Scolaris Content Display Scolaris Content Display

Cochrane Database of Systematic Reviews Review - Intervention

Parent‐mediated early intervention for young children with autism spectrum disorders (ASD)

Authors' declarations of interest

Version published: 30 April 2013 Version history

https://doi.org/10.1002/14651858.CD009774.pub2
Collapse all Expand all

Abstract

available in

Background

Young children with autism spectrum disorders (ASD) have impairments in the areas of communication and social interaction and often display repetitive or non‐compliant behaviour. This early pattern of difficulties is a challenge for parents. Therefore, approaches that help parents develop strategies for interaction and management of behaviour are an obvious route for early intervention in ASD. This review updates a Cochrane review first published in 2002 but is based on a new protocol.

Objectives

To assess the effectiveness of parent‐mediated early interventions in terms of the benefits for both children with ASD and their parents and to explore some potential moderators of treatment effect.

Search methods

We searched a range of psychological, educational and biomedical databases including CENTRAL, MEDLINE, Embase, PsycINFO and ERIC in August 2012. As this is an update of a previous review, we limited the search to the period following the original searches in 2002. Bibliographies and reference lists of key articles were searched, field experts were contacted and key journals were handsearched.

Selection criteria

We included only randomised controlled trials of early intervention for children with ASD. The interventions in the experimental condition were mediated by parents; the control conditions included no treatment, treatment as usual, waiting list, alternative child‐centred intervention not mediated by parents, or alternative parent‐mediated intervention of hypothesised lesser effect than the experimental condition.

Data collection and analysis

Two review authors (HM and IPO) independently screened articles identified in the search and decided which articles should be retrieved in full. For each included study, two review authors (IPO and EH) extracted and recorded data, using a piloted data collection form. Two review authors (IPO and HM) assessed the risk of bias in each study. We performed data synthesis and analysis using The Cochrane Collaboration's Review Manager 5.1 software.

Main results

The review includes 17 studies from six countries (USA, UK, Australia, Canada, Thailand and China), which recruited 919 children with ASD. Not all 17 studies could be compared directly or combined in meta‐analyses due to differences in the theoretical basis underpinning interventions, the duration and intensity of interventions, and the outcome measurement tools used. Data from subsets of 10 studies that evaluated interventions to enhance parent interaction style and thereby facilitate children's communication were included in meta‐analyses. The largest meta‐analysis combined data from 316 participants in six studies and the smallest combined data from 55 participants in two studies. Findings from the remaining seven studies were reported narratively.

High risk of bias was evident in the studies in relation to allocation concealment and incomplete outcome data; blinding of participants was not possible. Overall, we did not find statistical evidence of gains from parent‐mediated approaches in most of the primary outcomes assessed (most aspects of language and communication ‐ whether directly assessed or reported; frequency of child initiations in observed parent‐child interaction; child adaptive behaviour; parents' stress), with findings largely inconclusive and inconsistent across studies. However, the evidence for positive change in patterns of parent‐child interaction was strong and statistically significant (shared attention: standardised mean difference (SMD) 0.41; 95% confidence interval (CI) 0.14 to 0.68, P value < 0.05; parent synchrony: SMD 0.90; 95% CI 0.56 to 1.23, P value < 0.05). Furthermore, there is some evidence suggestive of improvement in child language comprehension, reported by parents (vocabulary comprehension: mean difference (MD 36.26; 95% CI 1.31 to 71.20, P value < 0.05). In addition, there was evidence suggesting a reduction in the severity of children's autism characteristics (SMD ‐0.30, 95% CI ‐0.52 to ‐0.08, P value < 0.05). However, this evidence of change in children's skills and difficulties as a consequence of parent‐mediated intervention is uncertain, with small effect sizes and wide CIs, and the conclusions are likely to change with future publication of high‐quality RCTs.

Authors' conclusions

The review finds some evidence for the effectiveness of parent‐mediated interventions, most particularly in proximal indicators within parent‐child interaction, but also in more distal indicators of child language comprehension and reduction in autism severity. Evidence of whether such interventions may reduce parent stress is inconclusive. The review reinforces the need for attention to be given to early intervention service models that enable parents to contribute skilfully to the treatment of their child with autism. However, practitioners supporting parent‐mediated intervention require to monitor levels of parent stress. The ability to draw conclusions from studies would be improved by researchers adopting a common set of outcome measures as the quality of the current evidence is low.

PICOs

Population
Intervention
Comparison
Outcome

The PICO model is widely used and taught in evidence-based health care as a strategy for formulating questions and search strategies and for characterizing clinical studies or meta-analyses. PICO stands for four different potential components of a clinical question: Patient, Population or Problem; Intervention; Comparison; Outcome.

See more on using PICO in the Cochrane Handbook.

Plain language summary

available in

Early intervention delivered by parents for young children with autism spectrum disorders

Autism spectrum disorders (ASD) affect more than 1% of children and is usually evident in behaviour before the age of three years. A child with ASD lacks understanding of how to interact with another person, may not have developed language or understand other people's communication, and may insist on routines and repetitive behaviours. This early pattern of difficulties is a challenge for parents. Therefore, helping parents to develop strategies for interaction and management of behaviour is an obvious route for early intervention. The present review brings up to date one published in 2003, which found only two well‐designed studies. This review, based on a new protocol, includes 17 randomised controlled trials, most published since 2010, in which interventions delivered by parents were compared with no treatment or local services, or alternative child‐centred intervention such as nursery attendance, or another parent‐delivered intervention that differed in some way from the main condition. We were able to combine outcome information and so increase confidence in the results. All the studies were rated on the quality of their evidence, which was then taken into account in judging how firmly conclusions could be drawn.

The studies varied in the content of what parents were trained to do, and over what length of time parents had contact with professionals. Parents received training either individually with their child or in groups with other parents. In the majority of the studies, the interventions aimed to help parents be more observant and responsive during interactions with their child in order to help their child develop communication skills. 

In summary, the review finds sufficient evidence that the ways in which parents interacted with their children did change as intended. The review also suggests improvement in child outcomes such as understanding of language and severity of autism characteristics as a result of interventions delivered by parents. However, important outcomes such as other aspects of children's language, children's adaptive skills and parent stress did not show change. The evidence is not yet strong for any outcome and would benefit from researchers measuring effects in the same ways.  

Authors' conclusions

Implications for practice

The findings of this review have a number of implications for practice. First, given a conclusion that there is sufficient evidence of the effectiveness of parent‐mediated interventions in treatment of ASD in young children, there is then a need for local service providers to pay increased attention to models and approaches that facilitate parent participation as part of available therapies. Three studies tested a combination of parent‐mediated interventions with other locally available services, and this might be expected to be the preferred pattern for local autism treatment provision so that the burden on parents is shared (Jocelyn 1998; Rickards 2007; Dawson 2010). Aspects of intervention that aim to reduce parents' stress might also be given more attention in future. Second, some of the studies reviewed hinted at the fact that not all therapies may suit all families and individuals, highlighting the need for treatment of ASD to be individualised. The benefits of an increase in parent interaction skills (such as a 'synchronous' style of interacting) will be likely to generalise therapy strategies into the natural environments of children with ASD, and therapist coaching of parents can be tailored to individual needs with some degree of ease and flexibility.

Implications for research

Most of the studies had methodological flaws that, if avoided, would improve the quality of evidence available, and thus increase our confidence in the effect estimates (Risk of bias in included studies; Figure 3; Figure 2). First, there is need for larger RCTs recruiting greater numbers of participants; only Green 2010 was adequately powered. Larger sample sizes will help increase power to detect the presence of a true effect if there is one, and to explore important moderators and mediators of effects. Second, most of the included studies failed to report on the process of allocation concealment, an important step in limiting bias in RCTs. Other weaknesses frequently observed were incomplete outcome data and variable blinding of assessors, and could be remediated in future studies. Third, the variability in outcome measures used, and the lack of evidence for their measurement properties, suggests there is an urgent need to explore what might be an optimum battery of outcome measures for early intervention (Cunningham 2012), with pre‐specification of the primary outcome and full reporting. There is a particular problem with assessment of improvement in the core social‐communication skills impaired in autism. Progress towards consistent and robust measurement requires further development and cross‐site agreement on measures of coding parent‐child interaction, and direct assessment of child social and communication skills, so as to ensure sensitive and objective outcome measures are used in future studies. Finally, researchers should include an estimate of the costs of their interventions, as well as any adverse effects, as these have important implications for future translation of research findings into practice. To arrive at more robust conclusions, there is need for studies of better quality, including attention paid to blinding, allocation concealment and fidelity of implementation, and more studies with outcome measure data that could readily be combined.

Summary of findings

Open in table viewer
Summary of findings for the main comparison.

Child and parent outcomes following parent‐mediated interventions for young children with autism spectrum disorders (ASD)

Patient or population: children aged 1 year to 6 years 11 months with diagnosis of ASD

Settings: home, clinic, or both, interventions delivered through group or 1‐to‐1 sessions

Intervention: parent‐mediated early intervention

Comparison: no treatment or treatment as usual (including alternative child‐centred intervention not mediated by parents, alternative parent‐mediated intervention that differed in some way from the experimental condition and waiting list control groups, for example, intensity)

Outcomes

Illustrative comparative risks* (95% CI)

No of participants
(studies)

Quality of the evidence
(GRADE)

Comments

Assumed risk

Corresponding risk

No treatment or treatment as usual

Parent‐mediated early intervention

Language ‐ joint (direct or independent assessment)

EIDP/PSDP, PLS‐IV/CASL1

‐

The mean language ‐ joint (independent assessment) in the intervention groups was
0.45 standard deviations higher
(0.05 lower to 0.95 higher)

64
(2 studies)

⊕⊕⊝⊝

low2

SMD 0.45 (‐0.05 to 0.95). Higher scores indicate improvement. Duration of intervention ranged from 12 weeks to 1 year. This is small effect 3

Communication (reported)

VABS1

The mean communication (reported) ranged across control groups from
60.93 to 67.7

The mean communication (reported) in the intervention groups was
5.31 higher
(6.77 lower to 17.39 higher)

228
(3 studies)

⊕⊕⊝⊝

low2

The mean value was 5.31 points higher for the intervention group. Higher scores indicate improvement. Duration of interventions ranged from 1 to 2 years

Language – expression (direct or independent assessment)

PLS‐IV, MSEL1

‐

The mean language – expression (direct assessment) in the intervention groups was
0.14 standard deviations higher
(0.16 lower to 0.45 higher)

264
(3 studies)

⊕⊕⊝⊝

low2

SMD 0.14 (‐0.16 to 0.45). Higher scores indicate improvement. Duration of interventions ranged from 12 weeks to 2 years. This is a small and uncertain effect 3

Language – Comprehension (direct or independent assessment)

PLS, MSEL1

‐

The mean language – comprehension (direct assessment) in the intervention groups was
0.29 standard deviations higher
(0.2 lower to 0.78 higher)

200
(2 studies)

⊕⊕⊝⊝

low2

SMD 0.29 (‐0.2 to 0.78). Higher scores indicate improvement. Duration of intervention ranged from 1 to 2 years in the studies. This effect size is small and uncertain 3

Parent‐child interaction (parent synchrony)

‐

The mean parent‐child interaction (parent synchrony) in the intervention groups was
0.9 standard deviations higher
(0.56 to 1.23 higher)

244
(3 studies)

⊕⊕⊝⊝

low2

SMD 0.9 (0.56 to 1.23). Higher scores indicate improvement. Duration of intervention ranged from 12 weeks to 1 year. This is large effect size 3

Parents'level of stress

PSI, SAC1

‐

The mean parents' level of stress in the intervention groups was
0.17 standard deviations lower
(0.7 lower to 0.36 higher)

55
(2 studies)

⊕⊕⊝⊝

low2

SMD ‐0.17 (‐0.7 to 0.36). Lower scores indicate improvement. Duration of interventions ranged from 12 weeks to 1 year. This is a small and uncertain effect 3

CI: confidence interval; SMD: standardised mean difference

GRADE Working Group grades of evidence
High quality: Further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: We are very uncertain about the estimate.

1 CASL: Comprehensive Assessment of Spoken Language, quotients; EIDP: Early Intervention Developmental Profile; MSEL: Mullen Scales of Early Learning; PLS‐IV: Preschool Language Scale‐IV; PSDP: Preschool Developmental Profile; PSI: Parental Stress Inventory; SAC: Stress Arousal Checklist; VABS: Vineland Adaptive Behavior Scales.
2 Quality of evidence has been rated as low across domains due to the small numbers of studies combined.
3 Cohen's interpretation of effect size (rule of thumb): 0.2 is considered a small effect, 0.5 is considered a moderate effect and 0.8 or higher is considered as large effect.

Background

Description of the condition

Autism is the core disorder of the pervasive developmental disorders as defined in the International Classification of Diseases and Related Health Problems, 10th edition (ICD‐10) (WHO 2010) and the Diagnostic and Statistical Manual of Mental Disorders IV, Text Revision (DSM‐IV‐TR) (APA 2000). It is evident before the age of three years and is characterised by impairments in communication and social interaction and the presence of restricted and repetitive behaviours and impaired imagination. The prevalence of core autism is accepted to be around four per 1000 (Baird 2006). Autism is, however, understood to be on a continuum rather than being a distinct category, with clinical agreement on a range of difficulties that can be classified as 'autism spectrum disorders' (ASD). The prevalence for all ASD is around 11 per 1000, with a male:female ratio of 3.3:1 (Baird 2006). Young children with difficulties on the autism spectrum pose challenges to family members and others who interact with them: they lack understanding of how to initiate and respond to joint attention with another person, have difficulties in social timing of communication and may not understand other people's intentions as expressed through language and gestures, even though they may appear affectionate and want to be with other people socially. They have difficulty organising their responses and inhibiting repetitive behaviours and interests. Children with ASD frequently pose considerable behaviour challenges to their parents and other family members.

