Interventions for anxiety in mainstream school-aged children with autism spectrum disorder: A systematic review

Types of studies

While the original strategy did not set limits on the types of studies to be reviewed, the results of initial searching proved so prolific that it was decided, on the basis of quality, to focus on two main types of quantitative studies—RCTs and quasi-experimental studies (in which a control group was employed but allocation was not strictly randomized). The mixed methods strategy proposed initially in the protocol (Lietz et al., 2018) was thus replaced with a purely quantitative review and meta-analysis. Otherwise, this review followed the approaches to search strategies and analyses specified in the study protocol (Lietz et al., 2018) which was published by the Campbell Collaboration prior to starting the research.

The studies could occur in schools or out-of-school settings (e.g., home, larger community) or clinical settings, as long as the intervention was designed to improve outcomes in real-world settings.

The comparison groups used in the majority of included studies were waitlist control groups or standard treatment/treatment-as-usual (TAU) groups. Two studies, namely vanSteensel_2015

Types of participants

The target population for the review is mainstream school-aged children, diagnosed with ASD (inclusive of autism, ASD, Autistic Disorder, Asperger's Disorder, Asperger Syndrome, atypical autism, PDD-NOS) by a professional eligible to diagnose these conditions, and also experiencing anxiety symptoms or a diagnosis of an anxiety disorder provided by a professional eligible to diagnose such conditions. The majority of included studies (21 of 24) used a screening instrument to confirm the existence of clinically significant levels of anxiety at intake, while the remaining studies relied on parent or teacher reports of elevated anxiety.

If studies included a sample of children in the target population as well as other children (e.g., the general population) and the findings were separated for the ASD subgroup, the study was included in the review whereby the type of ASD diagnosed did not matter. In contrast, if the study findings were not reported separately (e.g., the results for children with ASD and ADHD were combined for analysis), the study was excluded from the review as the impact of the intervention on only the ASD sample would be impossible to isolate.

To be included in the review, either all participants in a study had to be of mainstream school age or a majority of participants had to be of mainstream school age. This meant that while most studies involved young people aged 6–16 years, one study (Piravej_2009) included some younger children (minimum 3 years old) and six studies included slightly older participants (MacKinnon_2014, Pryor_2016, White_2013: max. 17 years; Murphy_2017, van Steensel_2015: max. 18 years; and Hepburn_2019: max. 19 years). No restrictions were imposed in terms of background variables such as socioeconomic status, or profiles of children with ASD with respect to characteristics such as level of cognitive functioning or ASD severity/classification, for example. However, given the types of interventions that were included and the requirement in some studies for the participating children and adolescents to report on their posttreatment anxiety levels, the majority of studies did include some requirements for minimum IQ or at least verbal IQ (VIQ) (generally a full scale or VIQ of 70), and behavioral standards (e.g., exclusion of violent subjects).

Types of interventions

This review included all treatments for anxiety where the large majority of participants were of mainstream school age, with ASD which occurred in schools, families or in clinical settings and that encompassed at least one of the elements outlined by Rotheram-Borus et al. (2012). As such, studies that focused solely on pharmacological interventions (e.g., selective serotonin reuptake inhibitors) were excluded from the review. Given the focus on psychoeducation and cognitive restructuring in Rotheram-Borus et al.'s (2012) recommendations, many of the included studies used a form of CBT. Those that did not examine CBT encompassed elements of relaxation, modeling, and self-monitoring.

Two examples of included studies identified by the initial search criteria are provided for illustrative purposes.

Chalfant, A. M., Rapee, R., & Carroll, L. (2006). Treating anxiety disorders in children with high functioning autism spectrum disorders: A controlled trial. Journal of Autism and Developmental Disorders, 37 (10), 1842-1857.

A family-based, cognitive behavioral treatment for anxiety in 47 children with comorbid anxiety disorders and High Functioning Autism Spectrum Disorder (HFA) was evaluated. Treatment involved 12 weekly group sessions and was compared with a waiting list condition. Changes between pre- and post-treatment were examined using clinical interviews as well as child-, parent- and teacher-report measures. Following treatment, 71.4% of the treated participants no longer fulfilled diagnostic criteria for an anxiety disorder. Comparisons between the two conditions indicated significant reductions in anxiety symptoms as measured by self-report, parent report and teacher report. Discussion focuses on the implications for the use of cognitive behavior therapy with HFA children, for theory of mind research and for further research on the treatment components.

Wood, J. J., Drahota, A., Sze, K., Har, K., Chiu, A., & Langer, D. A. (2009). Cognitive behavioral therapy for anxiety in children with autism spectrum disorders: A randomized, controlled trial. The Journal of Child Psychology and Psychiatry, 50 (3), 224–234.

