The Impact of Cognitive Behavioral Therapy on Sleep Problems in Autistic Children with Co-occurring Anxiety

Abstract

Purpose:

This study seeks to examine the relationship between anxiety-symptom severity and sleep behaviors in autistic children receiving cognitive behavioral therapy (CBT).

Methods:

We conducted a secondary-data analysis from a sample of 93 autistic youth, 4 to 14 years, participating in 24 weeks of CBT. Clinicians completed the Pediatric Anxiety Rating Scale (PARS) and parents completed the Children’s Sleep Habits Questionnaire, Abbreviated/Short Form (CSHQ-SF) at baseline, mid-treatment, post-treatment and 3-months post-treatment. Mediation analysis evaluated the role of anxiety symptoms in mediating the effect of time in treatment on sleep.

Results:

There was a negative association between time in treatment and scores on the CSHQ-SF (b = −3.23, SE = 0.493, t = −6.553, p < 0.001). Increased time in treatment was associated with decreased anxiety (b = −4.66, SE = 0.405, t = −11.507, p < 0.001), and anxiety symptoms decreased with CSHQ-SF scores (b = 0.322, SE = 0.112, t = 2.869, p = 0.005). The indirect effect of time in treatment on CSHQ-SF scores through PARS reduction was negative, but not statistically significant.

Conclusion:

Increased time in CBT was associated with decreased anxiety severity and improved sleep behaviors. Reductions in anxiety symptoms may mediate improvements in sleep problems, but larger sample sizes are necessary to explore this further.

Sufficient quality and duration of sleep are necessary for optimal neurodevelopmental functioning (Mason et al., 2021). Sleep problems occur frequently in autistic children, with prevalence rates ranging from 50–80% compared to 25–50% in typically developing children (Couturier et al., 2005; Krakowiak et al., 2008; Richdale & Schreck, 2009), and persist from childhood through adolescence (Humphreys et al., 2014; Goldman et al., 2012). Examples of sleep problems include difficulties with sleep onset and sleep maintenance, problematic bedtime routines, limited sleep duration, or restless sleep, among others. Additionally, sleep problems in autistic children have been associated with more severe externalizing and internalizing symptoms (Sikora et al., 2012; Goldman et al., 2011; Veatch et al., 2017; Mazurek et al., 2019).

Sleep problems correlate with additional features both inherent to and associated with autism (Cohen et al., 2014; Hollway & Aman, 2011). For example, in a large cross-sectional study of sleep disturbances in autistic children, anxiety was the strongest predictor of sleep disruption, while other characteristics (e.g., higher IQ scores, higher autism core-symptom severity, sensory sensitivities, gastrointestinal symptoms) further contributed to the observed variance in total sleep scores (Hollway & Aman, 2011). Similarly, Mazurek and Petroski also found anxiety to be highly correlated with sleep problems in autistic individuals (2015). In a recent meta-analysis, Han and colleagues (2022) examined the strength of associations between sleep problems and various domains of daytime functioning (adaptive functioning, executive functioning, and physical health) and clinical symptomatology (core autism symptoms, internalizing and externalizing conditions) in autistic individuals. Their analysis revealed that sleep problems were most strongly associated with externalizing and internalizing symptoms, particularly mood and anxiety symptoms (Han et al., 2022). Sleep problems are prevalent in children with anxiety, but the directionality of the relationship between sleep and anxiety remains unclear (Brown et al., 2018; Alvaro, Roberts & Harris, 2013). For example, a child may have difficulty with sleep initiation because s/he is perseverating on anxious thoughts. Alternatively, poor sleep may result in downstream consequences, such as affective dysregulation or increased stress responsivity, which could exacerbate anxiety-related symptoms (Cox & Olatunji, 2020). To further understand this relationship, comprehensive, repeated measures of anxiety and sleep over time are necessary, and repeated measurements of sleep and anxiety throughout the course of an intervention study would be particularly helpful to examine the presence of causal and bidirectional pathways between sleep and clinical symptomatology in autism (Mazurek et al., 2019; Han et al., 2022).