Description of the intervention

Previous literature reviews have suggested the effectiveness of a number of early intervention programmes (Dawson 1997; Rogers 1998a; Smith 1999), as have more recent reviews of the evidence (Ospina 2008; Rogers 2008), but the quality of the evidence base is still weak. The field of ASD remains controversial, with a range of questionable claims for efficacy of therapies and few successful replication studies (Diggle 2002).

ASD early intervention programmes vary considerably in their theoretical background (Prizant 1998). Some approaches use applied behaviour analysis (ABA) or early intensive behavioural interventions (EIBI) in intensive programmes that are based at home but delivered primarily by trained therapists (for example, Howlin 1987; McEachin 1993; Sallows 2005; Reichow 2012). Others have an educational framework, such as project TEACCH (Treatment and Education of Autistic and related Communication handicapped CHildren), with an emphasis on structuring class environments through visual cueing, communication routines and individual tasks (for example, Lord 1994). The TEACCH project aims to increase children's independence and is designed to work on their existing strengths rather than focusing on weaknesses. Other programmes focus more on communication partners such as parents and emphasise the creation of naturalistic communication opportunities, enhancing reciprocity between communication partners, enhancing children's motivation for social interaction and prompting specific social behaviours (for example, Rogers 1991; Koegel 1995; Aldred 2004).

Research in this area has typically involved children aged between one and seven years (for example, Koegel 1996; Jocelyn 1998; Smith 2010); and although these programmes may differ in content, they all advocate treatment implementation as early as possible as a matter of clinical urgency. The age at which intervention starts has been reducing since the early 2000s with the advent of earlier identification and diagnosis of ASD (Charman 2010). However, a diagnosis of ASD, based on behavioural observation, is not given before the age of 12 months, as children's impairments in social communication cannot be interpreted until the second year of life at the earliest. Stability of diagnosis before 30 months can be poor (Turner 2007).

Some reviews have suggested that successful early intervention programmes have important similarities, whatever their apparently different theoretical foundations (for example, Dawson 1997; Prizant 1998). Indeed, Rogers conducted a comparative analysis of the elements of apparently contrasting programmes to demonstrate how each may address the underlying neuropsychological processing difficulties evidenced by children with ASD, including intersubjectivity (interpersonal sharing including establishing joint attention to objects), and emotional functioning and imitation (Rogers 1998). It has been suggested that successful programmes are those in which the 'dose' is intensive (a significant number of hours per week and rigorous levels of structure and instruction are introduced into the child's world (Eldevik 2009)). Programmes may thus have to effect a change through the family system in order to treat ASD successfully, as well as offer expert educational support.

The involvement of parents in implementing intervention strategies designed to help their autistic children has a history stretching back to the 1970s (for example, Schopler 1971). Within the ASD treatment literature there have been a number of studies that evaluate specific parent training approaches in dealing with behaviour problems (for example, Howlin 1987; Vriend 2011), in improving parent‐child interactions (for example, Koegel 1996; Dawson 1997; McConachie 2005), in facilitating communication (for example, Prizant 1997; Aldred 2004), and in implementing an ABA approach (for example, Smith 2000). In addition, there are evaluations of the added value of parent involvement to a daycare or nursery programme (for example, Jocelyn 1998; Rickards 2007). A review by Boyd 2011 concludes that "many of the promising focused intervention practices and comprehensive treatment models (CTMs) involve components of naturalistic interventions for teaching pivotal skills in natural environments and parent‐implemented approaches where carers learn strategies to better support their children's development". Harnessing the skills of parents allows the opportunity for generalisation of child learning between environments, and consistency of management strategies.

How the intervention might work

Training parents as 'therapists' allows intervention to begin early, with the aim that parent interaction strategies help enhance children's earliest social relationships. It is important, given the nature of the children's impairments, that parents support the child in establishing shared interest in each other and in objects, and learn the power of imitation. If parents act in a way that is 'synchronous' with their child's focus and intentions, then language and communication are enhanced (Siller 2008). The secondary effects may be reduced frustration for the child, as well as increased parent confidence and skills. Increased parental skills allow for continual opportunities for child learning in a range of situations. Furthermore, training parents in new skills has frequently been carried out in groups, allowing for mutual support and potential reduction of parental stress. Where parent training is an adjunct to an education‐based intervention, the amount of ASD‐appropriate intervention is increased, with generalisation of child learning across different people and environments.

Why it is important to do this review

Since the mid 2000s, there has been an increase in research publications that seek to address issues surrounding early identification and early interventions for children with ASD (Boyd 2011). The earlier version of this Cochrane review, Diggle 2002, found only two randomised controlled trials (RCT) of parent‐mediated early intervention. From 2009 onwards, there has been a surge in the publication of RCTs of early intervention where parents are key to the delivery. In addition to improvement in research design, improvements in screening have enabled accurate detection of ASD at an age earlier than was documented about 30 years ago (that is, some children are now diagnosed as early as two years of age). However, doubt remains about which interventions are effective for young children with ASD and what constitutes an objective assessment of improvement. Furthermore, the increased number of studies raises a potential opportunity to link ASD and other characteristics to the outcomes of intervention. However, these developments have not yet led to consensus answers to important questions for healthcare policymakers and parents alike: What intervention should money be spent on? Which children, and which parents, will benefit most from what intervention? Is there evidence of greater benefit with earlier detection of ASD and thus earlier intervention? How long should interventions last?

Objectives

The primary aim of this review is to assess the effectiveness of parent‐mediated early interventions in terms of the benefits for both children and their parents. A secondary aim is to attempt to explore some potential moderators of treatment effect.

Methods

Criteria for considering studies for this review

Types of studies

We included only RCTs in this review as studies where participants are not randomly allocated to groups are more liable to bias.

Types of participants

  • Parents of children with ASD, aged between one year and six years eleven months. Studies that include child participants whose ages fall outside of this range of one year to six years eleven months may be included in the review, for example, if fewer than 5% of the children are above six years eleven months at the start of the study. The term 'parent' was used to include carers who take on a parental role.

  • Diagnosis of ASD to include autism, Asperger's syndrome, Pervasive Developmental Disorder (PDD) and PDD Not Otherwise Specified (PDD NOS).

  • Where a study included child participants with a variety of developmental disorders, it was included only if results were presented separately for the ASD group.

Types of interventions

Interventions in which parents were trained by professionals in strategies designed to improve the management of their child's ASD‐related difficulties. Parents must have received ongoing supervision and support from professionals. The training may have involved group or individual coaching of parents in a planned (potentially replicable) approach designed to help them promote their child's communicative and social development, learning, skills or control of behaviour.

The control conditions were no treatment, treatment as usual, a waiting list group, an alternative child‐centred intervention not mediated by parents or an alternative parent‐mediated intervention that differed in some way from the experimental condition. Where there are two alternative parent‐mediated interventions, one is likely to be a 'contact control' and would be regarded as the 'control' group; where one is of higher intensity, it would be regarded as the 'experimental' intervention.

Types of outcome measures

A range of measures is currently in use to confirm the diagnosis of ASD and characterise children with ASD. No particular measure or reporting of an outcome was used as an inclusion criterion for this review.

Primary outcomes
Child communication and social development

  • Language development (comprehension and expression)*

  • Social communication skills*

  • Skills in interaction with parent*

Parental outcome

  • Parents' level of stress*

Secondary outcomes
Child ability

  • Developmental/intellectual gains

  • Adaptive behaviour

Child problem behaviour

  • Severity of autism characteristics*

  • Maladaptive behaviour*

Parental outcomes

  • Parents' satisfaction with therapy

  • Parents' confidence in coping with child's disability and behaviour problems

Cost of intervention

  • Any cost information provided by the authors

The outcomes marked with (*) are used to populate the summary of findings Table for the main comparison.

Timing of outcome assessment

We collected pre‐treatment (baseline) and outcome data. Outcome data were collected immediately post‐treatment or at the time points closest to the following periods as appropriate: six months, one year, two years and four years.

Search methods for identification of studies

We considered published or unpublished RCTs, with no language restrictions.

Electronic searches

We searched the following databases between the year 2002 to August 2012 for relevant publications.

  • Cochrane Central Register of Controlled Trials (CENTRAL), part of The Cochrane Library ‐ from 2002 to August 2012.

  • Ovid MEDLINE(R) ‐ from 2002 to August 2012.

  • Embase ‐ from 2002 to August 2012.

  • ERIC (Educational Resources Information Centre) ‐ from 2002 to August 2012.

  • PsycINFO ‐ from 2002 to August 2012.

  • CINAHL (Cumulative Index to Nursing and Allied Health Literature) ‐ from 2002 to August 2012.

  • Dissertation Abstracts International ‐ from 2002 to August 2012.

  • Social Sciences Abstracts ‐ from 2002 to August 2012.

  • Sociological Abstracts ‐ from 2002 to August 2012.

  • Linguistics and Language Behavior Abstracts ‐ from 2002 to August 2012.

  • Cochrane Database of Systematic reviews (CDSR), part of The Cochrane Library ‐ from 2002 to August 2012.

  • Database of Abstracts of Reviews of Effects (DARE) ‐ from 2002 to August 2012.

  • National Research Register (NRR) Archive ‐ from 2002 to August 2012.

  • LILACS (Latin American Health Sciences Literature) ‐ from 2002 to August 2012.

  • Turning Research into Practice (TRIP) database ‐ from 2002 to August 2012.

  • OpenGrey ‐ from 2002 to August 2012.

  • ASSIA (Applied Social Sciences Index and Abstracts) ‐ from 2002 to August 2012.

  • IBSS (International Bibliography of the Social Sciences) ‐ from 2002 to August 2012.

  • National Criminal Justice Reference Service Abstracts ‐ from 2002 to August 2012.

  • WHO International Clinical trials Registry Platform (ICTRP) ‐ from 2002 to August 2012.

  • metaRegister of Controlled Trials (mRCT) ‐ from 2002 to August 2012.

  • CinicalTrials.gov ‐ from 2002 to August 2012.

Search terms were adapted for each database using appropriate syntax and controlled vocabulary. As this is an update of a previous review (Diggle 2002), we limited the search to the period from 2002 to August 2012. The search strategies used for this update are reported in Appendix 1. To identify studies that were not yet indexed in the above databases, Google Scholar was searched. In order to keep up to date with new articles that may be added to databases after our search, alerts were set up in databases that had provision for this facility (for example, Ovid MEDLINE, Google Scholar, etc.)

Searching other resources

We examined other sources of information including the bibliographies of systematic and non‐systematic reviews and reference lists of key articles identified through the search strategy. We contacted experts in the field by letter in order to identify unpublished studies. We also handsearched key journals to identify studies not yet electronically catalogued in the databases searched, in addition to using Google Scholar to search the World Wide Web.

Data collection and analysis

Selection of studies

All citations generated through the search strategy were transferred to the reference management program EndNote Web X5. Two review authors (HM, expert in early autism, and IPO, expert in systematic reviews) independently screened titles and abstracts identified in the search, and indicated which reports should be retrieved in full. The full report of any title or abstract for which there were insufficient data was retrieved. The same review authors independently read full reports and determined whether these studies met the specified inclusion criteria. Multiple reports of the same study were sought. Two review authors (HM and IPO) independently assessed studies for inclusion from the pool of remaining studies.

Data extraction and management

For each included study, two review authors (IPO, EH) extracted and recorded the following data, using a piloted data collection form: study location, funding source, study design, methods, participant details (diagnostic description and severity of impairments, parent characteristics), type of intervention (including the intensity and duration of intervention), measurement of adherence, outcome measures, any reported cost data, and key conclusions of study authors.

Assessment of risk of bias in included studies

We used The Cochrane Collaboration tool for assessing risk of bias (Higgins 2011). Two review authors (HM and IPO) independently assessed the risk of bias for each included study based on the following six domains; judgements were rated as 'high', 'low' and 'unclear' risk of bias. Final judgements of risk of bias were then reached by consensus. This is summarised and presented in Risk of bias in included studies, which can be found in the results section.

Random sequence generation: we described the method used to generate the allocation sequence in sufficient detail to allow an assessment of whether it should produce comparable groups.
Allocation concealment: we described the method used to conceal the allocation sequence in sufficient detail to determine whether intervention allocations could have been foreseen in advance of, or during, enrolment.
Blinding of participants and personnel: we described all measures used, if any, to blind personnel from knowledge of which intervention a participant received. As the focus of the review is on parent‐mediated early intervention, the parent participants could not be blind, so the judgement depended on whether the outcome was likely to have been influenced by lack of blinding.
Blinding of outcome assessment: we described all measures used, if any, to blind outcome assessors from knowledge of which intervention a participant received. We also provided any information relating to whether the intended blinding was effective.
Incomplete outcome data: we described the completeness of outcome data for each main outcome, including attrition and exclusions from the analysis. We also stated whether attrition and exclusions are reported, the numbers in each intervention group (compared with total randomised participants), reasons for attrition/exclusions where reported, and any re‐inclusions in analyses performed by the review authors.
Selective outcome reporting: we assessed the possibility of selective outcome reporting by the study authors by checking whether any of the stated outcomes were not reported at the end of the study.

In addition, for important outcomes included in summary of findings Table for the main comparison, we assessed the overall quality of evidence using the 'GRADE' approach (GRADE working group). Factors taken into consideration before judgements were made (the GRADEpro criteria) include limitations of detailed designs and execution (risk of bias) in studies, inconsistency (heterogeneity), indirectness (population, intervention, comparison and outcome), imprecision and publication bias (see also below).

Measures of treatment effect

Any procedures for making decisions set out in the protocol but not implemented in the review are summarised in Table 1 below.

Open in table viewer
Table 1. Table showing changes to methods described in the protocol

Section of the review

Issues

What was done

Measurement of treatment effect

No relevant categorical data encountered

No analysis using categorical data was done

Unit of analysis

No cluster randomised trials were identified

No analysis was done

Cross‐over trials

Wong 2010 presented a cross‐over design. Information from this study was used only in the narrative synthesis as the measures used in the study, the way in which results were reported and the theoretical background prohibited its inclusion in the meta‐analysis. Thus the study did not present any "unit of analysis" issue (see Appendix 2 and Characteristics of included studies sections). Only data in the first part of the study, before participants swapped groups, was used in the discussions. 

