A Systematic Review and Meta-analysis of Interventions to Improve Play Skills in Children with Autism Spectrum Disorder

Children with autism spectrum disorders (ASD) experience difficulty with play, and a number of different interventions have been developed and evaluated to address this deficit. This systematic review of randomized controlled trials identified 19 studies reporting on play-based interventions for children with ASD aged 2–12 years. The components of each study, including elements of the interventions and methodological quality, were examined. A meta-analysis was completed for 11 studies, and a small but significant treatment effect was identified (Hedges’ g = 0.439). The current review supports future development of interventions with a focus on the child with ASD across social environments. Outcome measures and comprehensive reporting of intervention components are important considerations in future intervention development and testing. Significance for clinicians and future research is discussed. PROSPERO registration number: RD42015026263.


Introduction
Children with autism spectrum disorders (ASD) often experience difficulties with play and forming and maintaining peer relationships. Research has demonstrated that these social difficulties persist into adolescence and young adulthood (Schall and McDonough 2010). This review will focus on interventions that target play in children with ASD. For the purpose of this review, play is defined as a transaction between the individual and the environment which includes "…the presence of three elements: intrinsic motivation, internal control, and the freedom to suspend reality" (Skard and Bundy 2008, p. 71). Additionally, it needs to be apparent to both the players and observers that this transaction is playful, by the cues the players give and read- Bundy (2012) identifies this as being in the play frame. This definition of play is both contemporary and comprehensive, is appropriate across different ages and stages of development, and has been used by many observation and intervention studies in the past. Play is an important aspect of childhood, and there are many benefits in promoting play. Play is the context in which most childhood friendships are formed, from early preschool years through to adolescence (Bundy 2012). Play is essential to childhood development and provides an ideal opportunity and context for parent and peer engagement. Play, as an independent occupation, not just a means to promote other skills or development, is a legitimate and necessary outcome because it is a critical element of the human experience (Parham et al. 1996). Despite this importance, play may have diminished social validity or priority (Foster and Mash 1999). Certainly, time available for play has been significantly reduced for some children as other areas of development, such as academic outcomes, are increasingly valued (Ginsburg 2007).
The three elements are not fixed, but rather move depending on the child's experience, and can tilt the transaction away from non-play to play. If children in the play frame have reduced internal control or intrinsic motivation, then play can tilt back to work (Bundy 2012). As play is defined as an intrinsically motivated transaction, simply providing toys in an engaging environment will not necessarily guarantee a child will play (Bundy 2011). Similarly, it is not enough for the study to say it was a play intervention if the transaction was highly structured or if the child was required to follow a set play routine. Furthermore, the complexity and constantly changing nature of play make measuring play ability difficult for educators, clinicians, and researchers (Brooke 2004). Frequently, measures are of readily observable social skills or children's behavior from the perspective of a parent or teacher, rather than a child's play ability (McAloney and Stagnitti 2009). Observation of unstructured play in a natural play context would support an accurate and authentic assessment (Ray-Kaeser and Lynch 2017). This assessment would be reliant on the assessor's definition and reporting of play skills to assure valid and reliable results and comparisons across individuals, context, and studies (Ray-Kaeser and Lynch 2017).
Children with ASD have difficulty with components of play, specifically, turn taking, changing activities away from preferred interests, reduced symbolic quality, and relinquishing control of preferred play activities (MacDonald et al. 2009). For children with ASD, improvements in play skills lead to increased positive social interactions, as well as decreased inappropriate behaviors (Jung and Sainato 2013). Specifically, children with ASD with average cognitive functioning have been reported to have difficulties in social initiation and in social-emotional understanding (Sigman et al. 1999). It is this reduced social understanding, rather than social disinterest or insensitivity, that is the primary deficit for social play (Sigman et al. 1999). As a result of their social play challenges and subsequent lack of opportunity, these children can be caught in a cycle of social isolation. Given that children with ASD who do not acquire age appropriate social skills may lack opportunities for positive peer interactions, explicit training in social play with peers is a necessary intervention (Bauminger 2002;Jordan 2003;Jung and Sainato 2013).
Several different approaches to interventions have been developed to address impaired social interactions and play in children with ASD. These different approaches include coaching the child with ASD, identifying and addressing individual play skills and interests, and developing supportive relationships and environments. Peer-focused interventions, including integrated play groups, peer buddy systems, and group interventions, represent the largest type of social play intervention for children with ASD (Bass and Mulick 2007). The inclusion of peers in the intervention helps support generalization of skills to other environments and creates a more authentic social environment for children with ASD to develop their social play skills (Chan et al. 2009). The use of typically developing peers in interventions is further supported as friendships are associated with prosocial behaviors and act as a protective factor against rejection and bullying, especially in the preschool and school environments (Chang et al. 2015). Although promoting play has been largely neglected in school contexts due to the focus by teachers on academic outcomes, some interventions focus on upskilling teachers to be able to provide intervention to the child with ASD and to create a supportive social environment (Kossyvaki and Papoudi 2016). Parents are also frequently a focus of interventions, especially with younger children (McConachie and Diggle 2007), as the social behavior of a child with ASD has been shown to be significantly enhanced by the interaction style of the parent or caregiver adapting to the child's play level (Freeman and Kasari 2013;Meirsschaut et al. 2010).
To date, no researchers have conducted systematic reviews of play-based interventions for children with ASD tested using randomized controlled trials (RCTs). Previous systematic reviews have investigated social developmental outcomes in children with ASD, including pragmatic language interventions (Parsons et al. 2017), after school programs for personal and social skills (Durlak et al. 2010), and early behavior interventions (Warren et al. 2011) with outcomes that included improvements in cognitive performance, language skills, and adaptive behavior skills. A recent systematic review of RCTs for preschool children with ASD included interventions addressing behavioral, communication-focused, and developmental outcomes of ASD general symptoms, but not interventions addressing play outcomes (Tachibana et al. 2017). A review of 13 play-based intervention studies for children with ASD identified improvements when the intervention built upon the child's existing play repertoire (Luckett et al. 2007). However, effectiveness of these interventions could not be determined as the majority of the 13 studies used single case study designs (Luckett et al. 2007). Since the Luckett et al. study was published (2007), more than a decade ago, a number of researchers have published RCTs of play interventions for children with ASD (Corbett et al. 2016, b;Kasari et al. 2014, b). To date, no systemic reviews have been conducted of play-based interventions for children with ASD that have been investigated using RCTs.