There is robust evidence that cognitive behavioral therapy (CBT) is an effective treatment modality for anxiety in autistic children and adolescents (Wood et al., 2009, 2015, 2020). While this has not yet been investigated in youth with autism and anxiety, sleep problems have been reduced following anxiety-focused CBT in youth without autism (Peterman et al., 2016; Storch et al., 2008; Caporino et al., 2017; Donovan, Spence & March, 2018). Interestingly, a recent pilot study found favorable results of telehealth-delivered CBT specifically for insomnia (CBT-I) in a cohort of school-aged autistic children with co-occurring insomnia (McCrae et Al., 2020). However, the study did not assess the extent to which anxiety co-occurred in the sample of autistic children with insomnia. Thus, the current study aimed to address this gap by examining the relationship between anxiety-symptom severity and sleep behaviors in a cohort of autistic children participating in anxiety-focused cognitive behavioral therapy (CBT). Our aim was to examine potential changes in sleep disturbances, as a function of CBT, hypothesizing that improvement in anxiety symptoms would mediate the effect of time in treatment on improvement in sleep-problem behaviors.

Methods

This study is a secondary analysis of data from a 24-week parent-led, stepped care CBT approach, outlined in Storch et al. (2022). The local Institutional Review Board approved the study (H-44081). Parents provided written informed consent for themselves and their child to participate in the study, and when possible, child participants provided written and/or verbal assent.

Participants

Ninety-six individuals (84.4% male), ages 4 to 14 years (mean age = 10.39 years, SD = 2.86 years), provided consent to be evaluated for study participation. Ninety-three completed baseline assessments, with 76 of those children entering treatment. Data collection occurred between January 2019 and November 2020. Participants had a previously established diagnosis of autism determined clinically using established methods (e.g., a validated instrument, detailed history, and behavioral observations) ) and documented by a review of records, along with a baseline score ≥ 65 on the Social Responsiveness Scale, 2nd Edition (Constantino & Gruber, 2012). Participants also had clinically significant symptoms of anxiety or OCD, defined by a clinical severity rating ≥ 4 for an anxiety/OCD diagnosis on the Anxiety Disorders Interview Schedule IV, Child/Parent Version (ADIS- IV-C/P; Silverman & Albano, 1996) with Autism Spectrum Addendum (ASA) (Kerns et al., 2017) and a score > 12 on the Pediatric Anxiety Rating Scale (PARS) (RUPP, 2002). Participants all had full scale and verbal comprehension IQ ≥ 70 as indicated by the Differential Ability Scales-II (DAS-II; Beran, 2007) or the Wechsler Abbreviated Scale of Intelligence-Second Edition (WASI-II; Weschler, 2011). The study excluded children who had DSM-5 diagnoses of bipolar disorder, psychotic disorder, and/or intellectual disability. Other exclusion criteria were (a) active suicidal/homicidal ideation or self-injury, (b) current receipt of psychotherapy for anxiety, (c) initiation of a psychiatric medication within 8 weeks of enrollment, or (d) having changed the dose of a medication within 4 weeks (antidepressant) or within 2 weeks (stimulant or benzodiazepine) of enrollment. We recruited participants through a variety of sources, including direct referrals from an autism specialty clinic, social media, community efforts, and the SPARK research match database (Feliciano et al., 2018).