Studies with multiple treatment groups

Two studies had multiple treatment groups and a control group. Roberts 2011 had a non‐randomised control group, so the two treatment conditions were compared (home based treated as the experimental intervention group, and centre‐based treated as the control group). Data from this study were included only in the narrative synthesis as the measures used in the study and the study methods prohibited its inclusion in the meta‐analysis (see Appendix 2 and Characteristics of included studies sections). Tonge 2006/Tonge 2012 also had 2 treatment arms, with hypothesised greater effects for the treatment with dedicated skills training versus a parent education‐only group. Therefore for this review, the 2 treatment arms were compared with the latter arm as the control.

In future revisions of this review, it is likely that there will be more examples of comparison of multiple‐treatment groups. In that case the authors will first examine studies that compare with no treatment, waiting‐list or treatment as usual, and then examine studies with other treatment controls, so as to interpret the latter in light of the evidence for the former group of studies.

Categorical data

Binary outcomes were not encountered in the present review. If encountered in future reviews, they will be analysed by calculation of the odds ratio (OR) with a 95% confidence interval (CI).

Continuous data

Continuous data encountered were analysed on the assumption that the means and the accompanying standard deviations (SD) were from a normally distributed sample with no evidence of skew. Where studies used different measurement scales for assessing outcomes of a similar construct, we used standardised mean differences (SMD) with the 95% CI as a summary statistic. In cases where studies used a uniform measurement scale for assessing outcomes, the mean difference (MD) was preferred. While a number of the studies examined reported change from baseline, all studies reported means (and SD) at start and end (or provided this information upon request).Thus, end point means and accompanying SD were extracted. We calculated SMDs using Hedges g.

Multiple outcomes

In studies that provided multiple, interchangeable measures of the same construct at the same point in time (for example, multiple measures of child language), we intended to combine results across those measures that are most similar in their properties, that is, where authors used the same or similar wording to describe the outcome construct, and the method of administration was the same (for example, for parent questionnaires about child maladaptive behaviour). However, in this review, it was frequently judged that methods of administration were different, and constructs were dissimilar, and so separate analyses were conducted (for example, direct assessment of children's language skills analysed separately from parent‐report; child comprehension of language separately from child expressive language). Where similar outcomes were combined, we used the reported average or combined means provided in eligible articles.

Unit of analysis issues

Cluster‐randomised trials

We did not find any cluster‐randomised trials.

Studies with multiple treatment groups

In cases where studies compare more than one intervention condition, we intended to combine results across eligible (parent‐mediated early intervention) intervention groups and compare these with the control group, making single, pair‐wise comparisons. However, none of the studies included in the meta‐analyses had multiple treatment groups.

Cross‐over studies

Similarly, the only study that had a cross‐over design (Wong 2010) was not included in the meta‐analysis as it differed significantly in its choice of measures and methodology.

Dealing with missing data

We contacted the authors of included studies to supply any unreported data (for example, group means and SDs, details of dropouts, and details of interventions received by the control group). We also described missing data and dropouts/attrition for each included study in the 'Risk of bias' table, and where necessary, discussed the extent to which these could impact upon the results/conclusions of the review and provided reasons (as stated by authors) for missing data in the review.

Assessment of heterogeneity

We intended to explore the possibility of assessing clinical variation across studies by comparing the distribution of duration and intensity of intervention. However, the information given in studies was not presented in such a way that hours of intervention could be estimated in order to make systematic comparisons. Heterogeneity according to focus of intervention, and whether the control condition was less (or more) intensive than the experimental intervention, was considered in choosing the studies included in meta‐analysis (see Appendix 2 study characteristics). Type of intervention is addressed in the narrative results.

Statistical heterogeneity was assessed by computing the I2 statistic (Higgins 2011), a quantity that describes approximately the proportion of variation in point estimates that is due to heterogeneity rather than sampling error. In cases where we combined only a few studies, we preferred direct comparison of any difference between papers over the I2 values owing to the small numbers. Heterogeneity found in the analyses could be explained by the way outcomes were assessed in the articles (direct assessment versus reports), the coding schemes used by researchers, differences in follow‐up intervals (which ranged from two to 13 months) and the varied nature of the interventions and outcomes reported. Thus, further statistical analyses were not carried out due to these differences.

Assessment of reporting biases

We used funnel plots (estimated differences in treatment effects against their standard error) to subjectively assess reporting bias in the studies.

The extent of bias in the interpretation of results, for example, where the interpretation is based on results from treatment completers alone rather than the intention‐to‐treat analysis, or the significance of an inadequately powered studies is over‐interpreted, or there is over‐reliance on non‐blinded measures, is addressed in the narrative results.

Data synthesis

We synthesised results in a meta‐analysis (using the inverse variance method) across interventions that were similar in their theoretical basis, the way in which parent‐training was delivered, duration and intensity, and that used outcome measures with similar psychometric and psychological properties. We have reported results as means with SDs. We used both a fixed‐effect and a random‐effects model and compared the results to assess the impact of statistical heterogeneity. Unless the model was contraindicated (for example, if there was funnel plot asymmetry or a large difference between the results obtained from fixed‐effect and random‐effects meta‐analyses), we have presented only the results obtained from the random‐effects model. Data synthesis was conducted using the Review Manager 5.1 software (RevMan 2011).

Subgroup analysis and investigation of heterogeneity

We conducted subgroup analyses to investigate the following possible sources of heterogeneity: how outcomes were assessed (direct reporting by outcome assessors such as trained psychologists, or reported by parents and carers of children). Due to variability in reporting methods, moderators such as duration, intensity, and type of intervention and parental education were not investigated directly in our analyses. Child intelligence quotient (IQ) could not be investigated as it was usually reported by group means rather than in ability bands. However, information relating to all these domains has been summarised and included in the narrative synthesis (Included studies).

Sensitivity analysis

We conducted sensitivity analyses to determine whether findings were sensitive to restricting the analyses to studies judged to be at low risk of bias based upon:

  • only studies judged to be of low risk of selection bias (associated with sequence generation or allocation concealment);

  • only studies judged to be of low risk of performance and detection bias (associated with issues of blinding);

  • only studies judged to be of low risk of attrition bias (associated with completeness of data).

In the analyses, we used the phrase 'No difference' to indicate instances where the overall conclusion or direction of effect remained unchanged. This does not take into account changes in the effect size estimate. The phrase 'There was a difference' was used to describe instances where there was a change in the overall conclusion that could be drawn from the initial estimates following sensitivity analysis (for example, a change from a statistically significant result to an inconclusive or non‐significant one). In such cases, effects estimates are provided in the footnotes. See Table 2 below.

Open in table viewer
Table 2. Sensitivity analyses

Analysis

Sequence generation

Allocation concealment

Blinding of outcome

(detection bias)

Attrition

Joint language (direct assessment) Analysis 1.1

Not applicable

Not done (only 1 study had low risk)

Not applicable

Not applicable

Communication (reported) Analysis 1.2

Not applicable

Not done (only 1 study had low risk)

Not applicable

Not applicable

Language – expression (direct assessment) Analysis 1.3

Not applicable

Not done (only 1 study had low risk)

Not applicable

Not applicable

Language – expression (reported) Analysis 1.4

Not applicable

Not done (only 1 study had low risk)

No difference

Not applicable

Language – comprehension (direct assessment) Analysis 1.5

Not done as there were only 2 studies

Language – comprehension (reported) Analysis 1.6

Not applicable

Not done (only 1 study had low risk)

There was a difference $

Not applicable

Parent‐child interaction (shared or joint attention time) Analysis 1.7

Not applicable

Not done (only 1 study had low risk)

Not applicable

Not applicable

Child initiations (coding of parent‐child interactions) Analysis 1.8

Not applicable

Not done (only 1 study had low risk)

Not applicable

No difference

Autism severity Analysis 1.9

Not applicable

There was a difference *

No difference

No difference

Parents' level of stress Analysis 2.1

Not done as there were only two studies

Parent‐child interaction (parent synchrony) Analysis 2.2

Not applicable

Not done (only 1 study had low risk)

Not applicable

Not applicable

$ New effect estimate mean difference 27.63 (95% CI ‐10.94 to 66.20) P value > 0.05.

* New effect estimate standardised mean difference ‐0.20 (95% CI ‐0.52 to 0.12).

Results

Description of studies

Results of the search

This review is an update of a previous review (Diggle 2002); as such the search was limited to papers published since 2002. (For details of the original 2002 search, see Appendix 3.) From a search of multiple databases, over 10,000 articles were initially identified, although this included a significant number of duplicate articles. All databases yielded relevant citations; however, most of the relevant articles originated from PsycINFO, ERIC and Ovid MEDLINE/PubMed. Twenty‐seven additional articles were identified through handsearching of relevant websites for grey literature, contacting authors for unpublished articles and searching of reference lists of publications. Two articles were carried forward from the previous review.

After removal of duplicates, the remaining 110 articles were screened by two review authors (HM and IPO) and 69 records were rejected based on their titles and abstracts. The remaining 41 full‐text articles were further assessed for eligibility and inclusion using a set of predefined inclusion and exclusion criteria. At this point 23 articles were excluded; the Cochrane Group were consulted about three articles and decision taken to exclude them (two had partial non‐randomisation; one was not analysed as randomised). The excluded articles and reasons for their exclusion are listed in the Characteristics of excluded studies table. The remaining 18 articles were included in this review. Of these 18 articles, two were separate accounts of the same research evaluation study, with the earlier focusing on the parent outcomes and the later article focusing on the child outcomes. These two papers (Tonge 2006 and Tonge 2012) were thus combined. As a result we had 17 studies for the review. Figure 1 is a flow diagram showing a summary of the search and synthesis described above. Further details about these articles can be found in the Included studies section and Characteristics of included studies table.

Figure 1
Study flow diagram.

Study flow diagram.

Included studies

Participants

There were a total of 919 children across the included studies. Fourteen studies (Jocelyn 1998; Drew 2002; Aldred 2004; Rickards 2007; Silva 2009; Dawson 2010; Green 2010; Kasari 2010; Smith 2010; Wong 2010; Carter 2011; Pajareya 2011; Roberts 2011; Siller 2012) reported on the numbers of males and females in their studies, with 645 males and 171 females (a male:female ratio of approximately 3.8:1). However, three studies (Tonge 2006/Tonge 2012; Nefdt 2010; Casenhiser 2011) did not report on the number of males and females. The smallest study in this review (Wong 2010) had 17 participants, while the largest (Green 2010) had 152 participants.

Children were aged between 17 months and six years with varied levels of functioning. In eight studies (Aldred 2004; Silva 2009; Nefdt 2010; Wong 2010; Carter 2011; Casenhiser 2011; Pajareya 2011; Roberts 2011), IQ was not directly assessed. However, in eight studies (Jocelyn 1998; Smith 2000; Drew 2002; Rickards 2007; Dawson 2010; Green 2010; Kasari 2010; Siller 2012), IQ was assessed in varying ways at baseline, and in three studies it was used as cut off for recruitment (for example, with participants having an IQ less than 35 being excluded from the studies) (Smith 2000; Dawson 2010; Green 2010). In one study (Tonge 2006/Tonge 2012) the Psychoeducational Profile–Revised (PEP‐R) was used to assess functioning, and it was reported that scores obtained on this measure approximate IQ estimates on the Stanford‐Binet Intelligence Scales. All child participants in the studies had a diagnosis of autism or ASD made by an assessing clinician or psychologist based on DSM‐IV, DSM III‐R and or ICD‐10 classification; several of the studies used 'gold standard' instruments such as Autism Diagnostic Interview ‐ Revised (ADI‐R) or Autism Diagnostic Observation Schedule (ADOS), or both, to confirm diagnosis. None of the children had any comorbid or debilitating illness such as cerebral palsy, genetic syndromes, diagnosed hearing impairment, diagnosed visual impairment or seizures, or severe psychiatric disorders. Children were from a wide range of ethnicities including white, Hispanic, African, Caribbean, Asian, Latino and other mixed races.

Intervention content

The content and theoretical basis of the interventions varied considerably. Fourteen of the studies had a principal focus on facilitating parent‐child interaction and child communication skills through coaching of parents by therapists. In some studies, this principal focus was supplemented by educational and developmental techniques, such as those taken from Pivotal Response Treatment (Nefdt 2010) or Developmental, Individual‐Difference, Relationship‐Based (DIR) techniques (Casenhiser 2011; Pajareya 2011). One study focused on a massage intervention (Silva 2009), one on management of challenging behaviour (Tonge 2006/Tonge 2012), and one on early intensive behavioural intervention (Smith 2000).

Control condition

The control conditions reported in the studies ranged from no treatment or 'treatment as usual' (local services) (Drew 2002; Aldred 2004; Tonge 2006/Tonge 2012; Green 2010; Kasari 2010; Carter 2011; Pajareya 2011) to alternative interventions (either not mediated by parents such as nursery attendance, or another parent‐mediated intervention that differed in some way such as content or intensity) (Jocelyn 1998; Rickards 2007; Dawson 2010; Nefdt 2010; Silva 2009; Wong 2010; Casenhiser 2011; Siller 2012). In two studies the control condition was more intensive than the parent‐mediated intervention condition (Smith 2000; Roberts 2011). In most of the studies (Jocelyn 1998; Smith 2000; Drew 2002; Tonge 2006/Tonge 2012; Rickards 2007; Dawson 2010; Green 2010; Kasari 2010; Casenhiser 2011; Roberts 2011; Siller 2012), the intervention group received the same local services as the control group. Thus it becomes difficult to judge the effectiveness of the experimental autism interventions in isolation from the services received in addition. Furthermore, there were multiple services available, with members of control groups being free to use any services of their choosing. In six studies, details of services received by the control condition were not given (Aldred 2004; Silva 2009; Nefdt 2010; Wong 2010; Carter 2011; Pajareya 2011).

