Associations between frontal lobe structure, parent-reported obstructive sleep disordered breathing and childhood behavior in the ABCD dataset

Abstract

Parents frequently report behavioral problems among children who snore. Our understanding of the relationship between symptoms of obstructive sleep disordered breathing (oSDB) and childhood behavioral problems associated with brain structural alterations is limited. Here, we examine the associations between oSDB symptoms, behavioral measures such as inattention, and brain morphometry in the Adolescent Brain Cognitive Development (ABCD) study comprising 10,140 preadolescents. We observe that parent-reported symptoms of oSDB are associated with composite and domain-specific problem behaviors measured by parent responses to the Child Behavior Checklist. Alterations of brain structure demonstrating the strongest negative associations with oSDB symptoms are within the frontal lobe. The relationships between oSDB symptoms and behavioral measures are mediated by significantly smaller volumes of multiple frontal lobe regions. These results provide population-level evidence for an association between regional structural alterations in cortical gray matter and problem behaviors reported in children with oSDB.

Fig. 1. Relationship between symptom burden of obstructive sleep disordered breathing (oSDB) and Child Behavior Checklist scores.

a–c show the prevalence of parent-reported frequency of symptoms of obstructive sleep disordered breathing in the Adolescent Brain and Cognitive Development Study (n = 10,140). Venn diagrams show the number of children with each symptom in non-intersecting areas or in combination with other symptoms in the intersecting areas. The frequency of symptoms was categorized as occasional (a, once or twice a month), sometimes (b, once or twice a week) or habitual (c, >2 nights a week). The relationship between oSDB burden measured by the sleep-related breathing disorders subscale within the Sleep Disturbance Scale for Children and the three composite measures (total, externalizing, and internalizing problems) from the Child Behavior Checklist (CBCL) was assessed using generalized additive models. Models were adjusted for age, sex, race/ethnicity, the presence of asthma and total household income before taxes as fixed effects and study site as a random effect. d–f The relationship between the oSDB factor score and the predicted marginal values of three CBCL scales—total, externalizing, and internalizing problems, respectively. The adjusted distributions were fitted with a smoothing spline, with the error bars spanning one standard deviation around the mean predicted value of each composite CBCL measure with increasing oSDB score grouped by sex. g compares the effects of the parent-reported frequency of oSDB symptoms (e.g., snoring frequency) on problem behaviors by measuring the change in the overall proportion of variance in the generalized additive model following their addition to the null, covariates-only model. The null model comprised solely of fixed effect (age, sex, race/ethnicity, asthma, and total household income before taxes) and random effect (recruitment site) covariates. The comparisons of the base and covariate-adjusted models using analysis of variance yielded statistically significant results for all CBCL categories (two-sided P < 10⁻¹⁶, unadjusted for multiple comparisons). The greatest effect was identified for the relationship between the oSDB symptom score and total problems. Among the individual symptoms, the frequency of snoring predicted CBCL scores better than others. Source data are provided as a Source Data file.

Fig. 2. Interaction between household income and obstructive sleep disordered breathing in predicting problem behaviors.

a–j Each contour plot, derived from 10,140 children in the Adolescent Brain Cognitive Development study, shows the interaction between household income and obstructive sleep disordered breathing (oSDB) factor score in predicting the CBCL scores. The three composite CBCL scales are shown in a–c while the individual syndrome CBCL scales are shown in d–j. The median of each income class was log-transformed due to positive skew. The parent-reported obstructive sleep disordered breathing (oSDB) factor score is shown on the x-axis and the natural log of income on the y-axis. The angled orientation of the contours indicates an interaction between income and oSDB on each CBCL score. All models included the summed effects of age, sex, race/ethnicity, asthma as well as the recruitment site as a random effect. k The effect size estimates were calculated by measuring the change in the overall proportion of variance (expressed as a percentage) by adding the interaction term to the no-interaction model linking oSDB factor score to each CBCL measure. Source data are provided as a source data file.

Fig. 3. Obstructive sleep disordered breathing (oSDB) is associated with regional cortical thinning.

a–d Atlas-based effect size maps projected on a cortical surface demonstrate the relationship between the frequencies of individual symptoms, the total oSDB factor score and cortical morphometric variables (average cortical thickness (left) and total volume (right)) measured using magnetic resonance imaging (MRI). Among the individual symptoms, the greatest effect size was identified for the relationship between the frequency of snoring and average cortical thickness (c). The top ten cortical regions of interest (ROI) wherein the strongest effects were identified for each morphometric variable are shown in e. Effect sizes were calculated by measuring the change in overall proportion of variance by adding the predictor to the covariates-only model comprising age, sex, race/ethnicity, and household income as fixed effects and MRI scanner as a random effect. f The covariate-adjusted model for the relationship between oSDB factor score and the mean cortical thickness within the left medial orbital sulcus in 10,140 children. This relationship was replicated for the frequency of snoring as a predictor (g). The error bars in f, g span 95% confidence intervals around the estimated marginal mean cortical thickness. h shows a potential threshold associated with cortical thinning within the left medial orbital sulcus with pairwise comparisons between 661 children who snored habitually (>2 nights a week), and 6040 children who did not snore. However, there was no significant difference between 3439 children who snored non-habitually, defined as less than three nights a week, compared to the non-snoring children. i A similar comparison for the average thickness of the right superior frontal sulcus. A P value threshold of 0.05 was applied to all effect size maps following correction for false discovery. All pairwise comparisons in h and i were adjusted for multiplicity using the Tukey method with the arrows on either side of the estimated mean highlighting the regions of overlap of the estimated marginal means. These tests were two-sided. Cortical surface area was not associated with oSDB or any of the individual symptoms. Source data are provided as a source data file.

Fig. 4. Frontal lobe regions mediate the relationship between obstructive sleep disordered breathing (oSDB) and Child Behavioral Checklist (CBCL) scores.

a General outline of the mediation model that estimates the extent of the covariate-adjusted relationship between oSDB factor score and the CBCL measures apportioned to alterations in cortical volume. These mediation effects expressed as a proportion of the total effect (% mediated) were projected on to atlas-based cortical effect size maps for CBCL scores (b). All mediation models included age, sex, race/ethnicity, history of asthma, and the total household income before taxes as fixed effects, and the recruitment site and the scanner serial number as random effects. The most widespread and strongest mediation effects were identified for attention problems (highlighted). c The top ten mediated effects for cortical regions of interest (ROI) for both oSDB factor score and the frequency of snoring showing similar effects. d Explores the clinical threshold for the frequency of snoring as a predictor for CBCL scores. Following recategorization of snoring frequency as none, non-habitual (less than three nights a week) and habitual (at least three nights a week), mediation effects were identified only for habitual snoring. All mediation effects were derived from jointly modeling two regressions—the first assessed the ROI using the predictor and the second assessed the CBCL score using the ROI as a predictor. The error bars span 95% confidence intervals associated with the average mediated effect and were obtained by bootstrapping 1000 replicates. A threshold of P < 0.05 following correction for false discovery was applied to all mediation models in addition to removing mediation effects whose confidence intervals crossed zero. Mediation effects for abnormal CBCL scores as a categorical variable are shown in Fig. S3. Source data are provided as a Source Data file.