Parent‐mediated intervention delivered through telehealth for children with autism spectrum disorder

Objectives

This is a protocol for a Cochrane Review (intervention). The objectives are as follows:

To assess the effects of telehealth‐based parent‐mediated interventions, for both children with ASD and their parents, compared with no treatment, wait‐list control, treatment as usual, an alternative parent‐mediated intervention not delivered by telehealth, or an alternative telehealth‐based intervention not mediated by parents.

A secondary objective is to explore the potential moderators of treatment effect.

Background

Description of the condition

Autism spectrum disorder (ASD) is a range of complex, pervasive, and heterogeneous conditions characterized by core features in two areas—deficits in social communication and restricted, repetitive behaviors—regardless of race, ethnicity, culture, or socioeconomic status (APA 2013). Although ASD is a biological disorder stemming from early abnormal brain development and altered neural connectivity, no reliable biomarkers have been identified to facilitate a diagnosis (Lord 2018). Observation of early appearing aberrant behavioral manifestations forms the basis of diagnosis, with symptoms in the two main domains (i.e. social communication and restricted, repetitive behaviors) ranged from slight to substantial impairment that can affect all daily living functions of the affected individual (e.g. language and motor development, cognitive functioning, adaptive behaviors; Masi 2017). As the most heritable of psychiatric disorders (Varghese 2017), ASD affects over 52 million individuals around the world (7.6 per 1000; Baxter 2015), and at least 1.5% of all populations in developed countries (15 per 1000; Lyall 2017), with a substantial annual economic cost expected to reach USD 461 billion by 2025 in the USA alone (Leigh 2015).

Heterogeneity in etiology, phenotype, and prognosis are hallmarks of ASD, leading to vast clinical heterogeneity and complicating the pursuit for a definitive treatment for ASD (Lord 2018). While the developmental trajectories of ASD vary significantly, with a minority of cases with previously diagnosed ASD no longer demonstrating any clear autistic impairments (Fein 2013), most individuals with ASD require lifelong support (Lord 2018). Deficits in social communication and aberrant behavioral patterns can lead to challenges and less desirable outcomes in youth and adulthood, including limited opportunity for community or social activities (Orsmond 2013), low participation in higher education (Roux 2013), high rates of unemployment (Shattuck 2018), and inability to maintain reciprocal friendships or intimate relationships (DaWalt 2019). Prognosis is likely to be predicted by childhood intelligence quotients (IQ; Fernell 2013), diagnostic severity (Gotham 2012), and early progress in language and non‐verbal skills (Pickles 2014). In addition, parent participation in early interventions and the family’s positive reactions to the diagnosis have also been shown to be critical predictors of better prognosis (Anderson 2014). There is an emerging consensus that early diagnosis and relevant resources are needed for families, schools, and communities to create an 'autism‐friendly' environment around the affected individuals (Fernell 2013).

Description of the intervention

Treatment for ASD is primarily based on education and behavioral interventions, with the use of pharmaceuticals as an important adjunct (Lord 2018). While there is wide consensus that the educational and behavioral service programs for children and youth with ASD should be based on the best scientific evidence (Steinbrenner 2020), it remains difficult to translate evidence‐based practices (EBPs) into the 'real world' (Vivanti 2018). Effective early intervention programs require individualized clinical management and skilled professional teams, which are expensive to implement, especially in areas with limited healthcare resources (Salomone 2017). In considering these barriers, parents or caregivers are expected to play a more vital role in the therapeutic process for children with ASD (Parsons 2017). Parent‐mediated intervention (PMI), where parents are trained to implement treatment with their children with ASD, has been increasingly considered as a prominent way to improve treatment access for the ASD population (Stahmer 2015).

In addition to its strength in improving treatment access for families with infants and toddlers with or at risk of ASD, PMI may also be a more favorable and naturalistic means of treatment (Nevill 2018). Indeed, most PMI research to date has been conducted with parents of young children with ASD (Wong 2015), with an emphasis on training or teaching parents to create opportunities for joint engagement and play, to avoid being too directive, and to encourage child initiations (e.g. Kasari 2014; Rogers 2014; Wetherby 2014). PMI is considered non‐intrusive for families, relatively low in cost, and adaptable for different settings and individual or group format (Lord 2018). As one of the 28 EBPs identified by the National Clearinghouse on Autism Evidence and Practice (Steinbrenner 2020), PMI has received relatively strong empirical support on its effectiveness for improved parent‐child interactions (Oono 2013), parenting efficacy (Tarver 2019), reduced child disruptive behaviors (Postorino 2017), symptom severity (Nevill 2018), and parental distress (Tarver 2019), among other things. However, barriers continue to prevent the use and accessibility of PMIs globally, including increased family financial and time burdens, a shortage of qualified clinicians, and geographic isolation (Parsons 2017).