Setting

The delivery of training to parents took place in a range of different settings across the included studies. In seven of the studies, interventions were home‐based (Smith 2000; Drew 2002; Rickards 2007; Dawson 2010; Pajareya 2011; Roberts 2011; Siller 2012). In six studies, interventions were centre‐ or clinic‐based (Jocelyn 1998; Aldred 2004; Silva 2009; Green 2010; Kasari 2010; Wong 2010), and Roberts 2011 contrasted the home‐based intervention with a centre‐based intervention. One study used a combination of both home and centre locations for the delivery of the intervention to parents (Carter 2011). In two studies, it was unclear which locations were predominantly used for intervention delivery (Tonge 2006/Tonge 2012; Casenhiser 2011). Most of the interventions were delivered on a one‐to‐one basis but, in three studies, intervention was predominantly delivered to groups of parents (Jocelyn 1998; Tonge 2006/Tonge 2012; Carter 2011). In one study, the mode of delivery was self training from a manual and videotapes (Nefdt 2010). Across the studies, the implementation of taught techniques by parents was not constrained to particular settings; indeed, most encouraged parents to apply techniques opportunistically, thus generalising interventions into the child's natural environment, including home and out of home locations.

Duration

The duration of intervention in the included studies varied greatly. Interventions lasted one week (Nefdt 2010), two weeks (Wong 2010), eight weeks (Kasari 2010), 12 weeks (Jocelyn 1998; Pajareya 2011; Siller 2012), 14 weeks (Carter 2011), 20 weeks (Tonge 2006/Tonge 2012; Silva 2009), 40 weeks (Roberts 2011), one year (Drew 2002; Aldred 2004; Rickards 2007; Green 2010; Smith 2010; Casenhiser 2011), and two years (Dawson 2010). The intensity of interventions varied considerable across studies and is presented in detail under the interventions section of individual studies in the Characteristics of included studies section of this review. Intensity of interventions mostly ranged between two to three hours per session, either in a group or for the parent and child with a therapist. Sessions were delivered at intervals ranging from twice a day through weekly and even monthly schedules. While most studies reported intensity and duration of interventions, many did not report outcome data immediately post‐intervention, rather reporting values that were obtained at some later specified follow‐up point.

Location

The studies were conducted in six countries: seven were conducted in the USA (Smith 2000; Silva 2009; Dawson 2010; Kasari 2010; Nefdt 2010; Carter 2011; Siller 2012), three in the UK (England) (Drew 2002; Aldred 2004; Green 2010), three in Australia (Tonge 2006/Tonge 2012; Rickards 2007; Roberts 2011), two in Canada (Jocelyn 1998; Casenhiser 2011), one in Thailand (Pajareya 2011), and one in China (Hong Kong) (Wong 2010).

Moderators

In eight of the studies (Drew 2002; Aldred 2004; Silva 2009; Dawson 2010; Nefdt 2010; Wong 2010; Pajareya 2011; Roberts 2011), small numbers precluded analyses to explore the potential impact of moderators such as child's age, child's baseline IQ and parental education or family socioeconomic status on reported outcomes. However, nine studies did report their attempts to explore the impact of certain baseline factors on the reported outcomes of interest. Carter 2011 reported that their intervention particularly facilitated communication development in children with lower levels of interest in objects at time 1. In Casenhiser 2011, changes in language scores were found to be significantly predicted by pre‐treatment developmental quotient (DQ), initiation of joint attention and involvement; they also identified enjoyment of interaction as a marginally significant predictor of outcome following intervention. In Green 2010, analyses were carried out to assess the impact of baseline characteristic such as age, socioeconomic status, parental education, child ability (non‐verbal ability and language level), centre where study was conducted and severity of disability on severity of autism symptoms. However, in this large study it was found that these variables did not significantly impact on the findings (which is a reduction of autism social‐communication symptom severity). Similarly, in Jocelyn 1998 analyses of the impact of baseline IQ and severity of autism symptoms as measured by the Childhood Autism Rating Scale (CARS) were conducted and these variables were not found to have any impact on the analysis. However, the level of reduction in autism symptomatology in this study failed to reach statistical level of significance and was only described as a trend. Analyses carried out by Kasari 2010 suggested that higher carer quality of involvement scores significantly predicted increased joint engagement scores post‐intervention. The relationship between family stress and outcome of intervention was explored by Rickards 2007, identifying that families with greater stress levels were likely to benefit most from the additional home‐based sessions added to nursery intervention for the child. Smith 2000 analysed the relationship between diagnostic grouping, intake IQ and early mastery of verbal skills; gains in IQ were not related to baseline IQ, but were greater for children who had a diagnosis of PDD‐NOS rather than autism, and for children who made greater progress within the first three months of treatment. In contrast, Siller 2012 suggested that only children with expressive language skills below 12 months evidenced reliable treatment effects on language outcomes following intervention. Furthermore, it was concluded that only parents classified as insightful at baseline effectively changed their communication in response to the experimental intervention. Finally, analyses conducted by Tonge 2006/Tonge 2012 suggested that the effect of treatment was dependent on pre‐test levels of communication skills; children whose parents were in the parent education and behaviour management (PEBM) group and who had lower pre‐treatment scores in communication skills domains (and by implication lower communication skills levels) registered greater gains in this domain post‐intervention when compared with the control group.

Training fidelity

Most studies reported on the fidelity of training delivered by therapists to parents. However, only five studies provided information on parent implementation of intervention procedures (adherence). For some studies this related to self report measures of either number of hours (Dawson 2010; Pajareya 2011) or weekly examination of carer diaries and carer involvement scales (Kasari 2010). In Nefdt 2010, parent‐recorded videos were scored for fidelity of implementation of techniques, and parents self rated their level of confidence. Similarly, in Casenhiser 2011, video recordings were scored for fidelity of implementation of intervention techniques. However, in the remaining 13 studies, the authors did not report on adherence/parent implementation. The picture of what intervention children actually received is further complicated by scant details provided in most studies about the additional services and interventions being received by children as 'treatment as usual'. Such information is provided only by Dawson 2010, Green 2010 and Kasari 2010.

Outcome measures

Another important factor to note about the studies included in this review is the wide variation in outcome measures used. Thus, combining the results from these measures in a meta‐analysis, and by implication the inferences that may follow from such analysis, can be potentially misleading. Since a number of these measures were not applied consistently across studies, data were aggregated only across outcome measures of similar psychometric properties that assess outcomes of similar constructs. A summary of the outcome measures aggregated and a definition of these outcome measures have been included (Table 3; Table 4).

Open in table viewer
Table 3. Principal meta‐analyses table

Outcome

Study ID (measure)

Intervention group

Control group

n

Follow‐up mean (SD)

n

Follow‐up mean (SD)

 

 

 

 

 

Child communication and social development

Language ‐ joint (direct assessment)

Jocelyn 1998 (EIDP/PSDP)

16

30.2 (15.0)

19

21.9 (9.6)

Casenhiser 2011 (PLS/CASL)

16

0.72 (0.39)

13

0.64 (0.32)

Communication (reported)

Green 2010 (Teacher VABS)

77

64·3 (17·7)

75

67·7 (17·5)

Aldred 2004 (VABS) ¤

14

71.0 (24.82)

14

60.93(20.26)

Dawson 2010 (VABS) ^^

24

82.1 (21.8)

24

69.4 (15.8)

Language – expression (direct assessment)

Green 2010 (PLS)

77

20·0 (11·2)

75

20·0 (11·3)

Dawson 2010 (MSEL) ^^

24

36.6 (13.6)

24

30.0 (9.2)

Siller 2012 (MSEL) ¶¶

34

19.35 (10.72)

30

18.57 (11.25)

Language – expression (reported)

Green 2010 (MCDI) $

77

171·9 (150·7)

75

163.8 (144·3)

Aldred 2004 (MCDI)

14

314.00 (230.98)

14

189.50 (237.81)

Drew 2002 (MCDI)

12

96.6 (118.8)

12

44.0 (50.2)

Language – comprehension (Direct assessment)

Green 2010 (PLS)

77

21·5 (13·0)

75

20·3 (12·8)

Dawson 2010 (MSEL) ^^

24

40.0 (16.3)

24

31.5 (10.6)

Language – comprehension (reported)

Green 2010 (MCDI) $

77

233·7 (129·6)

75

209·0 (131·3)

Aldred 2004 (MCDI)

14

265.43 (134.56)

14

219.14 (147.89)

Drew 2002 (MCDI)

12

176.1 (121.9)

12

100.3 (80.2)

Parent‐child interaction

Green 2010 (shared attention time)

77

64·0% (25·7)

75

55·6% (25·7)

Kasari 2010 (Joint engagement) *

19

42.85 (19.96)

16

27.87 (14.01)

Aldred 2004 (Parent Shared Attention)

14

88.60 (8.70)

14

80.30 (30.10)

Child initiations (coding of parent‐child interactions)

Green 2010 (Parent‐child interaction: child initiations)

77

34·9% (19·7)

75

26·0% (17·5)

Carter 2011 (Frequency of initiating joint attention ESCS)**

24

10.33 (9.82)

25

8.68 (9.26)

Kasari 2010 (Frequency of joint attention initiations)

19

3.11 (3.41)

19

3.77 (3.76)

Casenhiser 2011 (Initiation of Joint Attention ‐ mCBRS)*

16

1.84 (0.549)

13

1.23 (0.430)

Child problem behaviour

Severity of autism characteristics

Dawson 2010 (ADOS severity) ^^

24

7.0 (1.9)

24

7.3 (1.8)

Jocelyn 1998 (Autism behaviour checklist)

16

48.5 (18.4)

19

56.5 (20.1)

Pajareya 2011 (CARS)

16

34.50 (6.01)

16

37.13 (7.42)

Drew 2002 Reciprocal social interaction (ADI‐R)

12

18.3 (4.9)

12

20.1 (4.3)

Green 2010 (Social‐communication algorithm ADOS)

77

9·2 (3·0)

75

9·8 (2·9)

Aldred 2004 (Social‐communication algorithm ADOS)

14

7.70 (3.80)

14

10.70 (3.20)

Child ability

Adaptive behaviour

Dawson 2010 (VABS)

24

68.7 (15.9)

21

59.1 (8.8)

Green 2010 (VABS)

77

60.3 (15.2)

75

62.8 (14.8)

Parental outcome

Parent stress

Drew (PSI)

10

104.3 (20.0)

10

112.1 (20.1)

Jocelyn (Mothers Stress on the Stress‐Arousal Checklist)

16

5.9 (7.0)

19

6.3 (6.1)

Parent‐child interaction (parent synchrony)

Green 2010 (Parent synchrony)

77

51·3% (19·6)

75

32·6% (14·0)

Aldred 2004 (Parent synchrony)

14

65.10 (14.30)

14

49.50 (18.90)

Siller 2012 (Maternal synchronisation)

34

0.73 (0.21)

30

0.61 (0.20)

Notes: ^^ second‐year outcome, $ raw score , ¤ age equivalents in months was provided in the article but was transformed to standardised scores from individual data, ¶¶ time 3 values were extracted from the article, * coded child‐parent free play, ¶ only time 2 values were available in the study,** outcome data collected at time 3 (9 months into study).

ADI‐R: Autism Diagnostic Interview‐Revised; ADOS: Autism Diagnostic Observation Schedule; CARS: Childhood Autism Rating Scale; CASL: Comprehensive Assessment of Spoken Language; EIDP: Early Intervention Developmental Profile; MCDI: MacArthur Communicative Development Inventory; MSEL: Mullen Scales of Early Learning; PLS: Preschool Language Scales; PSDP: Preschool Developmental Profile; SD: standard deviation; VABS: Vineland Adaptive Behaviour Scales.

Open in table viewer
Table 4. Definition of abbreviations used in the principal meta‐analyses table and direction of scales

IMPROVEMENT INDICATED BY INCREASING SCORE

IMPROVEMENT INDICATED BY DECREASING SCORE

Comprehensive Assessment of Spoken Language, quotients (CASL)

Early Intervention Developmental Profile (EIDP)

Early Social Communication Scales (ESCS)

MacArthur Communicative Development Inventory (MCDI)

Modified Child Behaviour Rating Scales (mCRBS)

Mullen Scales of Early Learning (MSEL)

Parent child Free Play Procedures (PCFP)

Preschool developmental profile (PSDP)
Preschool Language Scale (PLS) IV

Vineland Adaptive Behavior Scales II (VABS)

Autism Behaviour Checklist (ABC)

Autism Diagnostic Interview – Revised (ADI ‐ R)

Autism Diagnostic Observation Schedule (ADOS)

Parenting Stress Index (PSI) (Short form)
Childhood Autism Rating Scale (CARS)

Stress‐Arousal Checklist (SACL)

Missing data

Missing data were judged to have been well accounted for in this review based on Cochrane guideline for assessing risk of bias (see 'Risk of bias' assessment of studies under Characteristics of included studies section and Figure 2 for details). In one paper (Carter 2011) included in the meta‐analysis, we thought the degree of missing data could lead to an alteration in our estimates and in this case we conducted a sensitivity analysis to assess the impact of missing data by excluding the study from the analysis (Analysis 1.8). It was observed that there was no change in the conclusions that could be made from the analysis before and after the sensitivity analysis. In the meta‐analyses, estimates from 499 participants (that is 54.2% of total study participants) were combined.

Excluded studies

Following the electronic search, handsearching of articles and reference lists and a process of sifting as described above, 23 articles were excluded. Eight articles were excluded as they were not "parent‐mediated" interventions and instead delivered by therapists or education staff (Warreyn (unpublished); Giarelli 2005; Sallows 2005; Kasari 2006; Yoder 2006; Whalen 2010; Landa 2011; Kaale 2012). Six were excluded as the primary focus was on child participants outside the age range of interest (Whittingham 2009; Frankel 2010; Lopata 2010; Sofronoff 2011; Sung 2011; Gantman 2012). Five articles were excluded as they did not provide results separately for the ASD group included in their studies (Rickards 2009; Shin 2009; Balkom 2010; Romski 2010; Tang 2011). Three were excluded as they were not full RCTs either in the process of recruitment or in the process of analysis (Oosterling 2010; Smith 2010; Silva 2011). Finally, one article (Gulsrud 2010) presents a descriptive analysis of participants in one of the included studies (Kasari 2010), not separated by condition. Further details about reasons for exclusion can be found in the Characteristics of excluded studies table.

