Implementation and effectiveness of non-specialist mediated interventions for children with Autism Spectrum Disorder: A systematic review and meta-analysis

Introduction In recent years, several non-specialist mediated interventions have been developed and tested to address problematic symptoms associated with autism. These can be implemented with a fraction of cost required for specialist delivered interventions. This review represents a robust evidence of clinical effectiveness of these interventions in improving the social, motor and communication deficits among children with autism. Methods An electronic search was conducted in eight academic databases from their inception to 31st December 2018. A total of 31 randomized controlled trials were published post-2010 while only 2 were published prior to it. Outcomes pertaining to communication, social skills and caregiver-child relationship were meta-analyzed when reported in > 2 studies. Results A significant improvement was noted in child distress (SMD = 0.55), communication (SMD = 0.23), expressive language (SMD = 0.47), joint engagement (SMD = 0.63), motor skills (SMD = 0.25), parental distress (SMD = 0.33) parental self-efficacy (SMD = 0.42) parent-child relationship (SMD = 0.67) repetitive behaviors (SMD = 0.33), self-regulation (SMD = 0.54), social skills (SMD = 0.53) symptom severity (SMD = 0.44) and visual reception (SMD = 0.29). Conclusion Non-specialist mediated interventions for autism spectrum disorder demonstrate effectiveness across a range of outcomes for children with autism and their caregivers.


Introduction
In recent years, several non-specialist mediated interventions have been developed and tested to address problematic symptoms associated with autism. These can be implemented with a fraction of cost required for specialist delivered interventions. This review represents a robust evidence of clinical effectiveness of these interventions in improving the social, motor and communication deficits among children with autism.

Methods
An electronic search was conducted in eight academic databases from their inception to 31 st December 2018. A total of 31 randomized controlled trials were published post-2010 while only 2 were published prior to it. Outcomes pertaining to communication, social skills and caregiver-child relationship were meta-analyzed when reported in > 2 studies. Introduction attention and engagement and communication [10]. SENSE Theater involves children with autism and their peers in theaters aiding in an understanding of socially expected behaviors. Video modelling involves videos depicting individuals demonstrating expected behavior to the children with autism [11]. And pivotal response treatment (PRT) trains children in pivotal behaviors required for daily functioning, such as social initiations and responsivity, self-efficacy, and motivation [12].

Results
There is a paucity of comprehensive systematic reviews and meta-analysis of randomized controlled trials detailing content of different interventions, their effectiveness across different outcomes and quality of available evidence. Therefore, the present review was designed to, a) assess the effectiveness of non-specialist delivered or mediated interventions in ASD; b) systematically evaluate relevant implementation processes involved in these non-specialists delivered interventions for ASD, and c) and to rate the quality of evidence across different outcomes using the World Health Organization's recommended Grading of Recommendations Assessment, Development and Evaluation (GRADE) criteria (described below).

Methods
This review was conducted as per the updated PRISMA guidelines [13] (S1 File), and the protocol registered in PROSPERO (CRD42017066009).

Search process & selection criteria
An academic search was conducted in eight electronic databases including PubMed, Scopus, Web of Science, POPLINE, New York Academy of Medicine, PsycINFO, Psycharticles, and CINAHL, from their inception to 31 st December 2018, using following search terms (S2 File): (("autism spectrum disorder" OR Asperger OR autis � OR "pervasive developmental disorder" OR "childhood disintegrative disorder") AND (intervention OR treatment OR RCT OR trial) AND (parent-mediated OR parent-delivered OR "non-specialist mediated" OR "non-specialist delivered" OR teacher-mediated OR "teacher delivered" OR "aide delivered" OR "aide mediated" OR "peer delivered" OR "peer mediated")). No restrictions or database filters regarding language, time period or publication year were applied.
Three independent reviewers screened the aforementioned databases for eligible studies based on their titles and abstracts, followed by screening of full texts. All discrepancies among reviewers were resolved through discussion between reviewers and senior authors. All studies were assessed for eligibility against following criteria: Inclusion criteria 1. Individuals with a clinical diagnosis of ASD, screened for ASD using questionnaires or clinician diagnosis, Asperger's syndrome, and childhood disintegrative disorder were included.
2. Only studies assessing the efficacy through randomized controlled trials were included. 3. No restriction to age, gender, language, country, socioeconomic status or time period was applied.
4. Studies focusing on the parent, caregiver, peer, teacher or any other non-specialist mediated or delivered interventions were included.

Exclusion criteria
1. Overlapping data sets reporting results from same study.
2. Studies which are not randomized controlled trials will be excluded.
3. Books, conference papers, theses, editorials, case reports, case series, reviews and articles without available full text will be excluded.
5. Non-original articles (reviews and analyses) 6. In Vitro studies and non-human trials.
7. Interventions conducted among adults with ASD were excluded.