In recent decades, the advancement of telehealth (also called ‘telemedicine’ and ‘telepractice’) has transformed the delivery of traditional healthcare services for a wide range of conditions, populations, and settings, and is becoming an effective means for increasing access to healthcare (Dorsey 2016). Telehealth includes a variety of different technologies (e.g. telephones, smartphones, video conferencing software and the Internet) that are not treatments or interventions themselves, but technologies to expand access and deliver service or education in alternate and convenient formats (Totten 2016), and when in‐person service is not feasible or desired, or both (Wijesooriya 2020). As one of the oldest technologies applied in healthcare, telehealth has a strong evidence base in remote monitoring and counseling for individuals with chronic conditions (Kamei 2013; Kotb 2015), and psychotherapy (Martin 2011). In the field of autism, the application of telehealth has also received preliminary yet promising results concerning diagnostic assessment (Nazneen 2015), psychiatric consultation and intervention (Hepburn 2016), and delivery of training to therapists, teachers, and parents (Boisvert 2010).

Compared with the general use of telehealth, PMI has a shorter history of adopting telehealth (or terms used more often in the field of education such as ‘online learning’, ‘distance learning’, and ‘remote education’) to deliver intervention for parentsh. While empirical research in this field has employed primarily single‐subject or quasi‐experimental designs (e.g. Meadan 2016; Vismara 2012; Wainer 2013), more recent attempts using randomized trials have suggested very promising results, with improvements in outcomes such as parents' knowledge and implementation of skills, child behavior and social communication capacity, and reductions of intervention cost (e.g. Ingersoll 2016; Lindgren 2016; Vismara 2018). Moreover, telehealth has been viewed as a feasible and favorable platform for training by parents living in communities with low resources (Parsons 2017).

How the intervention might work

Parents of children with ASD often suffer from a vicious circle of higher levels of distress (Hayes 2013), feelings of incompetence (Estes 2009), impaired parent‐child relationship (Woodman 2015), poor self‐perceived parenting behaviors (Ooi 2016), and a higher level of perceived behavioral problems in children (Kasari 2014). Empowering parents with the knowledge and skills to cope effectively with ASD‐related difficulties can help break this circle and result in a wide range of positive changes to the families. PMIs typically coach parents on how to interact with their children (e.g. Kasari 2014), or on how to prevent challenging behaviors (e.g. Bearss 2015a), or both. This can lead to immediate effects of improved parenting skills and parent‐child interactions, enhanced child social communication behaviors and reduced disruptive behaviors, with subsequent positive effects on parental stress, confidence, and self‐efficacy (Lord 2018). Additionally, parental involvement in early interventions compensates families’ limited access to early treatment services, which is particularly essential in low‐ or middle‐income countries and geographically isolated areas (Parsons 2017). Parents presenting as interventionists provide an alternative option of treatment, as they can support their children across different times and settings, thus facilitating skill generalization and sustainability (McConachie 2007).

However, many other studies examining the effectiveness of PMI have not yielded positive results (e.g. Carter 2011; Rogers 2012). In a meta‐analysis comparing the effect size of parent‐ and therapist‐implemented interventions, significantly better cognitive, social, and adaptive behavior outcomes were found in therapist‐implemented studies, and comparable communication outcomes were found in both types of interventions (Nahmias 2014). Given their relative lack of expertise and demanding schedules, parents may experience difficulty in reaching the high level of fidelity and intensity typically implemented by trained professionals (Stahmer 2015). In addition, for children with severe impairments who cannot handle toys or items well, training parents to avoid being directive and to encourage child initiations might not be beneficial (Lord 2018). Therefore, recent research efforts have suggested not to anticipate PMI as a replacement of intensive therapist‐implemented treatment, but to emphasize the role of PMI in improving family functioning and reducing parental distress, in addition to supporting child development (Stahmer 2015). 

The employment of telehealth for PMI delivery is rooted in a belief that telehealth has the potential to remove barriers to treatment use, leverage intervention cost, and produce positive benefits comparable to face‐to‐face training (Totten 2016). Telehealth technology can be both synchronous (e.g. real time consultation and video conferencing) and asynchronous (e.g. online programs that provide digital samples such as images or videos); either way, parents need not travel to a center for training. Training sessions can vary in frequency and intensity (for example, weekly 1.5 hour sessions for 12 weeks in Vismara 2018, or self‐paced modules completed within three weeks in Hamad 2010) and method of delivery (for example, live video conferencing with a therapist in Wacker 2013, self‐directed Web modules in Nefdt 2010, or a combination of both synchronous and asynchronous means in Ingersoll 2016). Most interventions were developed based on theories under the naturalistic developmental behavioral intervention paradigms (NDBI) that emphasized play, social interaction, child initiation, and natural consequences (for example, Parent Delivered Early Start Denver Model in Vismara 2018 and Pivotal Response Treatment in Nefdt 2010). Whilst communication technology allows parents of children with ASD to learn critical knowledge and skills regardless of time, financial and geographic limits, concerns have been raised with regard to participant acceptance, system usability, and ethical issues (Quigley 2019; Salomone 2017). Telehealth‐based research often faces an unavoidable challenge: to expand service coverage in communities with low resources and in target populations that are usually not technically savvy (Dinesen 2016). Considerable work is still needed to identify factors that promote telehealth acceptance, feasibility, and community implementations (Salomone 2017).

