4.79	17.05 ± 2.47
PPVT-III
Raw score	74.57 ± 20.76	85.46 ± 23.74
Standard score	96.36 ± 12.11	104.77 ± 13.14
Handedness	0.89 ± 0.33	0.80 ± 0.42
SES index	9.998 ± 2.834	10.66 ± 3.744
Racial category (number, %)
American-Indian	1, 7%	0, 0%
Black or African-American	2, 14%	3, 21%
White or Caucasian	7, 50%	11, 78%
More than once race	0, 0%	0, 0%
Unknown or not reported	4, 28%	0, 0%

^aBoth groups have n = 14.

Table 2.

Behavioural variables for OSA and control children.

Variable	OSA groupa	Control groupa
CBCL scales
Somatic complaints-T	58.62 ± 10.26*,†	51.85 ±3.05*,†
Internalising-T	56.85 ± 15.67*,†	43.69 ± 8.90*,†
Externalising-T	51.77 ± 12.60*,†	41.38 ±10.14*,†
Total problems-T	57.08 ± 13.73**,†	41.85 ±9.54**,†
Affective problems-T	61.23 ± 12.07*,†	51.23 ±2.74*,†
Anxiety problems-T	60.38 ± 12.84*,†	52.46 ± 4.74*,†
ADHD-T	56.15 ±6.57*,†	51.92 ±2.90*,†
Oppositional/defiant-T	55.92 ± 6.73*,†	51.08 ± 2.50*,†
NEPSY
Visual attention-raw	11.21 ± 4.32*†	17.21 ±9.64*,†
Memory/learning percentile	28.22 ± 28.79*,†	53.41 ± 33.72*,†

^aN = 28. Both groups have n = 14. Values given in mean + SD. Table has been abbreviated to show only those variables for which significant group differences were obtained.

^*p< 0.05.

^**p < 0.01.

^†t greater than critical t value obtained from Tukey’s HSD test where critical t = Tukey’s studentised q/(k) and k = the number of means tested.

3.2. PSG

Group means for relevant parameters from overnight PSG are shown in Table 3 with significant differences for several respiratory variables, arousal indices, and sleep architecture parameters. No correlations were found between OSA severity as indicated by variables such as AHI or O₂ saturation and the ERP variables.

Table 3.

PSG variables for OSA and control children.

Variable	OSA groupa	Control groupa
TST	476.64 ± 48.56	469.81 ± 36.961
% TST in stage 1	2.91 ± 2.00	4.39 ± 2.13
% TST in stage 2	53.38 ± 5.68**,†	50.32 ± 7.94**,†
% TST in stage 3	6.39 ± 8.48†	6.53 ± 1.68†
% TST in stage 4	17.51 ± 12.24**,†	21.64 ± 3.62**,†
Total apnoeas and hypopneas	55.86 ± 65.60*,†	3.38 ± 4.00*,†
Total apnoeas	30.00 ± 35.89*	2.13 ± 2.64*
Obstructive	16.00 ± 26.82*,†	0.00 ± 0.00*,†
Total hypopneas	25.86 ± 31.73*,†	1.25 ± 2.05*,†
AHI	7.17 ± 8.70*,†	0.41 ± 0.51*,†
Respiratory arousal index	1.94 ± 1.74**,†	0.29 ± 0.45**,†

^aNote: For the OSA group, n = 14. For the control group, n = 8. Of these, only eight children in the OSA group and five in the control group had accurate end tidal CO₂ recordings values given in mean ± SD.

^*p < 0.05.

^**p < 0.01.

^†t greater than critical t value obtained from Tukey’s HSD test where critical t = Tukey’s studentised q/(√_k) and k = the number of means tested.