Risk of bias in included studies

Risk of bias was assessed across a number of domains in the articles included in this review. Figure 3 provides a 'Risk of bias' graph showing review authors' judgements about each risk of bias item presented as percentages across all included studies. Figure 2 provides a 'Risk of bias' summary on review authors' judgements about each risk of bias item or domain for each included study.

Allocation

Eleven of the articles were judged to have conducted the randomisation process in a way that fell within the 'low risk' for bias category (Figure 2; Figure 3). Low and unclear levels of risk of bias were pre‐specified as acceptable for the inclusion of studies in a meta‐analysis. Only one study (Nefdt 2010) was judged to be at 'high risk' for bias in this domain (Figure 2). The authors did not report the randomisation sequence generation process, and they stated "Parents were randomly assigned to either [condition] based on the order in which the family information was received".

Blinding

The domain "Blinding of participants and personnel (performance bias)" was rated as 'high risk' for all the studies; as parents deliver the intervention, it is not logically possible to blind parents and training personnel from the intervention being delivered. Therefore, care was taken in the meta‐analysis to analyse parent‐report measures separately from direct (blinded) assessment measures. However, detection bias was not found to be a major issue; studies did ensure that the outcome assessors were blind to the group status and baseline characteristics of the participants. Only one article (Drew 2002) was judged to be at 'high risk' for detection bias.

Incomplete outcome data

Fourteen of the studies addressed attrition in ways judged to be at 'low risk' or 'unclear risk' of bias (Figure 2; Figure 3). Three studies (Nefdt 2010; Carter 2011; Roberts 2011) were judged to be at high risk of bias in this domain because they reported group results for those completing intervention, with considerable attrition.

Selective reporting

Fifteen studies reported their results in ways judged to be at 'low risk' or 'unclear risk' of bias. Two studies were judged to be at high risk of bias in this domain either because emphasis was placed upon an outcome ('change in diagnosis') not pre‐specified (Dawson 2010), or because results relating to only some of the measures used were reported (Rickards 2007).

Summary

In summary, only the study Jocelyn 1998 was found to have low risk of bias in all domains.

Effects of interventions

See: Summary of findings for the main comparison

Ten studies were included in meta‐analyses (Jocelyn 1998; Drew 2002; Aldred 2004; Dawson 2010; Green 2010; Kasari 2010; Carter 2011; Casenhiser 2011; Pajareya 2011; Siller 2012); this group of studies all evaluated intervention focusing on parent interaction style in facilitating children's communication and had a 'treatment as usual' comparison condition (that is, the intervention condition was more intensive than the control; Appendix 2). The numbers of participants involved in meta‐analyses ranged from 55 in the smallest analysis to 316 in the biggest (see Data and analyses section).

Primary outcomes

Child communication and social development
1. Language development

A range of measures were used to assess language development. Outcomes have been reported in studies either as a total score or as comprehension and expression scores separately. Measures used included either direct assessments by blinded outcome assessors or parent/carer reports and so have been grouped accordingly in analyses.

  • Comprehension (direct or independent assessment) : a random‐effects meta‐analysis (SMD) of end point means obtained from two studies (Dawson 2010; Green 2010) involving 200 children indicated a non‐significant difference between the parent‐mediated intervention group and the control condition (SMD 0.29; 95% CI ‐0.20 to 0.78, P value > 0.05) (Analysis 1.5). From the Forest plot, it is clear that the two studies are inconsistent in their effects size estimates (Figure 4). The quality of evidence for this outcome was assessed as 'low' based on the GRADE approach for assessing quality of evidence (summary of findings Table for the main comparison) (GRADE working group).

    Figure 4
    Forest plot of comparison: 1 Child communication and social development, outcome: 1.5 Language – Comprehension (direct or independent assessment).

    Forest plot of comparison: 1 Child communication and social development, outcome: 1.5 Language – Comprehension (direct or independent assessment).

  • Comprehension (parent report) : a random‐effects meta‐analysis of end point means obtained from three studies (Drew 2002; Aldred 2004; Green 2010) using the same measure (Macarthur Communicative Development Inventory) and involving 204 children found the presence of a statistically significant effect in favour of the parent‐mediated intervention group (MD 36.26; 95% CI 1.31 to 71.20, P value < 0.05) (Analysis 1.6). This estimate was fairly consistent across the studies combined in the meta‐analyses.

  • Expression (direct or independent assessment) : a random‐effects meta‐analysis (SMD) of end point means from three studies (Dawson 2010; Green 2010; Siller 2012) involving 264 children indicated a non‐significant effect on language expression between the parent‐mediated intervention group and the control condition (SMD 0.14; 95% CI ‐0.16 to 0.45), P value > 0.05) (Analysis 1.3). This estimate was consistent across the studies combined in this analysis. Again, the quality of evidence was rated as 'low' (summary of findings Table for the main comparison) (GRADE working group).

  • Expression (parent report) : the random‐effects meta‐analysis of data obtained from three studies (Drew 2002; Aldred 2004; Green 2010) involving 204 children, that used the same measure (Macarthur Communicative Development Inventory) a non‐significant difference between the parent‐mediated intervention group and the control condition (MD 29.44; 95% CI ‐14.99 to 73.86, P value > 0.05). This estimate was consistent across the studies combined in the meta‐analysis (Analysis 1.4).

  • Joint language (direct or independent assessment) :a random‐effects meta‐analysis of means obtained from two studies (Jocelyn 1998; Casenhiser 2011) involving 64 children indicated a statistically non‐significant effect on joint language between the parent‐mediated intervention group and the control condition (SMD 0.45; 95% CI ‐0.05 to 0.95, P value = 0.08). In the two studies combined, there was considerable overlap in their CIs and a fairly consistent effects size estimate (Analysis 1.1). The quality of evidence was rated as 'low' (summary of findings Table for the main comparison) (GRADE working group).

  • Child communication (parent or teacher report) : a random‐effects meta‐analysis of means obtained from three studies (Aldred 2004; Dawson 2010; Green 2010) that used the same measure (Vineland Adaptive Behavior Scales) and involving a total of 228 children indicated a statistically non‐significant effect between the parent‐mediated intervention group and the control condition (MD 5.31; 95% CI ‐6.77 to 17.39, P value > 0.05) (Analysis 1.2). From the Forest plot (Figure 5), it is evident that one study (Green 2010) had an effect size that is at odds with the other two studies (Aldred 2004; Dawson 2010). This difference may be explained by the method of reporting adopted (teacher reports in Green 2010 as opposed to parent reports in Aldred 2004 and Dawson 2010). When Green 2010 was excluded from the analysis, the results for parent‐reported child communication was a statistically significant difference in favour of the parent‐mediated group (MD 11.93; 95% CI 2.87 to 21.00, P value < 0.05) . The quality of evidence was rated as 'low' for this outcome (summary of findings Table for the main comparison) (GRADE working group). The overall evidence for this outcome is therefore inconclusive.

    Figure 5
    Forest plot of comparison: 1 Child communication and social development, outcome: 1.2 Communication (reported).

    Forest plot of comparison: 1 Child communication and social development, outcome: 1.2 Communication (reported).

These analyses indicate that directly assessed language effect size estimates were fairly consistent across studies, and though not statistically significant they suggest an effect of intervention upon children's language skills. Assessments of language development in the meta‐analyses of data obtained by parent report indicated a similar pattern (apart from report of expressive vocabulary) suggesting that lack of blinding of parents may not distort estimates of effect for language development.

2. Social communication skills

Carter 2011 used a directly observed assessment measure, the Early Social Communication Scales, and found no difference after intervention. Studies using a parent‐report of socialisation skills, the Vineland Adaptive Behavior Scales (Smith 2000; Roberts 2011; Tonge 2012) found a varied picture depending on the contrast condition, but suggesting some improvement within the more intensive treatment group. Despite the treatment focus being on physical massage, Silva 2009 also reported a significant improvement on a measure of teacher‐reported social and language skills.

3. Skills in interaction with parent

Several studies included a standard situation where parent and child played with toys for 10 to 20 minutes. The filmed session was then coded by observers blind to group and time. Three proportional measures were derived: duration of time in which parent and child showed shared attention to each other or joint attention to the same toy; proportion of child communicative acts that were spontaneous initiations of verbal, vocal or non‐verbal communication to the parent and proportion of parent communicative acts that were parent responses to the child that had a quality of synchrony, that is, picking up the child's focus of interest, making statements (and not making requests or changing focus).

  • Shared or joint attention (coding of parent‐child interactions) : the random‐effects meta‐analysis of means obtained from three studies (Aldred 2004; Green 2010; Kasari 2010), which recruited 215 children, indicated that there was a statistically significant small effect in favour of the parent‐mediated intervention group (SMD 0.41; 95% CI 0.14 to 0.68, P value < 0.05) (Analysis 1.7).

  • Child initiations (coding of parent‐child interactions) : a random‐effects meta‐analysis of means obtained from four studies (Green 2010; Kasari 2010; Carter 2011; Casenhiser 2011), which involved 268 children, indicated that there was no statistically significant difference between the intervention and control groups (SMD 0.38; 95% CI ‐0.07 to 0.82, P value > 0.05) (Analysis 1.8). There was moderate heterogeneity between the studies combined for this outcome (I2 = 60%).

Parents' level of stress

The reported findings from studies not included in the meta‐analysis had similar conclusions (Aldred 2004; Tonge 2006; Rickards 2007; Roberts 2011) (Table 5).

Open in table viewer
Table 5. Numerical data from studies not included in meta‐analyses

Outcomes

Study ID

Intervention group

(at follow up)

Control group (at follow up)

Comments

Language development (comprehension and expression)

Smith 2000

RDLS (total): 61.33 (SD 31.88)

RDLS (comprehension): 33.00 (SD 16.86)

RDLS (expressive): 36.23 (SD 21.19)

VABS (communication): 60.77 (SD 17.26)

 

RDLS (total): 87.40 (SD 46.21)

RDLS (comprehension):42.87 (SD 22.29)

RDLS (expressive): 44.53

(SD 23.48)

VABS (communication): 67.87 (SD 30.08)

 

Statistically significant difference was reported in favour of the intensive therapist‐delivered control group, only on directly assessed total language score

Roberts 2011

RDLS (comprehension standard score): 2.6

(SD 8.4)

RDLS (expression standard score): 2.8

(SD 7.5)

VABS (communication): 68.4 (SD 15.6)

RDLS (comprehension standard score): 10.5 (SD 17.4)

RDLS (expression standard score): 7.0 (SD 15.1)

VABS (communication): 76.1 (SD 17.1)

 Significant difference in favour of the centre‐based control group over the home‐based group reported for language comprehension (not expression or reported communication)

Tonge 2012

RDLS (comprehension): PEBM 14.06 (SD 19.67)

RDLS (expressive): 17.17 (SD 17.07)

VABS (communication): PEBM 71.71 (SD 19.83)

RDLS (comprehension): PEAC 5.45 (SD 12.29)

RDLS (expressive: PEAC 8.19  

(SD 13.65)

VABS (communication):                

PEAC 58.03 (SD 15.71)

Analyses conducted by the authors suggest that differences between groups at baseline could have accounted for observed differences at follow‐up

Social communication skills

 

Roberts 2011

VABS (social): Home‐based 66.4 (SD 7.7)

VABS (social): Centre‐based 72.6 (SD 11.2)

Significant difference in favour of the centre‐based control group over the home‐based group

Silva 2009  

PDDBI (language and social): 56.7 (SD 9.7)

PDDBI (language and social): 47.6 (SD 12.1)

Blinded teacher evaluations indicated that children treated with massage made significant classroom improvement in social and language skills

Smith 2000

VABS (social): 68.92 (SD 16.94)

VABS (social): 66.33 (SD 24.78)

No difference between groups

Tonge 2012

VABS (social): PEBM 73.31 (SD 16.59)

VABS (social): PEAC 63.03 (SD 15.53)

Significant difference in favour of the PEBM group

Parents'level of stress

Tonge: 2006

GHQ (total): PEBM 17.06 (SD 7.59)

GHQ (total): PEAC 16.70 (SD 10.51)

No difference between groups

Roberts 2011  

PSI: Home‐based 92.7 (SD 20.9)

PSI: Centre‐based 98.2 (SD 20.1)

No difference between groups

Rickards 2007

QRS‐F

QRS‐F

No significant difference between groups in family levels of resources and stress

Aldred 2004

PSI

PSI

Covarying for baseline scores, there was no significant difference between the groups in change in total PSI score (F = 0.29, P value = 0.597)

Developmental/intellectual gains

Tonge 2012

PEP‐R DQ: PEBM ‐ 72.18

(SD 24.77)

PEP‐R DQ: PEAC ‐  53.94

(SD 23.80)

No difference between groups

Smith 2000

IQ: 49.67 (SD 19.74)

 

IQ: 66.49 (SD 24.08)

 

Authors reported that the intensive therapist‐delivered control group outperformed the parent‐mediated group at the end of the study

Rickards 2007

IQ: 57.2 (SD 21.9)

IQ: 48.6 (SD 17.5)

The authors reported that there was non‐significant trend in favour of intervention

Adaptive behaviour

Rickards 2007

VABS (composite): 64.3 (SD 20.4)

VABS (composite): 59.2 (SD 19.7)

The authors reported that there was significant main effect in favour of intervention

Smith 2000

VABS (composite): 58.50

(SD 16.58)

VABS (composite): 61.19

(SD 29.72)

No observed difference between treatment and control groups on adaptive functioning at the end of the study

Restricted and repetitive behaviour

(autism severity included)

Tonge 2012

CARS: PEBM 35.86 (SD 6.14),

DBC ‐ ASA: PEBM 20.77 (SD 9.05)