Background: Children with autism spectrum disorders often present with comorbid anxiety disorders that cause significant functional impairment. This study tested a modular cognitive behavioral therapy (CBT) program for children with this profile. A standard CBT program was augmented with multiple treatment components designed to accommodate or remediate the social and adaptive skill deficits of children with ASD that could pose barriers to anxiety reduction. Method: Forty children (7-11 years old) were randomly assigned to 16 sessions of CBT or a 3-month waitlist (36 completed treatment or waitlist). Therapists worked with individual families. The CBT model emphasized behavioral experimentation, parent-training, and school consultation. Independent evaluators blind to treatment condition conducted structured diagnostic interviews and parents and children completed anxiety symptom checklists at baseline and posttreatment/postwaitlist. Results: In intent-to-treat analyses, 78.5% of the CBT group met Clinical Global Impressions-Improvement scale criteria for positive treatment response at posttreatment, as compared to only 8.7% of the waitlist group. CBT also outperformed the waitlist on diagnostic outcomes and parent reports of child anxiety, but not children's self-reports. Treatment gains were maintained at 3-month follow-up. Conclusions: The CBT manual employed in this study is one of the first adaptations of an evidence-based treatment for children with autism spectrum disorders. Remission of anxiety disorders appears to be an achievable goal among high-functioning children with autism.

The following is an example of study that was excluded due to it being a pharmacological only treatment:

Couturier, J., & Nicolson, R. (2002). A retrospective assessment of citalopram in children and adolescents with pervasive developmental disorders. Journal of Child and Adolescent Psychopharmacology, 12(3), 243–248.

Although selective serotonin reuptake inhibitors have been used to treat symptoms of aggression and anxiety in children and adolescents with pervasive developmental disorders (PDDs), there are no published reports of the use of citalopram in this population. The purpose of this study was to examine the benefits and adverse effects of citalopram in a group of children and adolescents with PDDs. Target behaviors included aggression, anxiety, stereotypies, and preoccupations. Seventeen patients with PDDs (14 with autistic disorder, three with Asperger's disorder) (mean age = 9.4 ± 2.9 years; range 4-15 years) were treated with citalopram for at least 2 months (mean duration of treatment =  7.4 ± 5.3 months; range 1-15 months). Treatment was initiated at a low dose (5 mg daily) and was increased by 5 mg weekly as tolerated and as necessary. The mean final dose was 19.7  ± 7.8 mg (range 5-40 mg). Outcome was based on a consensus between clinician and parents, using the Improvement item of the Clinical Global Impressions Scale as a guide. Ten (59%) children were judged to be much improved or very much improved regarding target behaviors. Core symptoms of PDDs (social interactions, communication) did not show clinically significant improvement. Citalopram was generally well tolerated, although four patients developed treatment-limiting adverse effects: two with increased agitation, one with insomnia, and one with possible tics. The results of this case series suggest that citalopram has beneficial effects on some interfering behaviors associated with PDDs with few adverse effects. Controlled trials are warranted.

Types of outcome measures

The primary outcome for included studies was anxiety, thus studies that focused on social skills interventions or other symptomatology of ASD as primary outcomes were excluded from this review. The measurement of anxiety (and related terms) had to be undertaken using valid and reliable approaches such as diagnostic interviews, screening instruments, observational ratings, and behavioral checklists—irrespective of the informant (e.g., student, parent, teacher).

Only the immediate posttreatment outcome is included in the current review, as the variety of follow-up schedules in the studies proved quite large, with 10 studies having no follow-up, 6 studies following-up less than 3 months after the end of the intervention, 4 studies after exactly 3 months, and 4 studies more than 3 months after the intervention (see Table 2). Compiling the results from different studies into ranges may have resulted in a loss of data integrity.

Types of settings

The settings in which the intervention was applied were real-world settings such as school or home. While 19 of the interventions were conducted in a clinical setting (either a university-based clinic or a community clinic, such as Child and Adolescent Mental Health), the intention of the studies was to address issues that were pertinent to the subjects’ lives—either at home or in school.

Electronic searches

A broad range of bibliographic databases were electronically searched for studies that matched our inclusion criteria:

Academic Search Complete (via EBSCO)

Our general search statement, set out below, was customized to fit the available search features of the bibliographic databases (see Appendix A in the Supporting Information Material for the customized statements). In this general search statement, the * symbol was used to indicate where our search covered variations in the root of the word.

(ASD OR Asperger* OR autis* OR Pervasive Developmental Disorder OR PDD NOS OR PDD unspecified) AND (Anxiety OR anxious OR internali* OR fear) AND (Student OR child* OR adolescen* OR preadolescen* OR pre adolescen* OR youth OR teen* OR teen age* OR young people OR young person OR boy OR girl) AND (Intervention OR treatment OR therap* OR psychotherap* OR evaluation OR outcome OR program OR trial* OR experimental OR control group OR random* OR best practi* or evidence based)

Searching other resources

Other resources that were search included gray literature, theses, conference proceedings, research reviews, purposely selected websites, reference lists from previously identified articles, and by contacting researchers and colleagues in the field and from the review's advisory group.