The exploration of potential moderators of the effects of PMIs and telehealth‐based PMIs has yielded variable conclusions. In a recent systematic review, Trembath 2019 identified 45 mediating and moderating factors that were examined across different individual PMI studies. Trembath and colleagues further categorized these factors as child and parent characteristics and contextual factors, all of which yielded mixed or non‐significant results. For example, one of their included studies (Green 2010) examined the impact of participant characteristics (e.g. autism severity, non‐verbal ability, family socioeconomic status), but none of these factors were found to have a significant impact on treatment effects. In contrast, other studies found that non‐verbal ability, communication skills, parent education and income, and family location (rural or urban) were significant moderators of treatment effect (Hardan 2015; Tonge 2014; Turner‐Brown 2019; Zhou 2018). Pickard 2016 also carried out a qualitative examination of parents’ perceptions of telehealth‐based PMI, and found that therapist support (as compared to self‐directed models), time pressures, and the variety of training materials all moderated treatment effect. Individual level data meta‐analysis is needed for the corroboration of evidence (Oono 2013).

Why it is important to do this review

Over the past 50 years, autism has gone from being regarded as a childhood psychiatric disorder largely attributed to unloving mothers, to being understood as a spectrum of neurodevelopmental disorder that is well‐researched and publicized. In the same span, researchers have made intriguing findings in genetics and neuroscience that have identified patterns of risk and may help to elucidate the contributors of heterogeneity (Lord 2018). What remains unchanged, however, are the geographic and socioeconomic disparities in autism research, policies, and service provision (Durkin 2015). The majority of people with ASD who live in low‐ or middle‐income countries are largely under‐represented in the autism literature (Elsabbagh 2012). The paucity of research targeting deprived populations has impeded the ability to apply existing tools and treatment models across social classes and cultures, and further prevented these populations from obtaining timely access to necessary services and supports (Liu 2020). Testing innovative and alternative models to expand service reach is a priority, and many experts argue that the use of telehealth in PMI programs may be one of the solutions (e.g. Ingersoll 2016).

Three previous reviews have examined the use of telehealth in PMIs for children with ASD (Boisvert 2014; Meadan 2015; Parsons 2017). Boisvert 2014 conducted a scoping review and identified only two telehealth‐based PMI studies published before October 2013, both of which adopted single‐subject research designs. Meadan 2015 selected six Internet‐based PMI programs, and among them, only one employed a randomized controlled experimental design (Jang 2012). Parsons 2017 conducted a systematic review and identified seven studies published between 2014 and 2016 that were specifically designed to target children with ASD living outside of urban areas (Parsons 2017). Among the seven studies, only one was a randomized controlled trial (RCT) (Ingersoll 2016). Taken together, the results of the identified studies indicated that telehealth was an effective platform for parent training, increasing both parents' knowledge and implementation skills. However, the lack of comparison groups prevented the calculation of overall effect sizes of telehealth‐based PMIs in the existing reviews. Also, restrictions on the research sample (i.e. the focus on remote areas in Parsons 2017) and time frame (i.e. all identified studies were published before 2016) in these reviews may have excluded possible telehealth‐based PMI studies that could have been examined. Furthermore, the small number of included studies and the lack of sufficiently powered experimental trials in the existing reviews have limited the capacity to determine what constitutes an effective telehealth‐based PMI and what family characteristics predict a positive response. 

Initial scoping searches for this review confirmed that there is an expanding literature consisting of controlled experimental studies about the application of telehealth in PMIs (e.g. Dai 2018; Ibañez 2018; Vismara 2018). It is therefore of interest to conduct a thorough systematic review and meta‐analysis to investigate and interpret the overall effects and potential moderators of telehealth‐based PMIs for children with ASD. With the rapid advancement of new technologies and policy‐relevant trends in the adoption of telehealth in public health care (Tuckson 2017), it is critical to identify, appraise, and synthesize this body of research, to ensure the evidence guides future research and practice, and informs the judicious use of resources to increase service availability for families of children with ASD.

Objectives

To assess the effects of telehealth‐based parent‐mediated interventions, for both children with ASD and their parents, compared with no treatment, wait‐list control, treatment as usual, an alternative parent‐mediated intervention not delivered by telehealth, or an alternative telehealth‐based intervention not mediated by parents.

A secondary objective is to explore the potential moderators of treatment effect.

Methods

Criteria for considering studies for this review

Types of studies

Randomized controlled trials, including cluster‐randomization and cross‐over designs. 

Types of participants

Children aged 18 years or under with a confirmed clinical diagnosis of ASD, with or without comorbid medical or psychiatric conditions and living with at least one parent or primary caregiver (i.e. the self‐identified main person regularly assisting in the care and support of the child with ASD).

The diagnosis of ASD includes ASD, autistic disorder, Asperger’s syndrome, pervasive developmental disorder‐not otherwise specified (PDD‐NOS), or atypical autism (APA 1994; APA 2013; WHO 1992). A confirmed clinical diagnoses is one made by a licensed psychologist or physician following standard procedures, with reference to the Diagnostic Statistical Manual of Mental Disorders‐Fourth Edition (DSM‐IV; APA 1994) or Fifth Edition (DSM‐5; APA 2013), or the tenth revision of the International Classification of Diseases (ICD‐10; WHO 1992).

We will exclude studies that include children with a number of different developmental disorders if the results are not presented separately for the group with ASD.