CARS: PEAC 40.06 (SD 5.44) DBC ‐ ASA: PEAC 23.90 (SD 8.23)

No difference between groups

Silva 2009

ABC: 33.9 (SD 18.6)

ABC: 59.4 (SD 35.4)

Blinded teacher evaluations indicated that children treated with massage made significant classroom reduction in autism characteristics

Maladaptive behaviour

Smith 2000

CBC (withdrawal): 55.0 (SD 4.40)

CBC (social problems): 59.43

(SD 8.02)

CBC (aggression): 55.71

(SD 5.53)

CBC (withdrawal): 61.89 (SD 7.04)

CBC (social problems): 59.78

(SD 9.59)

CBC (aggression): 60.0 (SD 10.81)

No significant differences reported by teachers between groups on this domain

Roberts 2011

DBC total: Home‐based 52.9

(SD 29.3)

DBC total: Centre‐based 55.7

(SD 19.5)

There were no statistically significant differences between the 2 groups

Rickards 2007

PBCL: 13.7 (SD 8.3)

PBCL: 21.2 (SD 6.7)

 

The authors reported that there was non‐significant trend in favour of intervention

Tonge 2012

DBC – TBPS: PEBM 53.29

(SD 24.36)

DBC – TBPS: PEAC 57.61

(SD 19.72),

No difference between groups

Parents'confidence in coping with child's disability and

behaviour problems

Nefdt 2010

‐

‐

Significant differences between the intervention group and control group; parents appeared more confident during interactions with their child following the self directed learning programmes (F = 16.37 and P value = 0.001 with an effect size of 1.28)

Any cost information provided by the authors

Roberts 2011

‐

‐

The cost per child was AUD6383.  Staff hours were the same for home based and centre based

Rickards 2007

‐

‐

Teacher's salary for 1 year plus some expenses for travelling (this travelling estimate was not provided)

ABC: Autism Behaviour Checklist; BRS: Bayley Behaviour Rating Scale; CARS: Childhood Autism Rating Scale; CBC: Child Behaviour Checklist; DBC: Developmental Behaviour Checklist; DBC‐ASA: Developmental Behaviour Checklist Autism Screening Algorithm; DBC‐TBPS: Developmental Behaviour Checklist Total Behaviour Problem Score; GHQ: General Health Questionnaire; IQ: intelligence quotient; PBC: Preschool Behaviour Checklist; PEAC: Parent Education and Counselling; PEBM: parent education and behaviour management; PBCL: Preschool Behaviour Checklist; PDDBI: Pervasive Developmental Disorders Behaviour Inventory; PEP‐R DQ: Psychoeducational Profile‐Revised – Developmental Quotient; PSI: Parenting Stress Index; QRS‐F: Questionnaire on Resources and Stress (Friedrich short form); RDLS: Reynell Developmental Language Scales; SD: standard deviation; VABS: Vineland Adaptive Behaviour Scales.

Secondary outcomes

Child ability

  • Developmental/intellectual gains: five studies (Smith 2000; Drew 2002; Rickards 2007; Dawson 2010; Tonge 2012) with varying theoretical basis and methods for assessing developmental/intellectual gains reported on this outcome. Dawson 2010 and Rickards 2007 suggest that small gains were made in this domain following intervention. However, Drew 2002 and Tonge 2012 (individual and group intervention, respectively) did not report any difference in this domain between intervention and control groups following intervention. Smith 2000 found greater gains for the intensive therapist‐delivered intervention condition. Evidence for gains from parent‐mediated intervention therefore may be suggested. However, gains in formal assessment may in part reflect child co‐operation.

  • Adaptive behaviour: based on random‐effects meta‐analysis of means from two studies (Dawson 2010; Green 2010), which recruited 197 children and used the Vineland Adaptive Behavior Scales composite score, found no statistically significant difference between the intervention and control groups following intervention (MD 1.06; 95% CI ‐2.95 to 5.06, P value > 0.05) (Analysis 1.10). Green 2010 reported teacher interview (after one year), and Dawson 2010 reported parent interview (after two years). Smith 2000 found no difference between parent‐mediated and intensive therapist‐mediated groups on adaptive behaviour (after four years), though Rickards 2007 did suggest a small effect in favour of additional parent‐mediated intervention (after one year).

Child problem behaviour

  • Maladaptive behaviour : four studies (Smith 2000; Tonge 2006/Tonge 2012; Rickards 2007; Roberts 2011) reported on this outcome. Due to significant and important differences between these studies in theoretical basis and outcome measures used, a meta‐analysis could not be conducted. None found a significant difference in maladaptive behaviour in favour of the intervention group, even where that was the focus of intervention (Tonge 2012).

  • Severity of autism characteristics : autism severity was assessed across the studies using a number of measures (Table 3; Table 4). A random‐effects meta‐analysis of means obtained from six studies (Jocelyn 1998; Drew 2002; Aldred 2004; Dawson 2010; Green 2010; Pajareya 2011), which recruited 316 children, indicated there was a statistically significant effect in reduced severity of autism characteristics after intervention between the parent‐mediated group and the control condition (SMD ‐0.30; 95% CI ‐0.52 to ‐0.08, P value < 0.05). There was no significant heterogeneity between the studies combined in the analysis (Analysis 1.9) (Figure 6).

    Figure 6
    Forest plot of comparison: 1 Child communication and social development, outcome: 1.9 Autism severity.

    Forest plot of comparison: 1 Child communication and social development, outcome: 1.9 Autism severity.

Silva 2009 found a significant differential reduction in autism characteristics reported by teachers after a massage intervention, though Tonge 2012 found no differential effect of their group intervention on two separate measures.

Parental outcomes

  • Parents' satisfaction with therapy : only one study reported on this outcome and it was stated that "customer" satisfaction was high with mean ratings of 5.48 (out of 6) on the group experience questionnaire and 3.46 (out of 4) on the group leader experience questionnaire (Carter 2011).

  • Parents' confidence in coping with child's disability and behaviour problems : in the Nefdt 2010 (Table 5) study, observers rated parent confidence in carrying out procedures with their child; parents in the intervention group appeared to be more confident compared with those in the control conditions.

Cost of intervention

  • The majority of studies did not provide information on the cost of intervention. One study stated that the cost of the home‐based programme was one teacher's salary for one year plus some expenses for travelling (this travelling estimate was not provided) (Rickards 2007). In another study, the authors estimated that the cost per child was AUD6383, adding that this is a small cost compared to a range of other interventions currently available to children and families with autism in that area (Roberts 2011). Casenhiser 2011 reported that the intervention programme (MEHRIT) cost is approximately CAD5000 per child per year, which they considered to be less than the estimates of therapy for most therapist‐delivered programmes that typically provide 20 to 30 hours treatment. In two studies, though no detailed cost analyses were reported, the authors hinted that the parent‐mediated interventions would cost less than available alternatives as they employed low‐cost and widely applicable "professional‐as‐consultant, parent‐as‐therapist" models (Drew 2002; Pajareya 2011). In general, most studies tended to report that the interventions would be more cost effective in the long run when compared with existing alternatives. However, to reach a definite conclusion about the cost effectiveness of intervention vis‐a‐vis alternative therapies, such arguments require health economic evaluation of the cost implications of the interventions. The study reported by Green 2010 included a health economics evaluation that is expected to be published.

Sensitivity analyses

  • Based on risk of bias

in this analysis, we restricted the included studies only to those we judged to be at low and unclear risk of bias in pre‐specified domains. For the results see Table 2. A statement of "no difference" in this table implies that the overall direction of inference was preserved following the analysis and one of "not applicable" was used to describe analyses in which all studies had low or unclear risk of bias.

Discussion

Summary of main results

Overall, the review and meta‐analyses demonstrate that children with ASD may make gains in language skills following parent‐mediated interventions. These gains have been reported fairly consistently across reviewed studies in directly (blinded) assessed composite language scores, and separately for expressive language, and language comprehension. This pattern of gains was also reported by parents, with statistically significant findings in reported language comprehension (Analysis 1.6). Furthermore, six studies reported on overall severity of children's autism characteristics, with meta‐analysis suggesting reduction in autism severity following intervention (Analysis 1.9). However, this finding certainly requires further investigation in future, especially to explore whether earlier intervention may be more efficacious. The strongest effect found was for the proximal effect of interaction coaching, that is, improvement in parent synchrony in observed interaction (Analysis 2.2). This is the assumed mediator for positive child outcomes from intervention (Aldred 2012), along with an increase in shared or joint attention during play. In this review, gains in child initiations within observed interaction, improvement in social communication skills, and reductions in child maladaptive behaviour have not been consistently found across studies, with effect estimates associated with significant degrees of uncertainty. The gains reported in the studies, especially in language development, could have the potential to enhance other areas of the child's development. Though the effect sizes for child outcomes were typically small, for a serious neurodevelopmental disorder such as autism even effect sizes of this magnitude could serve as pointers to what may be potentially effective approaches to managing autism in early childhood. How much benefit these gains will add to the longer‐term quality of life of children and families is a question that needs further investigation and follow‐up.

No significant reduction in parent stress was demonstrated in this review (Analysis 2.1). This could be due to the fact that the content of most of the interventions was primarily aimed at improving child outcomes rather than reducing parent stress. It was not possible to explore further questions about parent stress, such as whether the demands of receiving training and delivering intervention might lead to an increase in stress experienced. Rickards 2007 found that families with fewer resources and greater stress benefited most from the addition of home sessions with the parent to the child's experience of autism‐specific nursery.

Overall completeness and applicability of evidence

One strength of a parent‐mediated approach to intervention in autism is that it should be possible to create therapeutic experiences spread out across the child's day and natural environments. This approach, however, makes it difficult to quantify the amount of time spent in therapy. This is because many of the approaches encourage parents to use techniques learned during training sessions in their daily interactions with their children. In order to explore the relationship between duration and intensity of intervention and impact, a consistent system for report of parent implementation would be necessary. As each study had different numbers and distribution of contact training hours with parents, it was also not possible to describe intensity of intervention as delivered to the parent.

The studies reviewed varied in the extent to which they were embedded in community services, and thus approached an evaluation of 'effectiveness' rather than efficacy of the intervention (Smith 2007). It was notable that two community studies (Oosterling 2010; Smith 2010) had to be excluded as compromises had been made to randomisation. These studies raise questions about the feasibility of strict randomisation where young children with ASD share nursery provision, and parents meet in groups for support. Cluster‐randomised studies may be prohibitively expensive. A further issue identified is the tension between, on the one hand, good research design, which requires evaluation of an intervention where the key elements are clearly specified and controlled, and, on the other hand, the needs of families of children with ASD for comprehensive support and good educational provision for the child.

It was not possible to investigate any potential moderators through subgroup analyses. Though the evidence was variable, some of the individual studies suggested that children of lower ability (however measured) made greater progress in certain domains (for example, expressive language skills in Siller 2012 and pre‐treatment communication skills scores in Tonge 2006/Tonge 2012). This is generally in contrast to the findings of long‐term follow‐up of cohorts (for example, Magiati 2011) and reviews of early intensive behavioural intervention (for example, Howlin 2009) where greater gains are made by higher functioning children. Resolution of questions about moderators of treatment effect requires individual level data meta‐analysis (cf Eldevik 2009; Reichow 2012).

In the introduction we posed some important questions to which parents, professionals and policy makers would like to have answers. What intervention should money be spent on? Which children, and which parents, will benefit most from which intervention? Is there evidence of greater benefit with earlier detection of ASD and thus earlier intervention? How long should interventions last? The available evidence has enabled us to make very little preliminary exploration. The variability in interventions including length and intensity, variability in content and delivery, the lack of direct comparisons between interventions, variability in recruitment and in ways of describing characteristics of children and parents, means that these important questions cannot yet be answered. Therefore, these factors need to be considered when making any generalisations. However, the review provides a basis for concluding that parent‐mediated early intervention does appear on average to lead to positive changes in parent‐child interaction, and to possible gains for children in language, with reduction in severity of autism characteristics.

Quality of the evidence

This review considered 17 RCTs representing 919 young children with ASD. This is a considerable increase in numbers of RCTs since the original review (Diggle 2002), which included only two RCTs. In the meta‐analyses, estimates from 499 children were combined (252 of these estimates were from the individuals in 'intervention' conditions while 247 were from the 'control' conditions), though only a maximum of 316 in one meta‐analysis. For the studies included in the review, we used the Cochrane guideline for risk of bias assessment and it was observed that there were repeated problems with allocation concealment, incomplete outcome data, blinding and reporting bias (including emphasis on outcomes not pre‐specified) (Figure 2; Figure 3). Though all of the studies assessed failed to blind parents (not possible given the nature of the interventions), it is however suggested that this did not impact on the quality of the evidence as most studies also included outcome measures from 'blinded' assessors and these outcomes were considered separately. For important outcomes included in the summary of findings Table for the main comparison, we assessed the overall quality of evidence using the 'GRADE' approach (GRADE working group). Factors taken into consideration before judgements were made (the GRADEpro criteria) include limitations of detailed designs and execution (risk of bias) in studies, inconsistency (heterogeneity), indirectness (population, intervention, comparison and outcome), imprecision and publication bias.

Potential biases in the review process

We relied heavily on published data in the course of this review with limited access to individual participant records; however, attrition was low in 14 of the 17 studies. The individual studies aggregated in this review vary considerably in their design, tools used for measuring outcomes, degree of utilisation of community or local services, as well as definition of control and intervention conditions. This limited our capacity to make full use of published data in the meta‐analyses as we were only able to combine data from studies considered to be most similar in content and control condition.

Agreements and disagreements with other studies or reviews

The findings of this updated review add considerably to the original review (Diggle 2002) as the increased number of studies allows greater power to draw conclusions. However, many of the same difficulties with sample sizes, multiple outcome measures and variable follow‐up times remain. The general direction of findings presented is in keeping with those of other reviews of early intervention (see, for example, Ospina 2008; Warren 2011) though these do not focus on parent‐mediated intervention per se and include a range of ages and research designs. This review suggests more caution in the interpretation of results than is espoused by other reviewers (for example, Rogers 2008) on account of the rigorous consideration of weaknesses in study design and analysis. Even where studies have been well designed (for example, Jocelyn 1998; Dawson 2010; Green 2010) there remain some potential problems with design and biases in their analyses and reporting.