Procedures

In the first 12 weeks of treatment, participants engaged in 4 sessions of parent-led, therapist-assisted CBT. This treatment followed Helping Your Anxious Child (HYAC; Rapee et al., 2017) and the accompanying workbook, in addition to therapy modeled after a validated CBT protocol, either (a) The Cool Kids Anxiety Program: Autism Spectrum Disorder Adaptation (Cool Kids ASD), 2^nd Edition (Lyneham et al., 2016) for children ages 6 and older; or (b) Exposure-Focused, Family-Based CBT for Youth with ASD and Comorbid Anxiety for children ages 4–5 (Storch et al., 2020). After the first 12 weeks of parent-led CBT, participants who scored < 2 (Mild Symptoms) on the Clinical Global Impression-Severity (CGI-S) with a Clinical Global Impression-Improvement (CGI-I) rating of 6 or 7 (much or very much improved) were considered responders and continued in parent-led CBT for the subsequent 12 weeks. Non-responders were enrolled in therapist-led CBT sessions for the next 12 weeks, which consisted of 10 sessions using the same treatment protocols. Full details of the treatment and mid-point assessment criteria to determine responders versus non-responders are outlined in Storch et al. (2022).

Participants completed a 3- to 4-hour baseline assessment prior to enrolling in the study. Participants additionally completed a mid-point assessment, a post-treatment assessment, and a 3-month post-treatment assessment. Various parent- and clinician-report measures were completed over the course of the study (see Storch et al. [2022] for complete list of measures). The two measures of interest in the current analysis were completed at all timepoints.

Measures

Pediatric Anxiety Rating Scale (PARS)

The Pediatric Anxiety Rating Scale, or PARS, (RUPP 2022) is a clinician-reported instrument designed to measure anxiety-symptom severity in children. The measure has a 50-item symptom checklist with responses indicated as Yes/No regarding the presence of the symptom, along with a 7-item, severity-rating scale where items are rated on a 6-point Likert-type scale from Minimal (0) to Extreme (5) with regard to number of symptoms, frequency, symptom severity, avoidance, and interference. Questions are asked in an interview format, and respondents are to consider symptoms during the past seven days. A score ≥ 12 is considered clinically significant (Ginsburg et al., 2011). The PARS has demonstrated high inter-rater reliability, test-retest reliability, internal consistency, and validity in autistic children (Storch et al., 2012).

Children’s Sleep Habits Questionnaire-Abbreviated/Short Form (CSHQ-SF)

The Children’s Sleep Habits Questionnaire, Abbreviated/Short Form (CSHQ-SF) is a 22-item parent/caregiver-reported questionnaire that queries a child’s sleep behaviors over the past week. These symptoms are conceptually grouped into four domains: (1) Bedtime (2) Sleep Behavior (3) Waking During the Night and (4) Morning Wake Up. Parents or caregivers rate the symptoms on a 5-point scale based on the frequency of the sleep behavior (from 7/7 days or “Always” to 0/7 days or “Never”). A Total Sleep Disturbance score is calculated, with higher scores indicating more sleep problems. The CSHQ-SF is a reliable and valid tool to identify sleep problems in school-aged children (Bonuck et al., 2017) and the full CSHQ has been validated in autistic children (Katz et al., 2018 and Souders et al., 2009).

Data Analyses

Mediation analysis was conducted to evaluate the role of anxiety symptoms in mediating the effect of time in treatment on sleep. In this evaluation, time is the predictor, sleep behavior derived from the CSHQ-SF is the outcome, and anxiety as measured with the PARS is the potential mediator. Data were analyzed with a series of multilevel models using R (R Core Team, 2023) and the lme4 (Bates et al., 2015) package. Restricted maximum likelihood was used instead of maximum likelihood to produce more accurate estimates of variance parameters (Boedeker, 2017). Two-level models allowed for repeated measures to be nested within individuals. Linear and quadratic specifications for time were compared, and the linear model was found to be a better fit for the three time-points considered when comparing Bayesian information criterion (Schwarz, 1978). Fixed and random specifications of the coefficients of interest were compared. In all cases, the fixed-effect specification yielded lower BIC values and was therefore preferred.