Objectives
This systematic review focuses on the efficacy of play-based interventions to address the play skills of children with ASD. This systematic review aimed to summarize key characteristics of a range of play-based interventions for children with ASD and assess the quality of published RCTs. This metaanalysis addressed the following research questions: (1) are play interventions effective in improving play outcomes when compared to a non-play intervention or treatment as usual control group? and (2) do the following intervention characteristics mediate intervention effects: (a) focus of intervention (i.e., child, parent, peer, teacher or combination), (b) intervention setting, and (c) group or individual therapy?

Protocol and Registration
The methodology and reporting of this systematic review were based on the PRISMA and PRISMA-P statement (Moher et al. 2009;Shamseer et al. 2015), and the review was registered with PROSPERO (registration number RD42015026263; Booth 2013).

Eligibility Criteria and Study Selection
Studies were included if they met four inclusion criteria: (1) participants must include children who have a diagnosis of ASD according to the DSM-III-R, DSM-IV, or DSM-5 criteria; (2) study designs were RCTs; (3) the interventions included play as per the definition adopted in this study; and (4) treatment outcomes were assessed using play measures. Multimodal intervention programs in which the play-based intervention was part of a variety of social or behavioral components were also included. These criteria were selected to identify play-based intervention studies for children with ASD that are classified as level II on the National Health and Medical Research Council (NHMRC) Hierarchy of Evidence (NHMRC 2011). The Australian NHMRC developed the NHMRC Hierarchy of Evidence to rank and evaluate the evidence of healthcare interventions. According to the NHMRC Hierarchy of Evidence, level II studies are well-designed RCTs (NHMRC 2011).