3.3. ERP PCA and RMA

Temporal PCA produced five factors, accounting for 89.469% of model variance (Fig. 2). ERP waveforms are also presented in the classic way for comparison. Grand averaged waveform topographic plots are shown for the entire sample (Fig. 3) as well as by group (Fig. 4 shows waveform plots for the OSA group and Fig. 5 for the control group). Single channel recordings for the entire sample, the OSA group, and the control group are shown in Fig. 6. Table 4 contains summary tables for significant mixed-factorial omnibus ANOVA analyses performed on factor weights from temporal factors 2, 3, and 4. Other temporal factors had no significant effects by group. Effect sizes were moderate to large (Cohen’s d = 0.82, ε² = 0.13–0.18). Observed power ranged from 0.553 to 0.936.

Fig. 2.

Factor waveforms from temporal PCA. These are the component waveforms that contribute significant variation in the model independently as identified in factor analysis. Each waveform is plotted in mV (y axis) over ms (x axis) for the 900 ms following auditory stimulus onset. Each plot is labeled with the latency of maximal activation and the proportion of variance for which it accounts, n = 28.

Fig. 3.

Grand-averaged ERP waveforms for the entire sample are shown in a topographic plot. The averaged waveforms for both stimuli are overlapped. The blue line represents the response to the frequent stimulus. The red line is the response to the target stimulus, n = 28.

Fig. 4.

Grand-averaged ERP waveforms for the OSA group are shown in a topographic plot. The averaged waveforms for both stimuli are overlapped. The blue line represents the response to the frequent stimulus. The red line is the response to the target stimulus, n = 14.

Fig. 5.

Grand-averaged ERP waveforms for the control group are shown in a topographic plot. The averaged waveforms for both stimuli are overlapped. The blue line represents the response to the frequent stimulus. The red line is the response to the target stimulus, n = 14.

Fig. 6.

Single channel recordings are shown for all three groups. The electrode chosen (electrode 62 in the 128 channel geodesic sensor net) corresponds to the classic PZ electrode. The blue line represents the response to the frequent stimulus. The red line is the response to the target stimulus. For the grand average of the entire sample, n = 28. For the group-averaged ERP waveforms, n = 14 for both the OSA group and the control group.

Table 4.

Analysis of variance for ERP temporal factors 2, 3, and 4.

Sourcea	df	Mean square	F	p	η2	Observed power
Tests of between-subjects effects
Factor 4
Group	1	1.74	4.72	0.04*	0.15	0.55
Error	26	0.37
Tests of within-subjects effects
Factor 2
Stim*Ele*Gp	4	5.08	4.57	0.002**	0.15	0.94
Error	104	1.11
Factor 3
Sim*Hemi*Gp	1	6.00	5.55	0.03*	0.18	0.62
Error	26	1.08
Factor 4
Ele*Gp	4	4.50	3.77	0.01*	0.13	0.88
Error	104	1.19

^*p < 0.05.

^**p < 0.01.

^aNote: All nonsignificant interactions—including analyses for factor 1—were omitted.

The Group*Stimulus*Electrode interaction was significant for factor 2 (F = 4.57, p = 0.002) peaking at 484 ms. Post hoc analyses indicated the OSA group had more positive activation over occipital sites in response to the target (p = 0.019). Control participants had significantly different patterns of activation in response to frequent versus target stimuli (p = 0.0164). The OSA group did not. A significant Group*Stimulus*Hemisphere interaction for factor 3 (F = 5.552, p = 0.026) at 220 ms post-stimulus onset demonstrated significantly more positive activation over the left hemisphere in response to the frequent stimulus at 220 ms (p = 0.044) for apnoeic children. In factor 4 (peak = 108 ms), a main effect indicated a difference between groups (F = 4.72, p = 0.039). A significant Group*Electrode interaction (F = 3.772, p = 0.007) indicated increased positive activation over central sites for the OSA group (p = 0.036) while increased negative activation occurred over temporal sites (p = 0.011). No significant differences existed in number of correct responses or reaction time for correct responses (Table 5).

Table 5.