Google was used to identify gray literature from websites in the government, organization, and education domains (site:gov, site:edu, site:org). References were checked up to the first 200 results. The following search statement was used within the limits of the three specified domains was:

(autism OR autistic OR ASD OR asperger OR "PDD NOS" OR "PDD unspecified" OR "pervasive developmental") AND (anxiety OR anxious) AND (student OR child OR children OR adolescent OR youth OR teen OR boy OR girl) AND (intervention OR treatment OR therapy OR psychotherapy)-pubmed filetype:pdf

OpenGrey (European) was used to identify relevant European gray literature. The search statement was:

(asd OR Asperger* OR autis* OR Pervasive Developmental Disorder* OR PDD NOS OR PDD unspecified) AND (Anxiety OR anxious OR internali* OR fear) AND (Student OR child* OR adolescen* OR preadolescen* OR pre adolescen* OR youth OR teen* OR teenage* OR young people OR young person OR boy OR girl)

Institutional repositories: We searched the “Contents” of the Directory of Open Access Repositories (OpenDOAR) to identify research papers from institutional repositories. Our general search statement was customized to fit with the available search fields in the OpenDOAR Google Custom Search. Our search statement was:

(autism OR asd OR asperger OR "PDD NOS" OR "PDD unspecified" OR "pervasive developmental") AND (anxiety OR anxious) AND (student OR child OR children OR adolescent OR youth OR teen OR boy OR girl) AND (intervention OR treatment OR therapy OR psychotherapy) -pubmed –ncbi

Theses: For Networked Digital Library of Theses and Dissertations, a customized search strategy was used and the search limited by available population tags:

(asperger* OR autistic OR autism OR asd) AND (anxiety OR anxious) AND (treatment* OR intervention* OR therapy OR psychotherapy)

For WorldCat, a customized search strategy was used and the search limited by

Thesis/dissertation:

(Asperger* OR autistic OR autism OR asd) AND (anxiety OR anxious) AND (intervention* OR treatment* OR therap* OR psychotherapy*)

For American Doctoral Dissertations (EBSCO), the search statement was:

(Asperger* OR autism OR autistic OR ASD OR PDD OR "Pervasive Developmental") AND (anxiety OR anxious OR fear OR internal*)

Conference proceedings: In addition to conference proceedings and papers indexed in our selected databases, we identified conference literature via a search on SCOPUS which is a multidisciplinary database. This search was limited to conference papers in the collections other than the Social Sciences, Humanities, or Neuroscience as these collections were covered elsewhere. The search statement was:

TITLE-ABS-KEY((asperger* OR autis* OR asd OR "Pervasive Developmental" W/0 disorder* OR "PDD NOS" OR "PDD unspecified")) AND TITLE-ABS-KEY((anxiety OR anxious OR internali* OR fear)) AND TITLE-ABS-KEY((student OR child* OR adolescen* OR preadolescen* OR (pre W/0 adolescen*) OR youth OR teen* OR (teen W/0 age*) OR "young people" OR "young person" OR boy OR girl)) AND TITLE-ABS-KEY((intervention OR treatment OR therap* OR psychotherap* OR evaluation OR outcome OR program* OR trial* OR experimental OR (control W/0 group) OR random* OR (best W/0 practi*) OR "evidence based")))

Research reviews: Wherever possible, the following search statement was executed in the list of resources below:

(asperger OR autism OR autistic OR ASD OR "pervasive developmental" OR "PDD NOS")

Campbell Library

Where further refinement was necessary, we added additional search terms relating to the concept of anxiety.

Targeted searches of selected websites: We explored the websites of selected agencies, research centers and professional associations including the following:

Agency for Healthcare Research and Quality

Reference lists: We searched the reference lists of previously published reviews and meta-analyses that we identified as well as the reference lists of each of the studies identified for our analysis.

Current literature: After our initial search, we set up alerts in Google Scholar and, where possible, in the bibliographic databases, in order to identify any new literature within the time of our study. New table of content alerts were also set up for key journal titles including Research in Autism Spectrum Disorders, Autism, and the Journal of Autism and Developmental Disorders. Alerts were manually scanned for any new references that fitted our search criteria.

Colleagues: Contact was also made with researchers and colleagues in the field and from the review's advisory group, to identify any additional studies, particularly those that might have been ongoing or unpublished at the time of our work.

Ongoing trials: We identified current and ongoing trials via the following trial registries:

International Clinical Trials Registry Platform Search Portal

We searched for the following terms:

(Asperger OR autism OR autistic OR asd OR "pervasive developmental" OR "pdd nos")

Where further refinement was required, we included anxiety and/or trials limited to children.

Publication date range

Our searches were limited to a publication date range of 1996–2018. We selected 1996 as the earliest publication date in order to narrow the scope of interventions to current approaches used in the last 20 years. Given the development of understanding in this field, we believe that interventions before this date would be less progressive in their approach. The initial database searches were conducted between May 4, 2017, and June 7, 2017. Updates of database searches were conducted via alerts based on the original search statements or by rerunning search statements, up to December 31, 2018.

Other criteria

The searches in our selected sources were not restricted by geography, language, publication type, or by publication status. However, the selected sources are focused on the English language in keeping with our database subscriptions and the primary language of the authors.