Data extraction, risk of bias assessment & GRADE evidence
All data were extracted independently by three teams of reviewers using manualized data extraction forms and any disagreements among the reviewers, were resolved through discussion in conjunction with a senior author. Data pertaining to participant characteristics, study setting, nature of intervention and outcomes will be extracted. For outcomes, an apriori decision was taken to include all types of psychometric testing whether conducted by specialists, teachers or parents. A variety of psychometric scales used for measurement of symptoms of autism are reported in the literature. We conducted a thorough audit of included studies to identify the psychometric scales used and categorized them under a unifying category. For instance, total symptom severity comprised of several scales such as Autism Diagnostic Observation Schedule; Autism Behaviour Checklist; Vineland Adaptive Behaviour Scale and Childhood Autism Rating Scale among others.
If there was a trial with more than one publication, preference was given to the primary publication. A US board certified child psychiatrist also devised a taxonomy form for active ingredients of interventions with detailed instructions regarding content, strategies and elements of interventions. Moreover, two authors assessed the quality of the studies without blinding to authorship or journal, using The Cochrane tool for randomized controlled trials, against several matrices: a) sequence generation, b) allocation concealment, c) blinding of participants and personnel, d) blinding of outcome assessment, e) incomplete outcome data, f) selective reporting and g) other bias" [14].
The meta-analytical evidence in present review was assessed for its quality using the recommendations outlined by the Grading of Recommendations, Assessment, Development and Evaluations (GRADE) Working Group [15]. These recommendations allow for rating of meta-analytical ranging from high to very low based on its study design, risk of bias, inconsistency, indirectness in targeted population and suitability of intervention, imprecision, publication bias and magnitude of effect size [15]. The evidence is judged across outcomes, where each concern in aforementioned matrices is rated as serious or very serious, stepping down the quality of evidence by one or two levels respectively [15].

Data analysis
Descriptive statistics pertaining to characteristics of the study and implementation processes including elements of interventions were explored using Microsoft Excel. Thereafter, using Comprehensive meta-analysis software, a series of meta-analyses were run for similar outcomes assessed post-intervention or primary time points [16]. Only those outcomes were introduced in the meta-analysis that were reported in � 2 studies. Studies reporting similar outcomes were pooled together, weighted by employing inverse variance method, thus, estimating pooled effect sizes expressed as standardized mean differences with 95% confidence intervals (CI) [16]. Depending on the extent of heterogeneity, data were pooled together using either the fixed or random effects. Heterogeneity was considered significant at a cut off value � 40%. However, we applied random effects analysis for all of the outcomes because of heterogeneity in assessment of outcomes across included studies [17]. Sensitivity analyses was conducted by excluding individual studies individually to ascertain their effects on the pooled effect size. When � 5 studies reported an outcome, publication bias was assessed for asymmetry by visualizing the Begg's funnel plot and Egger's regression statistics (P � 0.10) [18]. Pooled effect sizes were then adjusted for publication bias using Duvall & Tweedie's Trim and Fill method [19]. To ensure an appropriate statistical power, when there were �4 studies reporting an outcome among different groups, subgroup analyses were conducted. Lastly, an outcome reported in � 10 studies allowed meta-regression analyses to identify potential moderators of intervention effects among children with autism [20].

Results
Searching of academic databases yielded a total of 659 non-duplicate references to be screened based on their titles and abstracts. Out of these, 596 citations were excluded, retaining 63 full texts. Thereafter, 33 randomized controlled trials were deemed eligible after screening of these full texts against inclusion and exclusion criteria. Detailed results have been presented in PRISMA flowchart (Fig 1). S3 File presents effect sizes, means (SD) and subgroups for individual studies.

Study characteristics
A total of 31 studies were published post-2010 while only 2 were published prior to it [21,22]. Only 2 studies was conducted in a low and middle income country [6] (Divan, 2018) while others were conducted in high income countries including USA (n = 21), Australia (n = 3), UK (n = 2), Canada (n = 2) and 1 each in Belgium, Norway, and Netherlands. Only two of the studies was a cluster randomized control trial while rest were individual RCTs [23] (Morgan et al, 2018). National Institutes of Health were the major funder of these trials (n = 11). A total of 24 studies were conducted in urban areas and 3 in both rural and urban areas (missing n = 3). Table 1 provides further details on these variables.