Mediation analysis with multilevel models requires specification of a series of separate models. In the first model, we estimated the overall effect of time in treatment on sleep behavior by regressing the repeated measures of sleep behavior on time in treatment. The indirect effect was estimated as the product of coefficients from two additional models. The first model fit assessed the effect of time in treatment on the mediator, PARS, yielding the a path. The second model assessed the effect of PARS on CSHQ-SF (the b path), controlling for time in treatment (the c path). The indirect effect was found by multiplying the a and b paths and the indirect effect’s confidence interval derived using the monte carlo method (Preacher & Selig, 2012). A 95% confidence interval that does not contain 0 indicates that the indirect effect is statistically significant. An effect size for mediation models is the proportion of the total effect that is attributable to the indirect effect (Preacher & Kelley, 2011). Final model analyses were conducted with both complete cases and using multiple imputation. Given the similarity across results, complete case analysis results are shown. All demographic variables were included as covariates in the analytic and imputation models.

Results

Ninety-six individuals consented for the study (Table 1), and 93 completed a baseline PARS and CSHQ-SF. Of these, 76 entered treatment (see Storch et al. [2020] for reasons why the n = 20 were excluded). Fifty-seven completed both measures at the mid-point of the treatment (12 weeks) and 48 individuals completed the measures at the end of the treatment (24 weeks). Thirty-four individuals completed both measures at all timepoints, including 3 months post-treatment (Figure 1). Little’s missing completely at random test was used to evaluate the hypothesis that data were missing randomly as opposed to systematically. The test was not significant (p = 0.481), indicating that listwise deletion of cases with missing data may not result in biased estimates.

Table 1.

Sample Demographics

Demographic Characteristic	N = 96 (%)
Gender
Male	81 (84.4%)
Female	15 (15.6%)
Ethnicity
Not Hispanic or Latinx	61 (63.5%)
Hispanic or Latinx	35 (36.5%)
Race
White	69 (71.9%)
Black or African American	7 (7.3%)
American Indian or Alaska Native	1 (1.0%)
Asian	7 (7.3%)
Native Hawaiian or Pacific Islander	1 (1.0%)
Mixed Race	9 (9.4%)
Other	2 (2.1%)
Maternal Education
High school diploma/GED	4 (4.2%)
Some college credit, no degree	22 (22.9%)
Trade/technical/vocational training	4 (4.2%)
Associate’s degree	3 (3.1%)
Bachelor’s degree	32 (33.3%)
Master’s degree	24 (25.0%)
Ph.D or M.D. or J.D.	7 (7.3%)
Income Category
$0 to $39,999	12 (12.5%)
$40,000 to $79,999	14 (14.6%)
$80,000 and over	55 (57.3%)
Not disclosed	15 (15.6%)
Language *
English	94 (97.9%)
Spanish	19 (19.8%)

^*Some households were bi-lingual; thus percentages do not add to 100%

Figure 1: Consort Diagram.

*Note: Excluded/Dropout rate refers to participating in entirety of treatment, and does not necessarily reflect those that completed the measures of interest in this secondary analysis

Baseline elevated PARS scores were positively correlated with elevated CSHQ-SF scores (Pearson’s r = 0.31, p = 0.003; Table 2). Evaluating the overall relationship between time in treatment and sleep, there was a negative association between time in treatment and scores on the CSHQ-SF (b = −3.23, SE = 0.49, t = −6.55, p < 0.001). To evaluate the indirect effect, the a and b paths were estimated and multiplied together. Path a was found by regressing anxiety symptoms on time and was statistically significant (b = −4.66, SE = 0.41, t = −11.51, p < 0.001), indicating that increased time in treatment was associated with decreased anxiety (Figure 2). Path b was also significant (b = 0.32, SE = 0.11, t = 2.87, p = 0.005) indicating that anxiety symptoms decreased with CSHQ-SF scores. The indirect effect of time in treatment on CSHQ-SF scores through PARS reduction was negative but not statistically significant (a*b = −1.51, 95% CI [−4.69, 1.55]) and accounted for 46.7% of the overall effect of time in treatment on CSHQ-SF. In those individuals with scores at all time points, there was not a statistically significant difference in scores immediately post-treatment and 3-months post-treatment (p = 0.32).

Table 2.