Selection of studies

As a first step in the screening process, four reviewers independently assessed titles and abstracts of a purposely heterogeneous subset of five studies identified through the searches. The purpose of this step was twofold: first, it determined their potential eligibility for inclusion in the review and second it served to develop a common understanding and application of inclusion criteria. Once a consensus regarding the application was reached, all abstracts were assessed by at least two reviewers. Where two reviewers disagreed regarding the inclusion of an abstract in the study, resolution was sought through discussion with the full project team. At the end of this step, studies that clearly did not meet the criteria, as well as duplicates, were removed.

Data extraction and management

Full-text articles were then retrieved for the included abstracts. Reviews of the full-text articles were undertaken independently by two reviewers (K. H. and K. D.). Any discrepancies were resolved through discussion and, where necessary, further details added to data definitions which may have been unclear. The checklists used during this review phase are presented in Appendices B and C in the Supporting Information Material, along with a summary of the appraisal results for excluded papers.

Once papers were selected for inclusion in the review, data were independently extracted in duplicate by authors K. D. and K. H. using the standardized data extraction tools from JBI SUMARI (see Appendix D in the Supporting Information Material for the quantitative data extraction form, which was operationalized in a spreadsheet). Discrepancies were checked and resolved. The data extracted included specific details about the interventions, populations, study methods, and outcomes of significance to the review question and specific objectives.

Assessment of risk of bias in included studies

The risk of bias assessment was carried out by authors K. H. and K. D. The assessment was informed by initial data extraction in JBI SUMARI and conducted using the Cochrane Collaboration's guidelines for assessing risk of bias (Higgins, Altman, & Sterne, 2011). Risk of bias in the selected studies was rated as high risk (bias that potentially reduces the reliability of the results), or low risk (bias that is unlikely to alter the results), with an unclear category used in cases in which there was insufficient information in the published study for the judgment of bias to be made. As the majority of included studies were RCTs, the risk of bias assessment focused on methodological issues pertaining to this form of study—sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, attrition, selective reporting, and other sources of bias. As a consequence, the quasi-experimental studies that were included in this review received higher ratings of risk of bias, particularly in terms of selection bias.

Measures of treatment effect

Only one outcome per respondent group was used in the quantitative syntheses to avoid double counting. In cases where there was more than one outcome measure per respondent available, we selected an outcome based on assessments of validity published in Wigham and McConachie (2014) and Lecavalier et al. (2014), the frequency of use across the included studies, and the availability of data appropriate for the meta-analyses.

Separate statistical analyses were conducted based on the informant for the outcome measures—namely parent, clinician, and self/student. While two of the included studies collected anxiety outcome measures from teachers of the subjects (Chalfant_2007; Luxford_2017), there were not enough teacher-informants to conduct a separate analysis. In addition, the extent of missing data for these measures was enough to raise concerns about reliability.

In accordance with the JBI SUMARI meta-synthesis program the SMD, reported as Cohen's d, and its 95% CI was used as the summary estimate of treatment effect size and based on the posttreatment/wait-list scores reported in each study. This summary statistic was selected as all studies included continuous measures and all measures were in the same direction (i.e., higher scores indicating higher levels of behavior or impact of symptoms) and thus no adjustments were required. In addition to Cohen's d—although not available in the JBI SUMARI program at the time of analyses—Hedges’ g is another method (i.e., formulae) commonly used for the computation of SMD. Cohen's d and Hedges’ g differ in that the latter uses the version of the standard deviation formula which divides by N-1, whereas the former divides by N. While, therefore, Hedge's g is often preferred for reviews involving studies with small sample sizes its use would not have led to different conclusions having been drawn from the results of the current review.

Unit of analysis issues

Some of the included studies deviated from standard treatment versus control comparisons, in employing crossover designs (Pryor_2016), inclusion of more than one treatment group (Sofronoff_2005) or inclusion of a control group without ASD who also received treatment (vanSteensal_2015).

For the crossover design, data from the baseline and the end of the first phase (prior to crossover) were used, effectively treating the alternative treatment group as a “TAU” control. Sofronoff_2005 included two treatment arms, one in which children received CBT on their own and another in which their parents participated in the treatment with them. While previous reviews have pooled the results of these two treatment arms (e.g., Kreslins et al., 2015; Ung et al., 2015), we elected to include them separately, as they do represent two different forms of intervention.

The study design of vanSteensal_2015 was relatively more complex, compared to other included studies, in that it included two treatment groups—one with ASD and anxiety disorders, one with anxiety disorders but no ASD diagnosis—and a wait-list control (WLC) subgroup of the ASD group. While the argument could have been made to exclude this study for a lack of formal diagnosis of ASD for those participants in the “anxiety disorders only” comparison group, it was decided to retain the study and focus on the immediate treatment versus WLC comparison within the ASD group.

Assessment of heterogeneity

Quantitative data were, where possible, pooled by way of statistical meta-analysis. Weighted mean differences and their 95% CIs were calculated for analysis. Heterogeneity was assessed statistically using the standard χ ² and also explored using subgroup analyses based on the different study designs included in this review.