Study flow diagram.
Figures and Tables -
Figure 1

Study flow diagram.

Navigate to figure in ReviewOpen in new tab

original image
Figures and Tables -
Figure 2
Navigate to figure in ReviewOpen in new tab

original image
Figures and Tables -
Figure 3
Navigate to figure in ReviewOpen in new tab

Forest plot of comparison: 1 Child communication and social development, outcome: 1.5 Language – Comprehension (direct or independent assessment).
Figures and Tables -
Figure 4

Forest plot of comparison: 1 Child communication and social development, outcome: 1.5 Language – Comprehension (direct or independent assessment).

Navigate to figure in ReviewOpen in new tab

Forest plot of comparison: 1 Child communication and social development, outcome: 1.2 Communication (reported).
Figures and Tables -
Figure 5

Forest plot of comparison: 1 Child communication and social development, outcome: 1.2 Communication (reported).

Navigate to figure in ReviewOpen in new tab

Forest plot of comparison: 1 Child communication and social development, outcome: 1.9 Autism severity.
Figures and Tables -
Figure 6

Forest plot of comparison: 1 Child communication and social development, outcome: 1.9 Autism severity.

Navigate to figure in ReviewOpen in new tab

Comparison 1 Child communication and social development, Outcome 1 Language ‐ Joint (direct or independent assessment).
Figures and Tables -
Analysis 1.1

Comparison 1 Child communication and social development, Outcome 1 Language ‐ Joint (direct or independent assessment).

Navigate to figure in ReviewOpen in new tab

Comparison 1 Child communication and social development, Outcome 2 Communication (reported).
Figures and Tables -
Analysis 1.2

Comparison 1 Child communication and social development, Outcome 2 Communication (reported).

Navigate to figure in ReviewOpen in new tab

Comparison 1 Child communication and social development, Outcome 3 Language – Expression (direct or independent assessment).
Figures and Tables -
Analysis 1.3

Comparison 1 Child communication and social development, Outcome 3 Language – Expression (direct or independent assessment).

Navigate to figure in ReviewOpen in new tab

Comparison 1 Child communication and social development, Outcome 4 Language – Expression (reported).
Figures and Tables -
Analysis 1.4

Comparison 1 Child communication and social development, Outcome 4 Language – Expression (reported).

Navigate to figure in ReviewOpen in new tab

Comparison 1 Child communication and social development, Outcome 5 Language – Comprehension (direct or independent assessment).
Figures and Tables -
Analysis 1.5

Comparison 1 Child communication and social development, Outcome 5 Language – Comprehension (direct or independent assessment).

Navigate to figure in ReviewOpen in new tab

Comparison 1 Child communication and social development, Outcome 6 Language – Comprehension (reported).
Figures and Tables -
Analysis 1.6

Comparison 1 Child communication and social development, Outcome 6 Language – Comprehension (reported).

Navigate to figure in ReviewOpen in new tab

Comparison 1 Child communication and social development, Outcome 7 Parent‐child interaction (Shared or joint attention time).
Figures and Tables -
Analysis 1.7

Comparison 1 Child communication and social development, Outcome 7 Parent‐child interaction (Shared or joint attention time).

Navigate to figure in ReviewOpen in new tab

Comparison 1 Child communication and social development, Outcome 8 Child initiations (coding of parent child interactions).
Figures and Tables -
Analysis 1.8

Comparison 1 Child communication and social development, Outcome 8 Child initiations (coding of parent child interactions).

Navigate to figure in ReviewOpen in new tab

Comparison 1 Child communication and social development, Outcome 9 Autism severity.
Figures and Tables -
Analysis 1.9

Comparison 1 Child communication and social development, Outcome 9 Autism severity.

Navigate to figure in ReviewOpen in new tab

Comparison 1 Child communication and social development, Outcome 10 Adaptive behaviour.
Figures and Tables -
Analysis 1.10

Comparison 1 Child communication and social development, Outcome 10 Adaptive behaviour.

Navigate to figure in ReviewOpen in new tab

Comparison 2 Parent outcomes, Outcome 1 Parents' level of stress.
Figures and Tables -
Analysis 2.1

Comparison 2 Parent outcomes, Outcome 1 Parents' level of stress.

Navigate to figure in ReviewOpen in new tab

Comparison 2 Parent outcomes, Outcome 2 Parent‐child interaction (parent synchrony).
Figures and Tables -
Analysis 2.2

Comparison 2 Parent outcomes, Outcome 2 Parent‐child interaction (parent synchrony).

Navigate to figure in ReviewOpen in new tab

Child and parent outcomes following parent‐mediated interventions for young children with autism spectrum disorders (ASD)

Patient or population: children aged 1 year to 6 years 11 months with diagnosis of ASD

Settings: home, clinic, or both, interventions delivered through group or 1‐to‐1 sessions

Intervention: parent‐mediated early intervention

Comparison: no treatment or treatment as usual (including alternative child‐centred intervention not mediated by parents, alternative parent‐mediated intervention that differed in some way from the experimental condition and waiting list control groups, for example, intensity)

Outcomes

Illustrative comparative risks* (95% CI)

No of participants
(studies)

Quality of the evidence
(GRADE)

Comments

Assumed risk

Corresponding risk

No treatment or treatment as usual

Parent‐mediated early intervention

Language ‐ joint (direct or independent assessment)

EIDP/PSDP, PLS‐IV/CASL1

‐

The mean language ‐ joint (independent assessment) in the intervention groups was
0.45 standard deviations higher
(0.05 lower to 0.95 higher)

64
(2 studies)

⊕⊕⊝⊝

low2

SMD 0.45 (‐0.05 to 0.95). Higher scores indicate improvement. Duration of intervention ranged from 12 weeks to 1 year. This is small effect 3

Communication (reported)

VABS1

The mean communication (reported) ranged across control groups from
60.93 to 67.7

The mean communication (reported) in the intervention groups was
5.31 higher
(6.77 lower to 17.39 higher)

228
(3 studies)

⊕⊕⊝⊝

low2

The mean value was 5.31 points higher for the intervention group. Higher scores indicate improvement. Duration of interventions ranged from 1 to 2 years

Language – expression (direct or independent assessment)

PLS‐IV, MSEL1

‐

The mean language – expression (direct assessment) in the intervention groups was
0.14 standard deviations higher
(0.16 lower to 0.45 higher)

264
(3 studies)

⊕⊕⊝⊝

low2

SMD 0.14 (‐0.16 to 0.45). Higher scores indicate improvement. Duration of interventions ranged from 12 weeks to 2 years. This is a small and uncertain effect 3

Language – Comprehension (direct or independent assessment)

PLS, MSEL1

‐

The mean language – comprehension (direct assessment) in the intervention groups was
0.29 standard deviations higher
(0.2 lower to 0.78 higher)

200
(2 studies)

⊕⊕⊝⊝

low2

SMD 0.29 (‐0.2 to 0.78). Higher scores indicate improvement. Duration of intervention ranged from 1 to 2 years in the studies. This effect size is small and uncertain 3

Parent‐child interaction (parent synchrony)

‐

The mean parent‐child interaction (parent synchrony) in the intervention groups was
0.9 standard deviations higher
(0.56 to 1.23 higher)

244
(3 studies)

⊕⊕⊝⊝

low2

SMD 0.9 (0.56 to 1.23). Higher scores indicate improvement. Duration of intervention ranged from 12 weeks to 1 year. This is large effect size 3

Parents'level of stress

PSI, SAC1

‐

The mean parents' level of stress in the intervention groups was
0.17 standard deviations lower
(0.7 lower to 0.36 higher)

55
(2 studies)

⊕⊕⊝⊝

low2

SMD ‐0.17 (‐0.7 to 0.36). Lower scores indicate improvement. Duration of interventions ranged from 12 weeks to 1 year. This is a small and uncertain effect 3

CI: confidence interval; SMD: standardised mean difference

GRADE Working Group grades of evidence
High quality: Further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: We are very uncertain about the estimate.

1 CASL: Comprehensive Assessment of Spoken Language, quotients; EIDP: Early Intervention Developmental Profile; MSEL: Mullen Scales of Early Learning; PLS‐IV: Preschool Language Scale‐IV; PSDP: Preschool Developmental Profile; PSI: Parental Stress Inventory; SAC: Stress Arousal Checklist; VABS: Vineland Adaptive Behavior Scales.
2 Quality of evidence has been rated as low across domains due to the small numbers of studies combined.
3 Cohen's interpretation of effect size (rule of thumb): 0.2 is considered a small effect, 0.5 is considered a moderate effect and 0.8 or higher is considered as large effect.

Figures and Tables -
Navigate to table in Review
Table 1. Table showing changes to methods described in the protocol

Section of the review

Issues

What was done

Measurement of treatment effect

No relevant categorical data encountered

No analysis using categorical data was done

Unit of analysis

No cluster randomised trials were identified

No analysis was done

Cross‐over trials

Wong 2010 presented a cross‐over design. Information from this study was used only in the narrative synthesis as the measures used in the study, the way in which results were reported and the theoretical background prohibited its inclusion in the meta‐analysis. Thus the study did not present any "unit of analysis" issue (see Appendix 2 and Characteristics of included studies sections). Only data in the first part of the study, before participants swapped groups, was used in the discussions. 

Studies with multiple treatment groups

Two studies had multiple treatment groups and a control group. Roberts 2011 had a non‐randomised control group, so the two treatment conditions were compared (home based treated as the experimental intervention group, and centre‐based treated as the control group). Data from this study were included only in the narrative synthesis as the measures used in the study and the study methods prohibited its inclusion in the meta‐analysis (see Appendix 2 and Characteristics of included studies sections). Tonge 2006/Tonge 2012 also had 2 treatment arms, with hypothesised greater effects for the treatment with dedicated skills training versus a parent education‐only group. Therefore for this review, the 2 treatment arms were compared with the latter arm as the control.

In future revisions of this review, it is likely that there will be more examples of comparison of multiple‐treatment groups. In that case the authors will first examine studies that compare with no treatment, waiting‐list or treatment as usual, and then examine studies with other treatment controls, so as to interpret the latter in light of the evidence for the former group of studies.

Figures and Tables -
Table 1. Table showing changes to methods described in the protocol
Navigate to table in Review
Table 2. Sensitivity analyses

Analysis

Sequence generation

Allocation concealment

Blinding of outcome

(detection bias)

Attrition

Joint language (direct assessment) Analysis 1.1

Not applicable

Not done (only 1 study had low risk)

Not applicable

Not applicable

Communication (reported) Analysis 1.2

Not applicable

Not done (only 1 study had low risk)

Not applicable

Not applicable

Language – expression (direct assessment) Analysis 1.3

Not applicable

Not done (only 1 study had low risk)

Not applicable

Not applicable

Language – expression (reported) Analysis 1.4

Not applicable

Not done (only 1 study had low risk)

No difference

Not applicable

Language – comprehension (direct assessment) Analysis 1.5

Not done as there were only 2 studies

Language – comprehension (reported) Analysis 1.6

Not applicable

Not done (only 1 study had low risk)

There was a difference $

Not applicable

Parent‐child interaction (shared or joint attention time) Analysis 1.7

Not applicable

Not done (only 1 study had low risk)

Not applicable

Not applicable

Child initiations (coding of parent‐child interactions) Analysis 1.8

Not applicable

Not done (only 1 study had low risk)

Not applicable

No difference

Autism severity Analysis 1.9

Not applicable

There was a difference *

No difference

No difference

Parents' level of stress Analysis 2.1

Not done as there were only two studies

Parent‐child interaction (parent synchrony) Analysis 2.2

Not applicable

Not done (only 1 study had low risk)

Not applicable

Not applicable

$ New effect estimate mean difference 27.63 (95% CI ‐10.94 to 66.20) P value > 0.05.