Correlations Between Variables

	1	2	3	4	5	6
1. PARS Baseline	20.8 (5.1)
2. PARS Mid	0.38**	16.4 (5.9)
3. PARS Post	0.32*	0.43**	12.1 (6.8)
4. CSHQ Baseline	0.31**	0.09	0.12	29.9 (11.0)
5. CSHQ Mid	0.39**	0.2	0.28	0.88***	25.7 (9.8)
6. CSHQ Post	0.2	−0.03	0.36*	0.72***	0.79***	22.9 (9.4)

Note. The main diagonal contains the mean and standard deviation of each variable and the lower triangle the bivariate correlations using pairwise deletion.

^*p < 0.05

^**p < 0.01

^***p < 0.001.

Figure 2: Mediation Model.

The mediation model shown as a path diagram. The a path connecting Time to PARS inidicates the associated effect between time in treatment on PARS. The b path connecting PARS to CSHQ-SF indicates the consistent associated effect of PARS on CSHQ-SF. The remaining path from Time to CSHQ-SF is the remaining direct effect after controlling for the mediating pathway. *p < 0.05; **p < 0.01, ***p < 0.001.

Discussion

This study examined the relationship between anxiety-symptom severity and sleep behaviors in a cohort of autistic children participating in anxiety-focused CBT. We found that higher scores on a clinician-rated anxiety measure correlated with higher scores on a parent-reported, sleep-behavior measure at baseline. Increased time in CBT was associated with decreased anxiety severity and improved sleep behavior. This immediate, post-treatment improvement in sleep scores persisted for three months post treatment. While increased time in treatment resulted in improved sleep behaviors, the indirect effect through reduction in anxiety-symptom severity was not statistically significant. However, the reduction in anxiety-symptom severity did account for 46.7% of the overall effect of time in treatment on improvement in sleep behavior, indicating the need for further exploration of this relationship in a larger sample size.

Given the correlation between baseline anxiety and sleep problems in this sample, clinicians should consider proactively assessing problematic sleep behaviors in autistic children with anxiety or conversely, assessing for anxiety in autistic children with sleep difficulties. Our findings, combined with findings from prior studies of the use of CBT in autistic youth, suggest that CBT is an effective treatment modality that may have secondary positive effects on sleep. In this study, parent-led CBT was effective for a proportion of the participants who did not require stepping up to therapist-led care—a finding that may have important implications for treatment accessibility. Additionally, recent work with autistic youth has demonstrated that CBT for childhood insomnia (CBT-CI) and CBT for anxiety can be delivered effectively via telehealth, which offers another platform that could serve to increase treatment accessibility (McCrae et al., 2021; Guzick et al., 2023).

There were several limitations to this secondary analysis. The sample size of individuals who completed assessments at all timepoints was small, as there was considerable participant attrition, higher than that in previously reported trials of more traditional CBT models for autistic youth with anxiety (Wood et al., 2015, 2020). The stepped-care approach itself and/or the onset of the COVID-19 pandemic may have impacted attrition, as much of the drop out occurred just before or at the start of the pandemic, when the treatment moved from in-person to virtual (Storch et al., 2022). Additionally, this study relied on self-, parent- or clinician-report of symptoms for multiple measures, including sleep behaviors/habits. Self-report, often needed for adequate evaluation of anxiety, was also unavailable for a small subset of the participants in the youngest age range. Though multi-modal assessment is best, clinician-rated assessment of anxiety has been considered gold standard and utilized as the primary outcome measure in major child anxiety trials (Compton et al., 2010; Wood et al., 2020). Thus, clinician rating of anxiety via the PARS was used for this secondary analysis. The CHSQ-SF included in this study is limited in the range of sleep problems assessed compared with the full CHSQ and has not been evaluated in autistic children in the same manner as the full version. Finally, individuals with IQ < 70 were excluded from this study given the verbal nature of CBT; thus, results may not generalize to individuals with co-occurring autism, anxiety, and intellectual disability. This study was a secondary-data analysis with limited sample size; therefore, we were unable to power the study to detect substantively meaningful indirect effects. Although the magnitudes of the relationships are estimated, we cannot make definitive claims based on statistical significance.