Assessment of reporting biases

Assessing risk of publication bias was an important task because of its potential influence on estimates of intervention effects. This review analyzed possible publication bias by implementing the trim-and-fill method (Duval & Tweedie, 2000; Schwarzer, 2007), providing an initial assessment of whether unpublished data on ASD and anxiety interventions (likely to have null results) was evident (Uljarević & Hamilton, 2013).

Data synthesis approach

Separate statistical analyses were carried out for clinician-reported, parent-reported, and self-reported outcome measures of anxiety. Studies were also coded dichotomously for two possible moderator variables: (a) family involvement or student-only and (b) group or individual format. SMD was chosen as the summary estimate of treatment effect, appropriate for the continuous outcome measures that were being analyzed. Often, SMD s of 0.2, 0.5, and 0.8 are taken to be indicative of small, moderate, and large effects, respectively (Cohen, 1988). However, heeding concerns by Valentine and Cooper (2003) and Lipsey et al. (2012) that effect in the field of education are likely to be small and risk being overlooked if based on Cohen's interpretation, effect sizes were interpreted in an educational context using the metric developed by Higgins et al. (2013) for the UK Education Endowment Foundation. Accordingly, SMDs of 0.05, 0.19, 0.45, and 0.70 were taken to be indicative of low, moderate, high, and very high effects, respectively. Moreover, these can be interpreted, respectively as, 1 month, 3 months, 6 months, and 9 months additional developmental progress. A random effects meta-analysis was employed due to the variability in outcome measurement instruments and interventions across the included studies.

Using the JBI SUMARI meta-synthesis program, SMD (Cohen's d) estimates were calculated based on the posttreatment mean scores and standard deviations provided in each study. Since the direction of the scales was the same for all outcome measures, no adjustments of the scores were required. The statistical significance level was set at p < .05. Forest plots were used to illustrate results from individual studies. In the case of multiple treatment arms, such as Sofronoff_2005, the scores of both intervention groups (ITs) were compared to the control group score. Similarly, if a study reported more than one outcome measure for a respondent, then both outcomes have been reported, rather than presenting an average score as previous systematic reviews have done (e.g., Kreslins et al., 2015; Lang, Regester, Lauderdale, Ashbaugh, & Haring, 2010; Perihan et al., 2019; Sukhodolsky et al., 2013; Ung et al., 2015).

Sensitivity analysis

Given the diversity of interventions and the potentially small sample of included studies within each intervention type, it was important to conduct a sensitivity analysis of the impact of a single study, particularly if it is an outlier, on the overall observed effect size for interventions in any meta-analysis. The main sensitivity analysis conducted, focused on excluding single studies which may have had an unduly large effect on the results. Results were then compared to provide an indication of the robustness of the review's findings.

Results of the search

The search of the databases yielded 3,417 records, with an additional 177 records identified through other sources. Removal of duplicates resulted in a total of 2,337 records, which were screened based on the title and abstract by three of the authors, with 2,218 records being excluded.

The remaining 119 full-text articles and theses were assessed for eligibility and 94 excluded as not meeting the search criteria. One further study was subsequently excluded as a reanalysis of data reported in an already included study. A total of 24 studies were thus included in the quantitative meta-synthesis. A flow diagram of the study selection is presented in Figure 1.

OPEN IN VIEWER

Figure 1

PRISMA flow diagram outlining process of study selection

Included studies

Twenty-four studies examining interventions for anxiety for children and adolescents with ASD are included in this review, identified by first author and publication date: Chalfant_2007; Clarke_2017; Conaughton_2017; Corbett_2017; Fujii_2013; Hepburn_2016; Luxford_2017; MacKinnon_2014; McConachie_2014; McNally-Keehn _2013; Murphy_2017; Ohan_2016; Piravej_2009; Pryor_2016; Reaven_2009; Reaven_2012; Sofronoff_2005; Storch_2013; Storch_2015; Sung_2011; vanSteensel_2015; White_2013; Wood_2009; Wood_2015.

Study location

Five of the studies were conducted in Australia, 12 in the United States, 4 in the United Kingdom (England), and 1 each in Singapore, Thailand, and The Netherlands. Twenty of the studies were set in clinical settings attached to medical or university institutions, while two were home-based interventions and two were school-based.

Study design

Six of the studies were classified as quasi-experimental designs, in that they employed a control or comparison group that was not randomly assigned, while the remaining 18 studies used a randomized WLC design (with varying degrees of fidelity).

Participants

A total of 931 (764 male and 167 females) participants most of whom were of mainstream school age (6–16) were included in these studies (note that subjects in pre-post studies only counted once). Six of the studies included older adolescents (up to 19 years of age) while one study included children as young as 3 years old. All studies had inclusion criteria of a documented diagnosis of ASD (often confirmed during intake assessments) as well as either a concurrent diagnosis of an anxiety disorder (again, confirmed during intake) or reports from parents or educators of significant levels of anxiety.