Types of interventions

Inclusion criteria

1. Parents trained or educated about strategies to improve their management of their child’s ASD‐related features, symptoms, or behaviors, or support their child’s skills, or both. Although Bearss 2015b has proposed a taxonomy that differentiates knowledge‐focused ‘parent support programs’ and skill‐focused ‘parent‐mediated interventions’, in real clinical settings many programs adopt ‘hybrid models’ that incorporate both knowledge‐focused and skill‐focused components. In the current review, in order to define a scope that corresponds to real clinical needs and practices, the term PMI includes both ‘purely skill‐focused’ models as defined by Bearss 2015b, and ‘hybrid models’ that incorporate both knowledge‐focused and skill‐focused training elements.

2. The treatment is remotely delivered to parents through a telehealth system. Both synchronous (e.g. real time video conference) and asynchronous (e.g. pre‐recorded videos, DVDs) types of telehealth will be considered, and both self‐directed (i.e. without consultation) and therapist‐assisted (i.e. with ongoing consultation) programs will be considered.

3. The control conditions include inactive comparisons (i.e. no treatment, wait‐list control, or treatment as usual), or active comparisons (i.e. an alternative PMI intervention not delivered by telehealth).

Exclusion criteria

1. We will exclude interventions focused entirely on parent knowledge or well‐being (e.g. psychotherapies or psycho‐education), or both, rather than their implementation of intervention techniques and management of ASD‐related features, symptoms, or behaviors.

2. We will exclude studies describing the use of technologies to support education and daily functioning of children with ASD (e.g. virtual reality, social robots, wearable sensors, mobile applications) that are not used for parent training.

3. We will exclude interventions delivered in a hybrid model combining telehealth‐based elements and in‐person elements.

Main comparisons

1. Telehealth‐based PMIs compared with inactive comparators (no treatment, wait‐list control, or treatment as usual)

2. Telehealth‐based PMIs compared with an alternative PMI intervention not delivered by telehealth, or an alternative telehealth‐based intervention not mediated by parents.

Types of outcome measures

The primary and secondary outcomes listed below are accompanied by examples of measures often used to evaluate treatment effects in PMI studies. None of the outcomes listed below are required as part of the eligibility criteria for including studies.

Primary outcomes

Parent outcomes

1. Parent implementation of intervention techniques, based on direct observations, including observational measures such as the Parent Delivered Early Start Denver Model Fidelity Checklist (P‐ESDM; Rogers 2010) and the Maternal Behavior Rating Scale (MBRS; Mahoney 1998), or direct observations such as coding of parent synchrony, shared attention, social responsiveness, or engagement.

2. Parent stress levels, measured with parent‐report scales such as the Parenting Stress Index (PSI; Abidin 1995).

3. Parent self‐efficacy, measured with parent‐report scales such as the Parent Sense of Competence Scale (PSOC; Gibaud‐Wallston 1978).

Child outcomes

1. Child social and communication development and skills, based on direct observations, including observational measures such as the Social and Communication domains of the Autism Diagnostic Observation Schedule (ADOS; Lord 2000), observed functional verbalizations, or behavioral observation measures developed by the study authors.

2. Child social and communication development and skills, measured by parent reports, using scales such as the Social Responsiveness Scale (SRS; Constantino 2005), or the Socialization and Communication domains of the Vineland Adaptive Behavior Scales (VABS; Sparrow 2005).

3. Adverse effects or events, such as an increase of parental stress, deterioration or worsening in symptom severity, or dropout from intervention.

Secondary outcomes

Child outcomes

1. Child symptom severity, measured with scales such as the Childhood Autism Rating Scale (CARS; Schopler 2010) or the Aberrant Behavior Checklist (ABC; Aman 1985). 

Parent outcomes

1. Parent ASD‐related knowledge, often assessed by a multiple choice quiz developed by the study authors.

2. Family quality of life, measured with parent‐report scales such as the Beach Center Family Quality of Life Scale (Summers 2005)

Process outcomes

1. Program satisfaction, measured by parent‐report scales such as the Client Satisfaction Questionnaire (Attkisson 1982) or scales developed by the study authors.

2. Program usability and acceptability, measured by scales such as the System Usability Scale (SUS; Bangor 2008), the Treatment Evaluation Inventory‐Short Form (TEI‐SF; Kelley 1989), or program usage data.

Economic outcomes 

1. Cost of intervention, provided by the study authors.

Outcomes must be measured using standardized assessments with established reliability and validity, parent‐report scales, and behavioral observations. We will collect data at pre‐treatment, immediately post‐treatment, medium‐term follow‐up (up to six months post‐treatment), and long‐term follow‐up (more than six months post‐treatment) time points. If a study has two or more time points measured within a time frame, we will choose the longest time point in that frame to allow each study to contribute only one effect estimate for a particular outcome within a specific time frame. We will document all eligible outcomes or measures in the ‘Characteristics of included studies’ table.

Search methods for identification of studies

We will consider both published and unpublished studies with no restrictions on the year of publication, language or publication type.

Electronic searches

We will search electronic databases and trials registers for relevant publications. The search strategy for MEDLINE is reported in Appendix 1, and we will adapt it for searches in other databases as appropriate.