* New effect estimate standardised mean difference ‐0.20 (95% CI ‐0.52 to 0.12).
Figures and Tables -
Table 2. Sensitivity analyses
Navigate to table in Review
Table 3. Principal meta‐analyses table

Outcome

Study ID (measure)

Intervention group

Control group

n

Follow‐up mean (SD)

n

Follow‐up mean (SD)

 

 

 

 

 

Child communication and social development

Language ‐ joint (direct assessment)

Jocelyn 1998 (EIDP/PSDP)

16

30.2 (15.0)

19

21.9 (9.6)

Casenhiser 2011 (PLS/CASL)

16

0.72 (0.39)

13

0.64 (0.32)

Communication (reported)

Green 2010 (Teacher VABS)

77

64·3 (17·7)

75

67·7 (17·5)

Aldred 2004 (VABS) ¤

14

71.0 (24.82)

14

60.93(20.26)

Dawson 2010 (VABS) ^^

24

82.1 (21.8)

24

69.4 (15.8)

Language – expression (direct assessment)

Green 2010 (PLS)

77

20·0 (11·2)

75

20·0 (11·3)

Dawson 2010 (MSEL) ^^

24

36.6 (13.6)

24

30.0 (9.2)

Siller 2012 (MSEL) ¶¶

34

19.35 (10.72)

30

18.57 (11.25)

Language – expression (reported)

Green 2010 (MCDI) $

77

171·9 (150·7)

75

163.8 (144·3)

Aldred 2004 (MCDI)

14

314.00 (230.98)

14

189.50 (237.81)

Drew 2002 (MCDI)

12

96.6 (118.8)

12

44.0 (50.2)

Language – comprehension (Direct assessment)

Green 2010 (PLS)

77

21·5 (13·0)

75

20·3 (12·8)

Dawson 2010 (MSEL) ^^

24

40.0 (16.3)

24

31.5 (10.6)

Language – comprehension (reported)

Green 2010 (MCDI) $

77

233·7 (129·6)

75

209·0 (131·3)

Aldred 2004 (MCDI)

14

265.43 (134.56)

14

219.14 (147.89)

Drew 2002 (MCDI)

12

176.1 (121.9)

12

100.3 (80.2)

Parent‐child interaction

Green 2010 (shared attention time)

77

64·0% (25·7)

75

55·6% (25·7)

Kasari 2010 (Joint engagement) *

19

42.85 (19.96)

16

27.87 (14.01)

Aldred 2004 (Parent Shared Attention)

14

88.60 (8.70)

14

80.30 (30.10)

Child initiations (coding of parent‐child interactions)

Green 2010 (Parent‐child interaction: child initiations)

77

34·9% (19·7)

75

26·0% (17·5)

Carter 2011 (Frequency of initiating joint attention ESCS)**

24

10.33 (9.82)

25

8.68 (9.26)

Kasari 2010 (Frequency of joint attention initiations)

19

3.11 (3.41)

19

3.77 (3.76)

Casenhiser 2011 (Initiation of Joint Attention ‐ mCBRS)*

16

1.84 (0.549)

13

1.23 (0.430)

Child problem behaviour

Severity of autism characteristics

Dawson 2010 (ADOS severity) ^^

24

7.0 (1.9)

24

7.3 (1.8)

Jocelyn 1998 (Autism behaviour checklist)

16

48.5 (18.4)

19

56.5 (20.1)

Pajareya 2011 (CARS)

16

34.50 (6.01)

16

37.13 (7.42)

Drew 2002 Reciprocal social interaction (ADI‐R)

12

18.3 (4.9)

12

20.1 (4.3)

Green 2010 (Social‐communication algorithm ADOS)

77

9·2 (3·0)

75

9·8 (2·9)

Aldred 2004 (Social‐communication algorithm ADOS)

14

7.70 (3.80)

14

10.70 (3.20)

Child ability

Adaptive behaviour

Dawson 2010 (VABS)

24

68.7 (15.9)

21

59.1 (8.8)

Green 2010 (VABS)

77

60.3 (15.2)

75

62.8 (14.8)

Parental outcome

Parent stress

Drew (PSI)

10

104.3 (20.0)

10

112.1 (20.1)

Jocelyn (Mothers Stress on the Stress‐Arousal Checklist)

16

5.9 (7.0)

19

6.3 (6.1)

Parent‐child interaction (parent synchrony)

Green 2010 (Parent synchrony)

77

51·3% (19·6)

75

32·6% (14·0)

Aldred 2004 (Parent synchrony)

14

65.10 (14.30)

14

49.50 (18.90)

Siller 2012 (Maternal synchronisation)

34

0.73 (0.21)

30

0.61 (0.20)

Notes: ^^ second‐year outcome, $ raw score , ¤ age equivalents in months was provided in the article but was transformed to standardised scores from individual data, ¶¶ time 3 values were extracted from the article, * coded child‐parent free play, ¶ only time 2 values were available in the study,** outcome data collected at time 3 (9 months into study).

ADI‐R: Autism Diagnostic Interview‐Revised; ADOS: Autism Diagnostic Observation Schedule; CARS: Childhood Autism Rating Scale; CASL: Comprehensive Assessment of Spoken Language; EIDP: Early Intervention Developmental Profile; MCDI: MacArthur Communicative Development Inventory; MSEL: Mullen Scales of Early Learning; PLS: Preschool Language Scales; PSDP: Preschool Developmental Profile; SD: standard deviation; VABS: Vineland Adaptive Behaviour Scales.

Figures and Tables -
Table 3. Principal meta‐analyses table
Navigate to table in Review
Table 4. Definition of abbreviations used in the principal meta‐analyses table and direction of scales

IMPROVEMENT INDICATED BY INCREASING SCORE

IMPROVEMENT INDICATED BY DECREASING SCORE

Comprehensive Assessment of Spoken Language, quotients (CASL)

Early Intervention Developmental Profile (EIDP)

Early Social Communication Scales (ESCS)

MacArthur Communicative Development Inventory (MCDI)

Modified Child Behaviour Rating Scales (mCRBS)

Mullen Scales of Early Learning (MSEL)

Parent child Free Play Procedures (PCFP)

Preschool developmental profile (PSDP)
Preschool Language Scale (PLS) IV

Vineland Adaptive Behavior Scales II (VABS)

Autism Behaviour Checklist (ABC)

Autism Diagnostic Interview – Revised (ADI ‐ R)

Autism Diagnostic Observation Schedule (ADOS)

Parenting Stress Index (PSI) (Short form)
Childhood Autism Rating Scale (CARS)

Stress‐Arousal Checklist (SACL)

Figures and Tables -
Table 4. Definition of abbreviations used in the principal meta‐analyses table and direction of scales
Navigate to table in Review
Table 5. Numerical data from studies not included in meta‐analyses

Outcomes

Study ID

Intervention group

(at follow up)

Control group (at follow up)

Comments

Language development (comprehension and expression)

Smith 2000

RDLS (total): 61.33 (SD 31.88)

RDLS (comprehension): 33.00 (SD 16.86)

RDLS (expressive): 36.23 (SD 21.19)

VABS (communication): 60.77 (SD 17.26)

 

RDLS (total): 87.40 (SD 46.21)

RDLS (comprehension):42.87 (SD 22.29)

RDLS (expressive): 44.53

(SD 23.48)

VABS (communication): 67.87 (SD 30.08)

 

Statistically significant difference was reported in favour of the intensive therapist‐delivered control group, only on directly assessed total language score

Roberts 2011

RDLS (comprehension standard score): 2.6

(SD 8.4)

RDLS (expression standard score): 2.8

(SD 7.5)

VABS (communication): 68.4 (SD 15.6)

RDLS (comprehension standard score): 10.5 (SD 17.4)

RDLS (expression standard score): 7.0 (SD 15.1)

VABS (communication): 76.1 (SD 17.1)

 Significant difference in favour of the centre‐based control group over the home‐based group reported for language comprehension (not expression or reported communication)

Tonge 2012

RDLS (comprehension): PEBM 14.06 (SD 19.67)

RDLS (expressive): 17.17 (SD 17.07)

VABS (communication): PEBM 71.71 (SD 19.83)

RDLS (comprehension): PEAC 5.45 (SD 12.29)

RDLS (expressive: PEAC 8.19  

(SD 13.65)

VABS (communication):                

PEAC 58.03 (SD 15.71)

Analyses conducted by the authors suggest that differences between groups at baseline could have accounted for observed differences at follow‐up

Social communication skills

 

Roberts 2011

VABS (social): Home‐based 66.4 (SD 7.7)

VABS (social): Centre‐based 72.6 (SD 11.2)

Significant difference in favour of the centre‐based control group over the home‐based group

Silva 2009  

PDDBI (language and social): 56.7 (SD 9.7)

PDDBI (language and social): 47.6 (SD 12.1)

Blinded teacher evaluations indicated that children treated with massage made significant classroom improvement in social and language skills

Smith 2000

VABS (social): 68.92 (SD 16.94)

VABS (social): 66.33 (SD 24.78)

No difference between groups

Tonge 2012

VABS (social): PEBM 73.31 (SD 16.59)

VABS (social): PEAC 63.03 (SD 15.53)

Significant difference in favour of the PEBM group

Parents'level of stress

Tonge: 2006

GHQ (total): PEBM 17.06 (SD 7.59)

GHQ (total): PEAC 16.70 (SD 10.51)

No difference between groups

Roberts 2011  

PSI: Home‐based 92.7 (SD 20.9)

PSI: Centre‐based 98.2 (SD 20.1)

No difference between groups

Rickards 2007

QRS‐F

QRS‐F

No significant difference between groups in family levels of resources and stress

Aldred 2004

PSI

PSI

Covarying for baseline scores, there was no significant difference between the groups in change in total PSI score (F = 0.29, P value = 0.597)

Developmental/intellectual gains

Tonge 2012

PEP‐R DQ: PEBM ‐ 72.18

(SD 24.77)

PEP‐R DQ: PEAC ‐  53.94

(SD 23.80)

No difference between groups

Smith 2000

IQ: 49.67 (SD 19.74)

 

IQ: 66.49 (SD 24.08)

 

Authors reported that the intensive therapist‐delivered control group outperformed the parent‐mediated group at the end of the study

Rickards 2007

IQ: 57.2 (SD 21.9)

IQ: 48.6 (SD 17.5)

The authors reported that there was non‐significant trend in favour of intervention

Adaptive behaviour

Rickards 2007

VABS (composite): 64.3 (SD 20.4)

VABS (composite): 59.2 (SD 19.7)

The authors reported that there was significant main effect in favour of intervention

Smith 2000

VABS (composite): 58.50

(SD 16.58)

VABS (composite): 61.19

(SD 29.72)

No observed difference between treatment and control groups on adaptive functioning at the end of the study

Restricted and repetitive behaviour

(autism severity included)

Tonge 2012

CARS: PEBM 35.86 (SD 6.14),

DBC ‐ ASA: PEBM 20.77 (SD 9.05)

CARS: PEAC 40.06 (SD 5.44) DBC ‐ ASA: PEAC 23.90 (SD 8.23)

No difference between groups

Silva 2009

ABC: 33.9 (SD 18.6)

ABC: 59.4 (SD 35.4)

Blinded teacher evaluations indicated that children treated with massage made significant classroom reduction in autism characteristics

Maladaptive behaviour

Smith 2000

CBC (withdrawal): 55.0 (SD 4.40)

CBC (social problems): 59.43

(SD 8.02)

CBC (aggression): 55.71

(SD 5.53)

CBC (withdrawal): 61.89 (SD 7.04)

CBC (social problems): 59.78

(SD 9.59)

CBC (aggression): 60.0 (SD 10.81)

No significant differences reported by teachers between groups on this domain

Roberts 2011

DBC total: Home‐based 52.9

(SD 29.3)

DBC total: Centre‐based 55.7

(SD 19.5)

There were no statistically significant differences between the 2 groups

Rickards 2007

PBCL: 13.7 (SD 8.3)

PBCL: 21.2 (SD 6.7)

 

The authors reported that there was non‐significant trend in favour of intervention

Tonge 2012

DBC – TBPS: PEBM 53.29

(SD 24.36)

DBC – TBPS: PEAC 57.61

(SD 19.72),

No difference between groups

Parents'confidence in coping with child's disability and

behaviour problems

Nefdt 2010

‐

‐

Significant differences between the intervention group and control group; parents appeared more confident during interactions with their child following the self directed learning programmes (F = 16.37 and P value = 0.001 with an effect size of 1.28)

Any cost information provided by the authors

Roberts 2011

‐

‐

The cost per child was AUD6383.  Staff hours were the same for home based and centre based

Rickards 2007

‐

‐

Teacher's salary for 1 year plus some expenses for travelling (this travelling estimate was not provided)

ABC: Autism Behaviour Checklist; BRS: Bayley Behaviour Rating Scale; CARS: Childhood Autism Rating Scale; CBC: Child Behaviour Checklist; DBC: Developmental Behaviour Checklist; DBC‐ASA: Developmental Behaviour Checklist Autism Screening Algorithm; DBC‐TBPS: Developmental Behaviour Checklist Total Behaviour Problem Score; GHQ: General Health Questionnaire; IQ: intelligence quotient; PBC: Preschool Behaviour Checklist; PEAC: Parent Education and Counselling; PEBM: parent education and behaviour management; PBCL: Preschool Behaviour Checklist; PDDBI: Pervasive Developmental Disorders Behaviour Inventory; PEP‐R DQ: Psychoeducational Profile‐Revised – Developmental Quotient; PSI: Parenting Stress Index; QRS‐F: Questionnaire on Resources and Stress (Friedrich short form); RDLS: Reynell Developmental Language Scales; SD: standard deviation; VABS: Vineland Adaptive Behaviour Scales.

Figures and Tables -
Table 5. Numerical data from studies not included in meta‐analyses
Navigate to table in Review
Comparison 1. Child communication and social development

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 Language ‐ Joint (direct or independent assessment) Show forest plot

2

64

Std. Mean Difference (IV, Random, 95% CI)

0.45 [‐0.05, 0.95]

2 Communication (reported) Show forest plot

3

228

Mean Difference (IV, Random, 95% CI)

5.31 [‐6.77, 17.39]

3 Language – Expression (direct or independent assessment) Show forest plot

3

264

Std. Mean Difference (IV, Random, 95% CI)

0.14 [‐0.16, 0.45]

4 Language – Expression (reported) Show forest plot

3

204

Mean Difference (IV, Random, 95% CI)

29.44 [‐14.99, 73.86]

5 Language – Comprehension (direct or independent assessment) Show forest plot

2

200

Std. Mean Difference (IV, Random, 95% CI)

0.29 [‐0.20, 0.78]

6 Language – Comprehension (reported) Show forest plot

3

204

Mean Difference (IV, Random, 95% CI)

36.26 [1.31, 71.20]

7 Parent‐child interaction (Shared or joint attention time) Show forest plot

3

215

Std. Mean Difference (IV, Random, 95% CI)

0.41 [0.14, 0.68]

8 Child initiations (coding of parent child interactions) Show forest plot

4

268

Std. Mean Difference (IV, Random, 95% CI)

0.38 [‐0.07, 0.82]

9 Autism severity Show forest plot

6

316

Std. Mean Difference (IV, Random, 95% CI)

‐0.30 [‐0.52, ‐0.08]

10 Adaptive behaviour Show forest plot

2

197

Mean Difference (IV, Fixed, 95% CI)

1.06 [‐2.95, 5.06]
Figures and Tables -
Comparison 1. Child communication and social development
Navigate to table in Review
Comparison 2. Parent outcomes

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 Parents' level of stress Show forest plot

2

55

Std. Mean Difference (IV, Random, 95% CI)

‐0.17 [‐0.70, 0.36]

2 Parent‐child interaction (parent synchrony) Show forest plot

3

244

Std. Mean Difference (IV, Random, 95% CI)

0.90 [0.56, 1.23]
Figures and Tables -
Comparison 2. Parent outcomes
Navigate to table in Review