The majority of the studies (21 of 24) limited participants to individuals with functioning above a certain level of cognitive ability, most commonly a full scale or verbal IQ of 70 or above, or in one case, the ability to read and write (in English) at an 8-year-old level as a minimum. Only three studies did not place restrictions (either explicitly or de facto by requiring that participants be attending a mainstream school) on the cognitive functioning of participants (Hepburn_2016; MacKinnon_2014; Piravej_2009). Further information, where available, on the cognitive functioning of participants in each study, is presented in Table 2.

Interventions

Twenty-two of the studies used a CBT intervention, with some developed specifically for use with participants with ASD (e.g., Exploring Feelings; BIACA and MASSI). Of these, 15 studies involved interventions that included parental involvement (Chalfant_2007; Fuji_2013; Hepburn_2016; MacKinnon_2014; McConachie_2014; McNally-Keehn_2013; Ohan_2016; Reaven_2009; Reaven_2012; Storch_2013; Storch_2015; van Steensel_2015; White_2013; Wood_2009, Wood_2015), and six studies involved student-only CBT treatments (Clarke_2017; Conaughton_2017; Luxford_2017; Murphy_2017; Pryor_2016; Sung_2011). One study (Sofronoff_2005) had two treatment arms, with and without parental involvement.

Rather than a CBT intervention, Corbett_2017 used peer-mediated theater therapy in a group context to address social anxiety by building the social-emotional skills of participating adolescents, while Piravej_2009 examined the benefits of Thai traditional massage in a one-on-one context to internalizing, externalizing, and sleep behaviors compared to the standard sensory integration treatment available to participants.

In development since previous reviews of treatments for anxiety in the ASD population, three studies examined the effectiveness of CBT treatments (either published or newly developed) designed for computer delivery (Conaughton_2017; Pryor_2016) or videoconferencing (Hepburn_2016).

Outcome measures

Across the 24 studies, 12 outcome measures of anxiety were used to varying extent and with different respondents (clinician, parent, child). Most of these measures are standardized, validated measures of anxiety for use by clinicians (e.g., psychologists). The measures, as indicated in Table 2, were the: Anxiety Disorders Interview Schedule (ADIS: Silverman & Albano, 1996); the Child Anxiety Life Interference Scale (CALIS: Lyneham et al., 2013); the Childhood Anxiety Sensitivity Index - Anxiety (CASI-Anx: Sukhodolsky et al., 2008); Conners’ Parent Rating Scales—Anxiety (CPRS: Conners, 1989); the Multidimensional Anxiety Scale for Children (MASC: March, Parker, Sullivan, Stallings, & Conners, 1997); the Pediatric Anxiety Rating Scale (PARS: RUPP Anxiety Study Group, 2002); the Revised Child Anxiety and Depression Scales (RCADS: Chorpita, Moffitt, & Gray, 2005); the Revised Children's Manifest Anxiety Scale—anxious arousal subscale (RCMAS: Reynolds & Richmond, 1978); the Screen for Child Anxiety Related Disorders (SCARED: Birmaher et al., 1999); the Spence Child Anxiety Scale (SCAS: Spence, 1998); the State-Trait Anxiety Inventory for Children (STAI: Spielberger, Gorsuch, Lushene, Vagg, & Jacobs, 1983); and the Social Worries Questionnaire (SWQ: Spence, 1995). A summary of some of the key elements of the included studies is presented in Table 2.

Comparisons to previous reviews

Eight of the studies had not been included in previous reviews of interventions for anxiety in the ASD population, predominantly due to being published after the reviews were completed and because the current review also included theses and dissertations that were experimental or quasi-experimental studies. Table 3 presents the prior reporting of studies in five previous systematic reviews undertaken by Lang et al. (2010), Sukhodolsky et al. (2013), Kreslins et al. (2015), Ung et al. (2015), and Perihan et al. (2019).

Table 4 lists the studies identified in the latest review by Perihan et al. (2019) that were not included in this review with reasons for noninclusion. As can be seen, the main reason for not including studies in this current review is their observational design.

Excluded studies

As indicated in Figure 1, 95 of the full-text papers retrieved were excluded from the current review: 19 did not report levels of anxiety among participants, 16 had no outcome measure of anxiety, 15 did not employ a control or comparison group, 10 had a participant age range out of scope (and no potential for isolating participants within scope), 10 studies were classified as purely observational (pre- and postintervention in single group only), seven had no formal diagnosis of ASD in their participants, six were secondary or follow-up analyses of previous studies that focused on aspect outside of inclusion criteria, four did not include an intervention that met criteria, two were proposals or study protocols only (no results were presented), and another six did not provide sufficient information to satisfy the selection criteria or had other issues. As summarized below and detailed in Appendices B and C in the Supporting Information Material, these 95 excluded articles included those studies that might reasonably have been expected to be included, such as those included in previous reviews, but which did not meet the inclusion criteria of the current review:

No anxiety present in participants=> 19 studies

Selection bias

Thirteen of the studies performed adequate random sequence generation, either manually (using staff unrelated to the study, Corbett_2017; Murphy_2017; Sung_2011; White_2013) or generated by computer (Clarke_2017; Luxford_2017; Reaven_2012; Storch_2013; Storch_2015; Wood_2009; Wood_2015). Block randomization procedures, stratified by demographic variables, were used in Fuji_2013, McNally-Keehn_2013, and Piravej_2009. Despite being identified as randomized controlled trials, Chalfant_2007, Pryor_2016, and Sofronoff_2005 did not provide suf ficient information about randomization methods or procedures to assess potential bias.