Despite these limitations, this study identifies a potential relationship between reduction in anxiety severity and reduction in sleep behavior problems over time in CBT with autistic youth, but larger sample sizes are necessary to explore this further. The next step in clarifying this relationship would be to engage this question as a primary aim, rather than examining it retrospectively via a secondary data analysis. This would involve measuring sleep habits via the full version of the CSHQ, or ideally a more objective measure of sleep such as actigraphy, in a larger sample of autistic youth receiving CBT. Conversely, research investigating interventions targeting sleep in autistic youth should also measure anxiety symptoms over the course of treatment, given the interplay between sleep and anxiety. It is imperative to further our understanding of the relationship between these symptoms, which are a common source of clinically significant distress, significantly impacting quality of life in autistic individuals and their families.

Table 3.

Model Results

	Model 1 c Path Outcome = CSHQ	Model 2 a Path Outcome = PARS	Model 3 b, c’ Paths Outcome = CSHQ
(Intercept)	47.82 (10.14)***	14.75 (5.10)**	42.72 (9.51)***
Time	−3.23 (0.49)***	−4.66 (0.41)***	−1.66 (0.75)*
PARS			0.32 (0.11)**
Age	−1.06 (0.47)*	0.37 (0.24)	−1.17 (0.44)*
Female (Reference: Male)	4.24 (3.06)	1.81 (1.58)	3.78 (2.83)
Income (Reference: $0 to $9,999)
$10,000 to $19,999	−7.77 (12.78)	11.76 (6.42)	−11.1 (11.8)
$20,000 to $29,999	−4.58 (11.11)	2.63 (5.69)	−5.06 (10.24)
$30,000 to $39,999	0.79 (9.51)	2.79 (4.80)	0.12 (8.75)
$40,000 to $49,999	−20.65 (11.98)	1.91 (6.07)	−20.42 (11.07)
$50,000 to $59,999	−10.41 (12.12)	3.99 (6.04)	−11.53 (11.13)
$60,000 to $69,999	−13.98 (9.57)	5.32 (4.75)	−15.57 (8.82)
$70,000 to $79,999	−16.34 (12.24)	−1.07 (6.34)	−14.96 (11.33)
$80,000 and over	−7.35 (8.73)	3.26 (4.36)	−8.21 (8.03)
Mother Employment (Reference: Employed Full Time)
Employed part time	−4.09 (3.37)	1.59 (1.74)	−4.64 (3.12)
Unemployed, looking for work	3.81 (4.74)	−0.14 (2.46)	3.76 (4.38)
Unemployed, not looking for work	−1.35 (2.82)	2.29 (1.43)	−1.93 (2.61)
Retired	−7.24 (7.36)	−0.97 (3.73)	−7.15 (6.79)
Student	21.22 (11.88)*	7.68 (6.54)*	18.52 (11.13)
Spanish (Reference: English)	1.71 (4.59)*	−0.04 (2.45)	1.41 (4.26)
Race (Reference: White)
Black or African American	−9.34 (4.58)	−2.78 (2.30)	−8.5 (4.23)
American Indian or Alaska Native	−34.34 (16.68)	−17.62 (8.75)	−28.42 (15.58)
Asian	−7.34 (4.63)	−3.66 (2.23)	−7.11 (4.28)
Native Hawaiian or Pacific Islander	10.60 (11.05)	4.61 (6.11)	9.39 (10.29)
Mixed Race	−0.89 (3.82)	−3.67 (2.01)	0.51 (3.57)
Other	−2.66 (8.33)	−1.99 (4.65)	−1.59 (7.77)
Hispanic or Latino (Reference: Not Hispanic or Latino)	−1.40 (2.78)	−2.32 (1.44)	−0.77 (2.58)