1. Cochrane Central Register of Controlled Trials (CENTRAL, current issue) in the Cochrane Library.

2. MEDLINE Ovid (1946 onwards).

3. MEDLINE In‐Process and Other Non‐Indexed Citations Ovid (1946 onwards).

4. MEDLINE EPub Ahead of Print Ovid (current issue).

5. Embase  Ovid(1974 onwards).

6. APA PsycINFO Ovid (1806 onwards)

7. ERIC EBSCOhost (Education Resources Information Center; 1966 onwards).

8. CINAHL EBSCOhost (Cumulative Index to Nursing and Allied Health Literature; 1937 onwards).

9. Web of Science Core Collection Clarivate (Science Citation Index ‐ Expanded; Social Science Citation Index; Conference Proceedings Citation Index‐Science; Conference Proceedings Citation Index‐Social Science & Humanities; 1970 onwards).

10. Cochrane Database of Systematic Reviews (CDSR, current issue), part of the Cochrane Library.

11. Epistemonikos (www.epistemonikos.org/en/)

12. Sociological Abstracts ProQuest (1952 onwards).

13. Applied Social Sciences Index and Abstracts ProQuest (ASSIA; 1987 onwards).

14. ProQuest Dissertations & Theses A&I (1743 onwards).

15. CinicalTrials.gov (www.clinicaltrials.gov).

16. WHO International Clinical Trials Registry Platform (ICTRP; apps.who.int/trialsearch).

17. OpenGrey (http://www.opengrey.eu).

Searching other resources

In addition to searching electronic sources, we will search Google Scholar (scholar.google.com) to identify studies that are not yet included in the above databases. As records of Google Scholar are sorted by relevance, we will screen only the first 100 records to identify any additional studies. We will also examine the reference sections of all included studies, guidelines, and relevant reviews to identify additional studies. In addition, we will search the tables of contents of recent autism journals to determine if new terminology has evolved. Finally, we will contact key authors in the field directly for any unpublished or ongoing trials.  Before publication, we will conduct  a search to make sure none of our included studies  have been retracted or found to be fraudulent. 

Data collection and analysis

Selection of studies

Having removed duplicates, the first two review authors (QL and WYH) will independently screen the titles and abstracts of all citations retrieved by the search strategy. We will move abstracts that appear to meet our inclusion criteria (Criteria for considering studies for this review), and abstracts for which we need further information, through to the full‐text stage of screening. We will seek additional information from the study authors, where necessary, to clarify eligibility for inclusion. We will regard multiple reports of the same study as one. We will seek full‐text translations of studies written in other languages. We will resolve disagreements at each stage through discussion and consultation with a third review author (GC) until we reach a consensus. We will record the reasons for excluding studies, and present the results of the selection process in a PRISMA flow diagram (Moher 2009). 

Data extraction and management

Two review authors (QL and WYH) will independently extract and record data from each included report using a data collection form, which we will pilot test on at least one study included in the review prior to use. We will enter data into Excel first, and one author (QL) will transfer the extracted data into Review Manager 5.4 (Review Manager 2020). A second author (WYH) will check the entered data for accuracy against the study report. We will resolve disagreements through discussion and by consulting a third author if necessary (GC). Following the Participants, Interventions, Comparison, and Outcomes (PICO) framework, the form will tabulate data in the following categories.

1. Study characteristics: location, language, study type, funding source, and notable conflicts of interest.

2. Participant characteristics: number of participants in intervention and control groups; parent characteristics, including age, gender, ethnicity, educational level, family income; child characteristics, including age, gender, diagnostic description and severity of impairments.

3. Intervention characteristics: content, duration, intensity/dosage, recipient of intervention, delivery of intervention and modalities (type of telehealth).

4. Control conditions: inactive comparators (no treatment, wait‐list control, and treatment as usual); or active comparators (PMI intervention not delivered by telehealth, or an alternative telehealth‐based intervention not mediated by parents).

5. Outcome measures: primary and secondary outcomes and time point(s) data are collected.

6. Implementation related factors: measurement of adherence, cultural adaptations, social validity, and any reported cost data.

7. Reported key conclusions.

Assessment of risk of bias in included studies

We will use the revised Cochrane 'Risk of bias' tool for randomized trials (RoB2; Sterne 2019) to assess the risk of bias in the included studies. Two review authors (QL and WYH) will independently assess the risk of bias for each study. We will resolve disagreements through discussion and, if necessary, consultation with a third review author (YY). We will use the RoB2 Excel tool to implement the assessment of risk of bias. For each included study, we will only assess the results that contribute to the review’s ‘Summary of findings’ table (i.e. parent implementation of intervention techniques based on direct observations; parent ASD‐related knowledge; parent stress levels; parent self‐efficacy; child social and communication development and skills based on direct observations; child social and communication development and skills as measured by parent report; and symptom severity) measured at post‐intervention. Our principal interest is on the effect of assignment at baseline (i.e. the ‘intention‐to‐treat effect’) (Higgins 2020c).