It should be noted that four of the included studies were classified as quasi-experimental studies, in that they included a comparison group but that allocation to groups was not random. Of the quasi-experimental studies, Hepburn_2016 used a pair-wise matching scheme for allocation to groups, Ohan_2016 claimed that the order of enrollment “approximated” randomization, while Reaven_2009 had subjects act as their own WLCs.

Allocation concealment was not detailed in the majority of studies, although Conaughton_2017 and Wood_2017 maintaining concealment by conducting baseline measures prior to randomization, and McConachie_2014, Murphy_2017, and Storch_2015 by concealing treatment group allocation from the researchers and independent evaluators throughout the studies.

Performance and detection bias

Performance bias was universally high, due to the nature of the interventions, as it was not possible to blind participants from their treatment group allocation.

Detection bias was higher among studies that used outcome measures based on the reports of participants themselves or their parents (many of whom had also participated in family-based interventions), but lower in studies that used reports from clinicians or teachers who were blinded to IT allocation. For those studies that used multiple outcome measures from different informants, the risk of performance bias was rated for each outcome measure separately.

Attrition bias

Attrition was rated as low in studies that had little to no attrition in subjects, who evaluated the effect of attrition by comparing therapy completer analyses with “Intent to Treat” analyses (i.e., analyses using original samples and data imputation techniques), or who compared the profile of study dropouts with completers with no statistically significant differences found. Attrition bias was fairly low across the studies, with only Ohan_2016 and Reaven_12 being rated as high. Ohan_2006 reported that six of the treatment families failed to complete their treatment (a loss of 25% of their sample), while Reaven_12 reported a treatment-completer sample of 47, but only provided baseline data for 43 participants (IT = 20 and TAU = 23) and used “last observation carried forward” imputation for missing data.

Reporting bias

There was no evidence or suggestion of reporting bias in any of the studies. However, a risk of bias assessment was undertaken by rating each study as high, unclear, or low risk of bias against five attributes. A summary of the risk of bias assessment for the included studies is presented alongside each forest-plot. In addition, publication bias was assessed using a funnel plot.

Treatment efficacy

Twenty-four studies reported outcome measures from one or more informants (clinicians, parents, student self-reports), with 60 informant reports in total, along with several teacher reports (see Table 2). These studies involved 1,020 participants (524 in the treatment group and 496 in the control group—note that participants in pre-post studies are counted twice). Rather than averaging all reports within each study across different measures to derive an overall treatment effect, the reports with lower risk of bias were selected. For half of the studies, the clinician report was used. For seven of the studies, the averaged parent- and student-reported SCAS (six instances) or SCARED (one instance) were used. For the other five studies, a parent-only or student-only report was used. A random effects meta-analysis of the 24 studies revealed a statistically significant treatment effect for interventions for reducing anxiety in mainstream school-aged students with ASD. The overall SMD was d = −0.83 (95% CI: −1.16, −0.51; z = −5.03, p < .01) which can be considered a very high effect. Based on these measures, the anxiety levels in the treatment groups were significantly lower than those seen in the control groups at posttreatment.

Considerable heterogeneity across the studies and assessments was detected (I ² = 83%). Figure 2 presents a forest plot illustrating the results. Visual inspection of the both the forest plot and the funnel plot, as presented in Figure 3, identified the SMD score reported by Chalfant_2007 as an outlier. A sensitivity analysis was carried out by removing this study, reducing the overall SMD to −0.71 (95% CI: −0.97, −0.46; z = −5.42, p < .01). While removal of the outlier did reduce the overall effect size, the difference between treatment and control conditions at posttreatment remained very high and statistically significant.

OPEN IN VIEWER

Figure 2

Forest plot of studies included in the meta-analysis, with risk of bias summary

OPEN IN VIEWER

Figure 3

Funnel plot of the 24 included studies

Clinician reported outcome measures

Thirteen studies involving a total of 526 students (267 in the treatment condition and 259 in the control condition) reported one or more clinician-reported outcome measure, with 18 reports in total. The three measures used by clinicians were the ADIS (A, 12 instances), the CASI-Anx (C, 2 instances), and the PARS (P, 4 instances). All studies reported greater improvements posttreatment in the treatment condition compared to the control condition, although Murphy_2017) reported nonsignificant mixed results. The overall SMD was d = −0.84 (95% CI: −1.15, −0.54; z = −5.43, p < .01) which can be considered a very high effect, potentially equivalent to 10 months developmental progress. Based on these measures, the anxiety levels in the treatment groups were significantly lower than those seen in the control groups at posttreatment. Considerable heterogeneity across the studies and assessments was detected (I ² = 70%). Figure 4 presents a forest plot illustrating the results.