Information Sources and Search
Studies were identified through the following two-step procedure. First, an electronic database search was conducted using PubMed, Embase, PsychINFO, CINAHL, and ERIC. These databases are where social interventions are most likely to be found. Two categories of subject headings were used in combination: (1) disorder (autism spectrum disorder; ASD) and (2) randomized controlled trials. Free text searches were also conducted for all four databases on September 4, 2017. Both subject headings and free text terms with limitations are described in Table 1. Secondly, identified studies were then  Embase: ((autism/OR "pervasive developmental disorder not otherwise specified"/OR Rett syndrome/OR childhood disintegrative disorder/) AND (randomization/or randomized controlled trial/OR "randomized controlled trial (topic)"/OR controlled clinical trial/)) OR((autism OR autistic OR ASD OR PDD OR PDD-NOS OR pervasive OR Asperger OR Rett OR (childhood AND disintegrative AND disorder*)) AND (RCT OR (Randomized AND Controlled AND Trial) OR (Randomised AND Controlled AND Trial) OR (Randomized AND Clinical AND Trial) OR (Randomised AND Clinical AND Trial) OR (Controlled AND Clinical AND Trial)) limit to yr="2017-Current") Eric ((DE "Autism" OR DE "Pervasive Developmental Disorders" OR DE "Asperger Syndrome") AND (RCT OR (Randomized AND Controlled AND Trial) OR (Randomised AND Controlled AND Trial) OR (Randomized AND Clinical AND Trial) OR (Randomised AND Clinical AND Trial) OR (Controlled AND Clinical AND Trial))) OR ((autism OR autistic OR ASD OR PDD OR PDD-NOS OR pervasive OR Asperger OR Rett OR (childhood AND disintegrative AND disorder*)) AND ( searched for inclusion of play (see Fig. 1). Gray literature was searched using Google Scholar for disorder, RCT, and play.

Synthesis of Results and Methodological Quality
Data across all studies were extracted independently by the first author using data extraction tables. Intervention characteristics were extracted for the following: (1) focus of the intervention and play skills targeted; (2) interventionists and procedure described in the study; and (3) setting, mode of delivery, and duration. Data on study characteristics were then extrapolated and synthesized into several categories: (1) group design and participant group numbers, (2) play as primary or secondary focus of study, (3) age range (means and standard deviations), (4) inclusion and exclusion criteria for participants, and (5) the play outcome measure used and results of the treatment. The QualSyst critical appraisal tool was used to assess the methodological quality of the included studies (Kmet et al. 2004). The 14-item checklist has a three-point ordinal scoring system (yes = 2, partial = 1, no = 0) that provides a systematic, reproducible, and quantitative means of assessing the quality of research. The total QualSyst score can be converted to a percentage score, with a QualSyst score of ≥ 80% considered strong quality, a score of 60-79% considered good quality, a score of 50-59% considered adequate quality, and a score of < 50% was considered to have poor methodological quality. All included studies were reviewed by two assessors and interrater reliability was established for ratings.

Meta-analysis
A meta-analysis and overall treatment effects were calculated for play-based interventions on pre-post outcome measures. Between-group analyses were also conducted to compare post-intervention scores with control groups that included another intervention or treatment as usual comparator group. Studies that included no treatment or delayed control group were removed from between-group analyses (Corbett et al. 2016, b;Frankel et al. 2010, b;Kasari et al. 2010, b). Subgroup analyses were conducted to compare the effect as a function of intervention characteristics: (1) groups or individual, (2) focus of intervention (i.e., child, parent, peer, or teacher), and (3) setting (i.e., clinic, home, or school setting).
A meta-regression analysis was conducted to determine whether focus of intervention, setting, or group or individual therapy mediated intervention effects. The study sample size (eight studies) allowed for multivariate analysis involving up to two covariates without compromising power (Borenstein et al. 2011), so one model addressed the interaction between group vs individual and setting and the other addressed the interaction between the focus of the intervention vs setting.
To compare effect sizes, pre-and post-intervention means, standard deviations, and sample sizes were extracted. If the data required for meta-analysis calculations were not reported, attempts were made to contact authors to request the data. When multiple outcome measures of play were reported for one intervention, the measure that evaluated the highest level of play skills was extracted for analysis (e.g., symbolic play types were selected over functional play types in a structured play assessment).
Extracted means, standard deviations, and sample sizes for pre-and post-intervention measures were entered into comprehensive meta-analysis, version 3.3.070 (Borenstein et al. 2005). A random effects model was used to generate effect size. The Hedges' g formula for standardized mean difference with a confidence interval of 95% was used to report effect size. Using Cohen's d convention for interpretation, an effect size of < 0.2 reflects negligible difference, between 0.2 and 0.49 is small, between 0.5 and 0.79 is moderate, and > 0.8 is large (Cohen 1988).
Given that studies that report large and significant treatment effects are more likely to be selected for publication (Borenstein et al. 2005), it is possible that some low-effect or non-significant interventions are missing from the metaanalysis. The presence of publication bias was assessed using classic fail-safe N. The test calculates the number of additional studies that, if included in the analysis, would nullify the measured effect (N). If N is large, it can be considered unlikely that there would be so many unpublished low-effect studies and it can be assumed that the meta-analysis is not compromised by publication bias (Borenstein et al. 2005).