For individually‐randomized, parallel group trials and cross‐over trials, we will address five domains of bias: bias arising from the randomization process; bias due to deviations from intended interventions; bias due to missing outcome data; bias in the measurement of the outcome; and bias in the selection of the reported result. For cluster‐randomized trials, we will consider an additional domain, namely the bias arising from the timing of identification and recruitment of individual participants in relation to the timing of randomization. For each domain, the RoB2 tool provides a series of ‘signaling questions’ aimed at eliciting trial information relevant to risk of bias (e.g. was the allocation sequence concealed until participants were enrolled and assigned to interventions?). Based on the answers to the signaling questions in each domain, we will generate an algorithm and a corresponding judgment about the risk of bias for each domain, which can be ‘low’, ‘high’ or ‘some concerns’ (Sterne 2019). We will support the answers to the signaling questions and our judgments about the risk of bias with written justifications.

The overall risk of bias for each result will be the least favorable assessment across the domains of bias. In addition, if we assess one result as having 'some concerns' in four domains or more, we will consider raising the overall risk of bias to 'high' (Sterne 2019). We will provide narrative descriptions in the relevant results section.

Measures of treatment effect

Dichotomous data

For dichotomous outcomes (e.g. clinical improvement versus no clinical improvement), we will calculate risk ratios (RRs) and present these with 95% confidence intervals (CI). We choose to present the RR over the odds ratio (OR) because the OR tends to be misinterpreted as the RR, which may lead to an overestimation of the intervention effect (Higgins 2020b).

Continuous data

For continuous data, when the same outcomes are measured using different measurement scales, we will calculate standardized mean differences (SMDs) with 95% CI. When the same measurement scale is used, we will calculate mean differences (MDs) with 95% CI.

If data for the same outcome are presented in some studies as dichotomous data and in other studies as continuous data, we will summarize the means and standard deviations as continuous outcomes and the binary data as dichotomous outcomes, and re‐express ORs as SMDs using the algorithm described in Section 10.6 of the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2020) for later meta‐synthesis. 

If the outcome distribution is skewed (reported as medians or interquartile ranges), we will provide a narrative description of the skewed data.

Unit of analysis issues

Cluster‐randomized trials

Studies with a cluster‐randomized design should ideally report a direct effect estimate from analyses that properly account for cluster allocations. In such cases, we will extract reported effect estimates and their standard errors (SE) for meta‐analysis using the generic inverse‐variance approach in RevMan 5.4 (Review Manager 2020). If such an appropriate analysis is not performed in identified cluster‐randomized trials, we will correct the analysis by reducing the size of each trial to its ‘effective sample size’, using the intra‐cluster correlation coefficient (ICC) reported by the studies. If the ICC is not available in the identified reports, we will use external ICC estimates obtained from similar studies (Higgins 2020a).

Cross‐over trials

Due to the developmental nature of ASD and the long‐term effects of PMIs, we assume that PMI studies adopting cross‐over designs may inevitably face the problems of carry‐over and period effects (Higgins 2020a). Therefore, if any cross‐over trials are identified, we will include only the data from the first period of the trials in this review.

Studies with multiple treatment groups

If included trials randomize participants to multiple intervention groups, we will include only the relevant comparisons in this review (i.e. telehealth‐based PMI versus no treatment, wait‐list control, treatment as usual, or an alternative PMI intervention not delivered by telehealth). If all intervention groups are relevant, we will combine the results of all the eligible treatment groups into a single treatment group and combine all relevant control groups into one control group to create a single pair‐wise comparison.  

Dealing with missing data

In accordance with Section 10.12.2 in the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2020), we will handle missing data as follows.

1. If data of interest or study details are not reported in the included studies, we will contact the authors and request them to supply any unreported data or missing information.

2. For each individual study, we will assess whether data are missing at random or not. If data are missing at random, and we are unable to make contact with the study authors to obtain the missing information (if no response is received after three emails or one week after the first email attempt), we will only use the reported information in the original reports and ignore the missing data (complete case analysis). Where data are not missing at random, we will impute the missing data with replacement values using multiple imputation (Rubin 1987), and treat these as if they were observed. We will beware of the uncertainty in the imputed values and results, typically in CIs that are too narrow. We will address this issue in the Discussion section.

3. We will describe missing data and attrition or dropout for each study in the ‘Risk of bias’ tables.

4. We will perform a sensitivity analysis to assess how sensitive the results are to reasonable changes in assumptions made (Sensitivity analysis).

5. We will address the potential impact of missing data on the findings of the review in the Discussion section.

Assessment of heterogeneity

We will assess clinical heterogeneity by comparing intervention characteristics, participants, and outcomes used across studies. We will examine methodological heterogeneity by comparing differences in methods (e.g. randomization procedures, blinding of assessors) across studies. To assess statistical heterogeneity, we will perform the Chi² test, where a low P value (e.g. lower than 0.10) indicates statistical heterogeneity of treatment effects. We will examine estimates of between‐studies variance using Tau². We will also compute the I² statistic to determine the percentage of variability that is due to heterogeneity rather than sampling error: an I² of 0% to 40% indicates that heterogeneity is unlikely to be present or is low; 30% to 60% indicates moderate heterogeneity; 50% to 90% indicates substantial heterogeneity; and 75% to 100% indicates considerable heterogeneity (Deeks 2020). In cases where only a small number of studies are synthesized for a certain construct, we will compare the difference between studies directly rather than using I², given that I² is not good at estimating heterogeneity for analysis with few studies (Deeks 2020). If heterogeneity is present, we will explore possible reasons for heterogeneity by conducting subgroup analyses, where the data permit (Subgroup analysis and investigation of heterogeneity).