Behavioural responses from oddball ERP task.

	OSA groupa	Control groupa
Correct responses	65.42 ± 24.73	53.14 ± 25.62
Response time (ms)	849.13 ± 147.12	794.26 ± 265.20

^aNote: Differences in number of correct responses or response time for correct responses were not significant. N = 28. Both groups have n = 14. Values are given in means ± SD.

3.4. Source localisation

Topographic maps of activation (see Figs. 3–6) from scalp electrodes were used to produce the source localisation solutions by group presented in Figs. 7–9. These figures illustrate projected sources of the observed electrophysiological patterns at specific peak latencies. For children with OSA, the source at 108 ms (peak latency of factor 4) was in the right temporal lobe (Brodmann 39). For control children, this source was in Brodmann area 18 in the lingual gyrus (Fig. 7).

Fig. 7.

Source localisation at 108 ms (peak latency of factor 4). The figure shows the estimated source of the electrical activation patterns measured from the high-density electrode arrays. For the OSA group, the average estimated source of the recorded ERPs was in Brodmann area 39 in the right hemisphere (middle temporal gyrus). For control children, the average estimated source was centrally located in Brodmann area 18 in the lingual gyrus, n = 28. Scale units in mV.

Fig. 9.

Source localisation at 484 ms (peak latency of factor 2). The figure shows the estimated source of the electrical activation patterns measured from the high-density electrode arrays. Estimated source for OSA group was in the amygdala in response to frequent stimuli and superior temporal gyrus in response to target stimuli. The control group’s estimated source was in superior temporal gyrus during frequent stimuli and precuneus (Brodmann area 31) for target stimuli, n = 28. Scale units in mV.

Fig. 8 illustrates average activity for both groups at 220 ms post-stimulus onset (peak latency of factor 3). For apnoeic children, the estimated source in response to target stimuli was the amygdala, and the estimated source in response to the frequent stimuli was within Brodmann area 38 (left superior temporal gyrus). For control children, the estimated source in response to both stimuli was in the limbic lobe (amygdala).

Fig. 8.

Source localisation at 220 ms (peak latency of factor 3). The figure shows the estimated source of the electrical activation patterns measured from the high-density electrode arrays. The averaged estimated source in response to the auditory target stimulus was localised in limbic lobe of amygdala for OSA group. For frequent stimuli, it occurred in left superior temporal gyrus (Brodmann area 38). For control group, the averaged estimated source to both stimuli was amygdala (limbic), n = 28. Scale units in mV.

Fig. 9 shows source localisation at 484 ms (factor 2 peak). At this latency, corresponding to the P300 for this age range, the activation pattern is reversed in the OSA group. Apnoeic children have an estimated source in the uncus of the amygdala in response to frequent stimuli and the superior temporal gyrus in response to targets. Control children exhibited an estimated source in the superior temporal gyrus in response to frequent tone. The estimated source generator in response to target tone was the precuneus of the parietal lobe.

3.5. Multiple regression

Stepwise multiple regression analyses for variables on which groups significantly differed are shown in Tables 6 and 7. Two ERP variables predicted raw scores on NEPSY Visual Attention sub-test: temporal activation for factor 4 and parietal activation to frequent stimuli for factor 2. Factor loadings for these two variables account for 22.6% of the variance in raw scores (adjusted R² = 0.226, F = 4.939, p = 0.016). Activation at central sites to target stimuli for factor 2 (484 ms) significantly predicted percentile scores for NEPSY Memory and Learning domain (adjusted R² = 0.122, F = 4.742, p = 0.039). Factor loadings at central sites account for 12.2% of variance.

Table 6.