Intervention characteristics
Most of the interventions took place in the community (n = 9) and educational settings, (n = 10) followed by home (n = 7), healthcare setting (n = 2), videoconferencing (n = 1), and the rest in mixed settings (n = 3). The majority of the interventions (n = 14) had employed the Autism Diagnostic Observation Schedule (ADOS) for screening of children with autism. Other scales such as Childhood Autism Rating Scale (CARS), Modified Checklist for Autism in Toddlers (M-CHAT) and Autism Behavior Checklist (ABC), Checklist for Autism Spectrum Disorder (CASD) were also employed (Fig 2 and Fig 3).

Density of dosage
The mean number of sessions was 53.29 (SD = 158.07), ranging from 5 [29] to 834 [23] sessions, while this information was not reported by two studies [33]. Duration of each session was averaged at 130.63 minutes (SD = 281.47) and ranged from 10 minutes (daily) to 1500 minutes per week [26,27], and was not reported by two studies [33]. Mean duration for program was 22.62 weeks (SD = 33.34), lasting a minimum of 6 weeks [34] and maximum of 182.50 weeks [23]. Booster sessions were conducted in a total of six trials [26,27,29,35] (Brian and Matthews). These variables are reported in greater detail in Table 2.
Sensitivity analyses revealed that removal of specific trials led to significant effect sizes pertaining to adaptive living (Silva,

Moderator analyses
Initially, meta-regression analysis was run inclusive for all outcomes. It did not reveal any significant effects of age, year of publication or duration of program and session or number of sessions or quality of trials on the significance of these interventions. Meta-regression plots have been presented as S4 File. Subgroup analyses was run when specific outcomes reported in � four studies. It did not reveal any significant differences among interventions delivered by different agents on outcomes of symptom severity and joint attention. While significant subgroup differences were observed in reporting of joint engagement with parent mediated interventions reporting highest effect sizes (Table 3).

Quality rating & strength of evidence
Significant publication bias was revealed in reporting of social skills and symptom severity outcomes (Eggers statistics, P < 0.1). However, adjusted effect sizes for Social skills SMD = 0.42 (0.30 to 0.55) and symptom severity 0.38 (0.22 to 0.54) remained statistically significant (S4 File).
Cochrane's tool for risk of bias assessment among the included trials revealed an overall low risk of bias among majority of the studies. Random sequence generation was at a high/ unclear risk of bias among 8 trials, allocation concealment (n = 13). Frequency of studies reporting a high risk across other domains of Cochrane risk of bias tool were: Blinding of outcome assessors (n = 14), other sources of bias (n = 9), attrition bias (n = 8), selective reporting (n = 4) and blinding of participants and personnel (n = 0). A total of 11 studies were rated as having as having a high risk of overall bias i.e. � 3 matrices of risk of bias tool were rated as having unclear or high risk of bias for these studies [22,23,26,27,32,41] (Fig 7 and S4 File). Fig  7 presents a clustered bar chart exhibiting frequencies of high, unclear and low risk bias across all matrices of Cochrane risk of bias tool. S4 File presents study wise risk of bias across all matrices of Cochrane risk of bias tool.
According to the GRADE criteria, evidence for four outcomes was rated as: High for communication skills, expressive language, motor skills, repetitive behaviors, and parental distress. The evidence was moderate for adaptive behaviors, severity of symptoms, receptive language, social skills, and improvement in parent child relationship. It was found to be low for joint engagement, self-efficacy and competence, and very low for joint attention (Table 4).

Summary of results
We identified 33 studies comparing the effects of non-specialist mediated interventions with control groups among children with autism spectrum disorder. The meta-analyses demonstrated the effectiveness of non-specialist mediated interventions across several outcomes pertaining to social skills, motor and communication among children with autism. These were also associated with an improvement in parent-child relationship and parenting stress. The risk of bias among the studies assessed was generally low, albeit the overall strength of evidence varied across outcomes. This means that further research may change the effect estimates for a few of the outcomes reported in this review.
All of the interventions reviewed had varying density of dosage, but meta-regression analyses did not generally reveal any significant effects on the effectiveness of interventions. An outlier intervention program was the LEAP intervention program, run for 2 years spanning 834 sessions of naturalistic and incidental teaching among children with autism [23]. This  intervention reported the highest effect sizes, hence, we opine that sustained and long term interventions might achieve long sustaining results. A total of three outcomes including joint engagement, parent child relationship and joint attention exhibited substantial heterogeneity (I 2 >70%). Rest of the outcomes presented no to moderate heterogeneity. We opine that this may be because of two main reasons. For the outcome of joint engagement, this substantial heterogeneity is due to differences in intervention content as well as different delivery agents as shown in subgroup analysis ( Table 3). The outcomes of parent child relationship and joint attention were only reported in parent mediated interventions. The heterogeneity in these outcomes may be accounted for by use of different rating scale or methods of measurement. The studies reporting these outcomes used varying methods for measurement of both the joint attention and parent child relationship.