Assessment of reporting biases

If 10 or more studies are included in a meta‐analysis, we will use funnel plots (i.e. estimated differences in treatment effects against their standard error) to subjectively detect whether reporting bias is present. Asymmetry could be caused by publication bias, but could also be caused by other factors such as poor methodological quality, true heterogeneity and artefactual correlations between effect size (e.g. SMDs and odds ratios) and its standard error (Page 2020). If funnel plot asymmetry is detected, we will attempt to distinguish the different possible reasons for the asymmetry, and conduct sensitivity analyses to explore different assumptions about the causes of funnel plot asymmetry (Sensitivity analysis).

Data synthesis

Where two or more studies are eligible for inclusion and they are considered homogenous regarding intervention focus, participant characteristics, intervention delivery and outcome measures used, we will conduct a meta‐analysis. Due to the heterogeneous nature of ASD characteristics and PMI programs, we will use the random‐effects model. For outcomes presented either as dichotomous data or continuous data, or a combination of dichotomous and continuous data where ORs will be re‐expressed as SMDs, we will use the inverse‐variance approach in RevMan 5.4 (Review Manager 2020). We choose the inverse‐variance approach over the Mantel‐Haenszel method for dichotomous data because the latter is preferable only in fixed‐effect meta‐analyses; in random‐effects model the two methods give similar estimates (Deeks 2020).

If a study only presents change‐from‐baseline data or post‐intervention data in the published report, we will contact the study authors to obtain data for each time point. If only one type of data is available upon request, we will combine change scores and post‐intervention values in meta‐analysis using unstandardized MD method, and synthesize the two types of data separately when the effect measure is SMD (Deeks 2020). If a meta‐analysis is not appropriate, due to an insufficient number of eligible studies or considerable heterogeneity across studies, we will provide a narrative description of the study characteristics and results. 

Subgroup analysis and investigation of heterogeneity

We will explore possible causes of heterogeneity using subgroup analysis, if there are adequate studies to justify subgroup analyses or meta‐regressions (at least 10 studies in meta‐analysis per outcome). Exploratory subgroup analysis will include:

1. therapist involvement (self‐directed models versus therapist‐assisted or hybrid models; Pickard 2016);

2. intervention dosage (less than 20 hours versus 20 hours or more; Nevill 2018); and

3. pre‐treatment participant characteristics (i.e. symptom severity, parent educational level, and family socioeconomic status; Trembath 2019).

Sensitivity analysis

We will perform sensitivity analyses to test the impact of methodological decisions made within the systematic review process on the overall estimate of effects. We will consider the following aspects.

1. Risk of bias: we will remove studies with an overall high risk of bias for each outcome and re‐analyze the remaining studies to determine whether the results are affected by these factors.

2. Method of analysis: we will compare fixed‐effect and random‐effects estimates of the intervention effect to determine whether the results are affected by method of analysis.

3. Imputation of missing data: we will remove studies where we impute missing data with replacement values.

4. Cluster‐randomized trials: we will compare different values of the intra‐cluster correlation coefficient (ICC) when trial analyses have not been adjusted for clustering.

5. Funnel plot asymmetry: we will compare the fixed‐effect and random‐effects estimates of the intervention effect to assess the influence of ‘small‐study effects’ (Page 2020).

Summary of findings and assessment of the certainty of the evidence

We will create a 'Summary of findings' table using the GRADEpro GDT software for each of the main comparisons in this review:

1. telehealth‐based PMIs compared with inactive comparators (no treatment, wait‐list control, or treatment as usual); and

2. telehealth‐based PMIs compared with an alternative PMI intervention not delivered by telehealth, or an alternative telehealth‐based intervention not mediated by parents.

We will include the following outcomes, measured immediately post‐treatment, in the 'Summary of findings' tables.

1. Parent implementation of intervention techniques based on direct observations.

2. Parent stress levels.

3. Parent self‐efficacy.

4. Child social and communication development and skills based on direct observations.

5. Child social and communication development and skills as measured by parent report.

6. Adverse events.

We will assess the overall certainty of the evidence for each outcome following the methods and recommendations described in Chapter 14 of the Cochrane Handbook for Systematic Reviews (Schünemann 2020), using the GRADE approach (Schünemann 2013). We will consider the following factors: the overall risk of bias related to study design and implementation; inconsistency of results with unexplained heterogeneity; indirectness of evidence related to the target population, intervention, and outcomes of interest; imprecision of results such as small sample size and wide Cls; and publication bias. Two review authors (QL and YY) will independently rate the certainty of the evidence for each outcome as one of four levels: high, moderate, low or very low. We will resolve disagreements through discussion and, if necessary, consultation with a third author. We will include supplemental tables that present the evidence in more detail, reporting study characteristics, summary statistics, and risk of bias with consensus decisions for the signaling questions and summary of judgments. 