Stepwise multiple regression for NEPSY Visual Attention raw scores.^a

Predictor Variable	B	SE B	β	p
Temporal 4	4.96	1.86	0.47	0.01
Parietal Frq 2	5.83	2.63	0.39	0.04

^aNote. Adjusted R² = 0.226. F(2,27) = 4.939 with p = 0.016. Temporal 4 = loading scores for factor 4 from temporal electrode sites. Parietal Frq 2 = loading scores for factor 3 from parietal electrode sites in response to frequent stimuli. Dependent variable = raw scores for Visual Attention subtest of NEPSY.

Table 7.

Stepwise MR for NEPSY Memory and Learning percentile.^a

Predictor Variable	B	SE B	β	p
Central target 2	19.89	9.13	0.39	0.04

^aNote: Adjusted R² = 0.12. F(1,27) = 4.74 with p = 0.04. Central target 2 = loading scores for factor 2 from central electrode sites in response to target stimuli. Dependent variable = percentile for Memory and Learning Domain of NEPSY.

4. Discussion

4.1. Behavioural findings

Children with OSA performed more poorly on Visual Attention and Memory and Learning domains of the NEPSY, and such findings cannot be due to variability in age, sex, SES, or other confounders. The children studied herein not only exemplify the anticipated cross-sectional profile of children around the world, but further confirm previous studies on the presence of behavioural sequelae of OSA in children [2,3,50].

4.2. Electrophysiological findings

By 108 ms following stimulus onset, OSA children exhibited significantly increased – perhaps more effortful – activation relative to their control counterparts, with positive activation over central sites and increased negative activation at temporal sites. Source localisation for this time period identified an electrical source in the right temporal lobe of OSA children – an area classically associated with pure tone sensory processing. Conversely, the control group already exhibited activation in the lingual gyrus, an area near the occipital lobe implicated in attention [51]. This suggests decreased cognitive reserve in the OSA group, requiring these children to recruit additional areas to complete basic sensory processing before performing attention goals. By 220 ms, control children engaged the amygdala during both stimulus conditions, perhaps indicating memory consolidation, and learning [52]. In children with OSA, amygdala activation did occur to the target stimuli. However, their response to frequent tones occurred within the left temporal lobe, indicating that early sensory processing was still taking place. This activation was significantly greater in OSA children, possibly analogous to the sustained activation reported in apnoeic adults and interpreted as recruitment of alternate – less effective – attentional resources [29].

By 484 ms – approximate P300 at this age – the control group showed activation in the precuneus, a region of the parietal lobe implicated in integrated tasks involving working memory and executive functioning [53]. The OSA group exhibited a pattern of reversal from the one they displayed at 220 ms with activation occurring in the amygdala during frequent stimuli and the temporal lobe during target stimuli. At this late latency, the OSA children continued to process the task as a basic sensory perception exercise, at least in part. These differences were found in the absence of grossly observable phenomena, including impaired reaction time or accuracy.

4.3. Regression findings

ERP variables predicted scores for NEPSY variables on which the groups differed significantly. Specifically, as temporoparietal activation increased, so did Visual Attention scores. Increased temporal activation at 108 ms and increased parietal activation at 484 ms were significant predictors of success for the attention task, accounting for 22.6% of variance. The OSA group performed more poorly on this task, and had more diffuse, poorly localised patterns of activation. This could indicate a deficit in basic attention processing, in working memory and executive functioning components of the task, or both. In the context of the reported source localisation, it seems that – though additional resources are recruited to assist with auditory attention processes – the centralised executive presumably engaged when control children activate parietal areas is adversely impacted by SDB in the OSA group.

Memory and Learning scores were predicted by factor loadings for central sites during target stimuli. Increased central activation positively correlated with scores, accounting for 12.2% of variance. At 484 ms – the classic P300 peak – central activation during novel stimuli predicted success for Memory and Learning. Deficit in centroparietal attention and executive regions could reflect neurobehavioural problems previously reported in paediatric OSA, including inattention and poor school performance. Significant relationships between ERP variables and standardised neuropsychological tests further support construct validity of ERP techniques.