Recommendation for task shifting
Autism spectrum disorder is a major global health concern accounting for a large disease burden, health loss, disability adjusted life years (DALYs) and high specialist treatment costs. Moreover, the scarce availability of psychiatrists and psychologists in low-and middle-income countries further aggravate this public health issue. The WHO report on global mental health infrastructure, estimated the number of psychiatrists at less than one and 7.7 mental health nurses for 100,000 people in countries inhabited by 45% of the world's population [2,43]. In 2009, there was only one registered occupational therapist in Pakistan, highlighting the bleak situation in poorly resourced countries [44].
The proven effectiveness of non-specialized autism care in present review is thus, of particular relevance to low resourced settings, where access to specialist mental health interventionists specializing in autism treatment is poor [45]. However, only one good quality randomized controlled trial conducted jointly in India & Pakistan limits the evidence for clinical and costeffectiveness of these interventions in the region [6]. One of these studies was an adapted version of the PACT trial developed in Manchester [35] and was tested in a multi-site study conducted in Rawalpindi, Pakistan and Goa, India [6]. While, the second study revised this intervention and added a "plus" module pertaining to psychoeducation and assessment of the most disruptive comorbidity for the family [31]. Therefore, more research is required to ascertain the suitability of these interventions in the context of low-and middle-income countries. Moreover, there are no frameworks for recruitment, role descriptions and financial compensation for non-specialists, which creates a barrier in scale up and sustainability of these interventions [5,6]. Based on the findings of this systematic review, we cannot recommend one nonspecialist mediated therapy for autism. PACT, PASS and PASS plus; JASPER, SENSE and Hanen's more than words were tested in at least two studies and settings. Therefore, we recommend that future investigators, implementors and policy makers consult these therapy programs for development of interventions suitable for their settings.

Strengths and limitations
This study has several strengths. Firstly, the inclusion of randomized controlled studies ensured the internal validity of results. Previously, review studies had reported evidence for single cases, non-randomized controlled trials, specialist interventions or homogenous interventions based on specific strategies only [7,8,[46][47][48][49][50][51]. Lastly, our study was inclusive of children of all ages, confirmed diagnoses of autism spectrum disorder, psychosocial functioning, languages and time period, thereby, improving the generalizability of the results to this study population. Lastly, the subgroup analysis based on intervention mediators led to meaningful subgroup analyses.
Despite of its strengths, there are several limitations of this review and therefore, these results should be interpreted with caution. For instance, none of the studies had reported standardized outcomes pertaining to the IQ and psychological functioning of the study sample. Meta-regression analysis accounting for IQ of the children is a necessary analysis for studying moderating effects on the intervention effects. The interventions differed in their content and strategies, study settings, and intervention mediators, leading to substantial methodological heterogeneity in the meta-analyses. Several diagnostic methods such as ADOS and different updates of DSM criteria for diagnoses of autism were employed in included RCTs, further adding heterogeneity in the results. Psychological interventions limit the blinding status of participants and personnel as well as outcome assessors that is a serious limitation. The present systematic review was based on searching of a limited number of databases, we encourage investigators to search more databases in future studies. Moreover, investigators should also consider using a more comprehensive search strategy encompassing different terms for RCT. Combining results from diverse measures applied for heterogeneous study samples is another limitation of this systematic review.

Implications for practice and future research
The present review provides an overall good quality evidence of effectiveness of non-specialist mediated interventions among the children with autism. Most of the studies were mediated by parents and caregivers and presented low risk of bias. However, the evidence for peer and teacher mediated interventions was poor due to a limited number of studies. The sample size was low among individual studies and only a few interventions were tested in long term follow-up studies. The economic feasibility and cost-effectiveness were not reported in most of the interventions; an important metric for evaluating their suitability for task shifting and scaling up. The standardized instruments differed in studies, adding to the methodological heterogeneity among studies. Future studies should be designed keeping these limitations in context, emphasizing the introduction of standardized and cross-culturally validated instruments for assessment of symptomatology.

Recommendations
Despite of the aforementioned limitations, a small to moderate improvements in several debilitating symptoms of autism were noticed. These interventions also reduced care-giver stress and improved parent-child relationship. Based on the clinical effectiveness and good quality of evidence for these interventions, we recommend up-scaling of these interventions in high income countries. However, more research is required to ascertain the suitability of these interventions in context of low-and middle-income countries. Based on the findings of this systematic review, we cannot recommend one non-specialist mediated therapy for autism. PACT, PASS and PASS plus; JASPER, SENSE and Hanen's more than words were tested in at least two studies and settings.