History

Protocol first published: Issue 3, 2021

Appendices

Appendix 1. Search strategy for Ovid MEDLINE

1     exp Child Development Disorders, Pervasive/
2     Developmental Disabilities/ 
3     Neurodevelopmental Disorders/ 
4     autis$.tw,kf. 
5     Asperger$.tw,kf. 
6     pervasive development$ disorder$.tw,kf. 
7     (PDD or PDDs or PDD‐NOS or ASD or ASDs).tw,kf.
8     or/1‐7 
9     exp Parents/ 
10     Parenting/ 
11     exp Caregivers/
12     exp Family/ 
13     exp parent‐child relations/
14     Patient Education as Topic/ 
15     (parent$ or family or families or mother$ or father$ or maternal$ or paternal$).tw,kf. 
16     (carer$ or care‐giver$ or care giver$ or caregiver$).tw,kf. 
17     (at home or (in adj3 home) or home‐based or home based or homebased).tw,kf. 
18     or/9‐17 
19     exp Telemedicine/ 
20     exp Telecommunications/ 
21     Communications Media/ 
22     Social Media/ 
23     Therapy, Computer‐Assisted/ 
24     Video Games/
25     (app or apps or electronic device$ or mobile device$ or mobile phone$ or telephone$ or smart phone$ or
smartphone$).tw,kf. 
26     ((digital adj5 clinic$) or (digital adj5 consult$) or (digital adj5 health) or (digital adj5 intervention$) or
(digital adj5 servic$) or (digital$ adj5 therap$) or (digital adj5 training)).tw,kf. (7439)
27     (eclinic$ or e‐clinic$ or econsult$ or e‐consult$ or e‐health or ehealth or e‐medicine or emedicine or epsychol$
or e‐psycholog$ or epsychiat$ or e‐psychiat$ or etherap$ or e‐therap$ or electronic therap$).tw,kf. (6477)
28     (eLearn$ or e‐Learn$ or distance learning or distance education or (remote adj3 consult$) or remote learning or
remote education or DVD$ or video$ or videoconferenc$ or video‐conferenc$ or Zoom or multimedia or social media$ or
online$ or internet$ or web or web‐based or computer or computer‐based or tech$ or self‐directed).tw,kf. (2219642)
29     (telecare or tele‐care or teleclinic$ or tele‐clinic$ or teleconferenc$ or tele‐conferenc$ or teleconsult$ or
tele‐consult$ or telehealth or tele‐health or teleintervention or tele‐intervention or telemedicine or tele‐medicine or
telepractice or tele‐practice or telepsycholog$ or tele‐psycholog$ or tele‐psychiat$ or telepsychiat$ or telerehab$ or
tele‐rehab$ or teletherap$ or tele‐therap$ or teletreat$ or tele‐treat$).tw,kf. (19544)
30     or/19‐29 
31     randomized controlled trial.pt. 
32     controlled clinical trial.pt.
33     randomi#ed.ab.
34     placebo$.ab. 
35     drug therapy.fs. 
36     randomly.ab. 
37     trial.ab.
38     groups.ab. 
39     or/31‐38 
40     exp animals/ not humans.sh. 
41     39 not 40
42     8 and 18 and 30 and 41 
43     ("Early Start Denver Model" or "Pivotal Response Treatment" or "Naturalistic Developmental Behav$ Intervention"
or "Naturalistic Intervention" or "Enhanced Milieu Teaching" or "Incidental Teaching" or "Joint Attention Symbolic Play
Engagement and Regulation" or JASPER or "Improving Parents As Communication Teachers" or "Project ImPACT" or "Reciprocal
Imitation Training" or "Early Social Interaction" or "Applied Behav$ Analysis" or ABA or "Discrete Trial Training" or
"Functional Behavioral Assessment" or "Functional Communication Training" or "Social Communications Emotional Regulation
Transactional Support" or SCERTS or "Picture Exchange Communication System" or PECS or "Social Narratives" or "Social
Stories" or "video model*ing").tw,kf. 
44     8 and 30 and 41 and 43
45     42 or 44 

Contributions of authors

• Qing Liu (QL) has overall responsibility for the review.

• QL developed a draft of the present protocol and search strategy; Wu‐Ying Hsieh (WYH), Gregory A Cheatham (GC), and Yue Yin (YY) all revised the draft and agreed with its submission. 

• QL and WYH will conduct the literature searches, screen the results for eligibility, extract data independently, enter data into a data extraction form, and assess each study for risk of bias with help from GC and YY in all difficult cases.

• QL and YY will carry out the meta‐analysis and interpret the analysis.

• All four authors will contribute to the write‐up of the review.

Sources of support

Internal sources

• Faculty of Education, University of Hong Kong, Hong Kong

  Postgraduate Scholarship (Qing Liu)

• Department of Special Education, University of Northern Iowa, USA

  Salary (Wu‐Ying Hsieh)

• Department of Special Education, University of Kansas, USA

  Salary (Gregory A Cheatham)

• Department of Educational Psychology, University of Illinois at Chicago, USA

  Salary (Yue Yin)

External sources

• None, Other

Declarations of interest

Qing Liu ‐ none known.

Wu‐Ying Hsieh ‐ none known.

Gregory A Cheatham ‐ none known.

Yue Yin ‐ none known.

New