OPEN IN VIEWER

Figure 4

Forest plot of clinician-reported outcome measures with risk of bias summary

Parent reported outcome measures

Nineteen studies reported one or more parent reported outcome measures, with 21 reports in total. These studies involved 819 participants (412 in the treatment group and 407 in the control group). The five measures used with parents were the SCAS (S, 11 instances), the MASC (M, 5 instances), SCARED (D, 3 instances), CALIS (C, 1 instance), and the CPRS (N, 1 instance). The overall SMD was d = −0.68 (95% CI: −1.05, −0.31); z = −3.64, p < .01), indicating that the difference between the treatment and control groups at posttreatment reached significance. There was significant heterogeneity across the included studies and measures (I ² = 85%). Figure 5 presents a forest plot of the results, along with a summary of the risk of bias. Like other systematic reviews (e.g., Kreslins et al., 2015), the SMD reported by Chalfant_2007 was assessed as an outlier, being substantially higher than the others reported. A sensitivity analysis was carried out. Once this outlier was removed, the overall SMD decreased to −0.53 (95%CI −0.76, −0.31; z = −4.73, p < .01). Although the summary estimate decreased following removal of the outlying study, the high treatment effect remained statistically significant and potentially equivalent to 7-month developmental progress.

OPEN IN VIEWER

Figure 5

Forest plot of parent-reported outcome measures with risk of bias summary

Student self-reported outcome measures

Self-reported outcome data from 734 student participants (370 in the treatment group and 364 in the control group) from 21 outcome reports across 17 studies were synthesized. The main student outcome measure used by 10 studies was the SCAS (S). The SCARED (D), MASC (M), RCMAS (R), and RCADS (V) had each been used by two of the studies, while the CALIS (L), STAI (T), and SWQ (W) had each been used by one of the studies. The overall SMD was d = −0.58 (95% CI: −0.95, −0.21; z = −3.06, p = .002) with a significant difference (high effect size, equivalent to 7-month progress) between the treatment and control conditions at posttreatment. Figure 6 presents a forest plot of the student results. There were high levels of heterogeneity across the studies (I ² = 85%). The two SMD scores reported by Chalfant_2007 were again identified as outliers and a sensitivity analysis was undertaken. Removal of this study reduced the overall SMD to −0.35 (95% CI: −0.55, − 0.15; z = −3.41, p < .01). This difference between treatment and control conditions at posttreatment of a moderate effect was significant and equivalent to 4-month progress.

OPEN IN VIEWER

Figure 6

Forest plot of self-reported outcome measures with risk of bias summary

Moderator analysis

Also of interest in this study, was the potential moderating effects of a) family involvement in sessions and b) the individual or group nature of sessions. These two moderators were analyzed to help address the lack of similar investigations in previous reviews published, and to draw comparison to the most recent review by Perihan et al. (2019), which did consider the potential moderator of parental involvement.

Family involvement

Two groups were created based on family involvement, in order to compare the outcomes of treatments with parental involvement (n = 15, excluding outlier Chalfant_2007) to treatments without parental involvement (n = 9). It should be noted that the Piravej et al. (2009) study involving Thai massage was categorized as being without family involvement even though for safety reasons, the parent was present in the room. Similar to Perihan et al.'s (2019) findings, treatments that had family involvement resulted in a larger overall effect size than treatments without parental involvement, a difference equivalent to 2-month additional progress. The SMD with family involvement yielded a very high significant effect size of d = −0.74 (95% CI: −1.06, −0.42; z = −4.55, p < .01). In comparison, the SMD without family involvement resulted in a high significant effect size of d = −0.60 (95% CI: −1.03, −0.17; z = −2.73, p = .006). There were high levels of heterogeneity across the two groups of studies (I ² = 70% with involvement; 75% without involvement). Figure 7 presents a forest plot of the moderating effect of family involvement.

OPEN IN VIEWER

Figure 7

Forest plot comparing treatments with and without family involvement

Individual or group treatment

In order to compare the potential moderating effect of treatment format, two groups of studies were formed and assessed based on whether the treatment was administered on an individual basis (n = 9), like the Thai massage (Piravej et al., 2009), or in a group setting (n = 14), such as the peer-mediated, theater-based intervention (Corbett et al., 2017). The forest plot presented in Figure 8 suggests that individually based treatments resulted in a larger overall effect size than group-based treatments, a difference equivalent to 7-month additional progress. The SMD yielded a very high significant effect size of d = −1.24 (95% CI: −1.75, −0.74; z = −4.87, p < .01). In comparison, the SMD for group treatments resulted in a moderate significant effect size of d = −0.37 (95% CI: −0.54, −0.19; z = −4.10, p < .01). There was a high level of heterogeneity across studies involving individual treatments (I ² = 80%) and a low level of heterogeneity across the studies involving group treatments (I ² = 0%).

OPEN IN VIEWER

Figure 8

Forest plot comparing treatments that involve individual or group participation