5. Conclusion

We report brain changes associated with OSA that can be used to determine which children might require earlier diagnosis or treatment. Children with OSA exhibit substantial alterations in behavioural tests, ERP temporal trajectories, and localisation patterns during an attention task. Activation in centroparietal areas is selectively impaired in OSA. These findings are remarkably similar to affected brain regions identified in a recent functional magnetic resonance imaging (fMRI) study of apnoeic adults [54]. In our OSA group, increased, diffuse activation of other neural resources is effort intensive, similar to apnoeic adults who exhibit compensatory neural responses with increasing disease severity [55]. Diffuse, unfocussed brain activation patterns as observed herein in children with OSA is a similar compensatory mechanism that is not necessarily more efficient, considering it is negatively correlated with success on standardised neuropsychological assessments. The apparently selective disruption of cognitive processes occurring in the cortical substrates of centroparietal brain areas could account for parentally reported problems with attention and school performance in paediatric OSA. That these children had no significant differences in socioeconomic parameters, age, race, or sex, further supports that the ERP changes observed do relate to the alterations in cognitive function associated with OSA. Although the mechanism by which OSA impacts cognition and behaviour remains unknown, younger children are less likely to be included in such studies despite the fact that they are at greatest risk for developing OSA. Early intervention could be key in optimising treatment outcomes. Increased arousal indices in snoring infants are sufficient to impose declines in development [56]. It is highly desirable to identify patients requiring prioritised intervention secondary to greatest vulnerability using objective, brief testing approaches such as ERP, particularly considering the inconsistency of findings across published studies [57]. Although the results reported are most valid at the group level, in time and with expanded exploration of the concepts examined herein, electrophysiological methods might become valid and applicable at an individual level, similar to current applications of retinal and auditory brainstem ERP applications.

Potential weaknesses of this study include the use of a relatively novel technique that involves advanced statistical analyses of electrophysiological data. The authors concede that – at the current state of the art – few laboratories are able to perform such advanced analyses, particularly with children who might be low performing or less collaborative. However, we are optimistic that – with the developing technology and increased use of such technology by clinicians, as well as the advancement of commercially available statistical analysis programs – ease and frequency of use will increase, as evidenced by the extensive clinical use of ERP in the audiology or visual fields. Furthermore, although research of these methods does require demanding statistical analyses, the demand on parents is negligible, and the time taken to complete ERP tasks with children is lesser than the duration of many neuropsychological assessments. It is in appealing to these facts that the authors optimistically maintain that ERP could provide a more objective and less effortful measure for patients and their families. These current findings provide foundation for more extensive exploration of neurocognitive function in apnoeic children, particularly considering the absence of robust relationships between PSG variables and neurocognitive impairments.

Abbreviations:

A/EF

attention/executive functioning domain

ADHD

attention deficit/hyperactivity disorder

AHI

apnoea hypopnea index

ANOVA

analysis of variance

AT

adenotonsillectomy

BMI

body mass index

CBCL

Child Behaviour Checklist

CSI-4

Childhood Symptom Inventory

DSM-IV

Diagnostic and Statistical Manual of Mental Disorders-4

ECI-4

Early Childhood Inventory-4

EEG

electroencephalogram

ERP

event-related potential

fMRI

functional magnetic resonance imaging

IQ

Intelligence Quotient

LAURA

local autoregressive average

NEPSY

neuropsychological assessment

NIH

National Institutes of Health

OSA/OSAS

obstructive sleep apnoea/obstructive sleep apnoea syndrome

PCA

principal components analysis

PETCO₂

end tidal carbon dioxide

PPVT-III

Peabody Picture Vocabulary Test-III

PSG/NPSG

polysomnography/nocturnal polysomnography

RMA

repeated measures ANOVA

SBQ

Sleep Behaviour Questionnaire

SD

standard deviation

SDB

sleep disordered breathing

SES

socioeconomic status

SPL

sound pressure level

TST

total sleep time