Study Selection
A total of 327 papers were identified and screened through the subject heading and free text searches (see Fig. 1). The first author assessed all 327 abstracts for inclusion, and the fourth author assessed a random selection of 40% for interrater reliability; weighted Kappa 0.88 (95% CI [0.7, 61.00]). A total of 82 full text articles were accessed to determine if the studies met the inclusion criteria. Specifically, further information was needed regarding the description of the intervention and the outcome measures to determine if studies met the inclusion criteria. Of these, 63 studies were excluded for one or more than one of the following reasons: 10 did not include children aged 2 to 12 years of age; 3 did not include participants with ASD; 26 were not a RCT study design; 26 did not include a play-based intervention as defined by our study; and 46 did not have an outcome measure for play (see Table 2).
A total of 19 studies were selected for this systematic review based on the inclusion criteria. All the selected studies included participants aged 2-12 years with a diagnosis of ASD, used an RCT study design, investigated a play-based intervention, and reported on play outcomes that aligned with the definition of play adopted for this review.

Study Characteristics
Participants The 19 studies that met the eligibility criteria included a total of 1149 participants aged between 2 and 12 years. Of these, 11 studies included only preschool-aged children (2 to 5 years of age) involving 670 participants and nine studies included only primary school-aged children (5 to 12 years of age) with a total of 479 participants. Treatment group sample size ranged from 4 to 76 participants.  T., et al. (2005). Outcome at 7 years of children diagnosed with autism at age 2: predictive validity of assessments conducted at 2 and 3 years of age and pattern of symptom change over time. J Child Psychol Psychiatry 46(5): 500-513.   (9)    Intervention A detailed description of each intervention is provided in Table 3. Interventions focused on the child with ASD, a parent or caregiver, teacher, or typically developing peers of the child with ASD. Ten interventions occurred in the preschool or school setting, one in the community, five in the clinic, two in the home, and one with a combination of both clinic and home sessions.
Comparator Group All participants included in control groups had a diagnosis of ASD. Across the 19 studies, there were three different types of comparator group: wait-list control group, non-play-based intervention control group, and an alternative play-based intervention control group. Seven studies assigned control participants to wait-list control groups who served as a no-treatment comparison during the intervention phase of the project then went on to receive the intervention at a later stage. Control participants in four studies attended intervention for the same duration as the intervention group but participated in activities that did not meet the definition for a play-based intervention. Control groups in six studies were assigned to an alternative play-based treatment. A further three studies included both an alternative play-based intervention and a wait-list comparator group.
Outcome Measures All outcome measures reported on play outcomes that matched the definition of play used in this study. Of the studies included, one used a parent-report questionnaire and 18 used observations of the child's behavior, 13 of which used a validated outcome measure with published psychometric properties. Fifteen studies showed significant improvements in treatment outcomes between groups for their selected play outcome measure; four did not identify any significant difference between the groups. Further details on characteristics of included studies are reported in Table 4.

Meta-Analysis: Synthesis of Results
Eleven of the 19 studies eligible for the systematic review were included in the meta-analysis (see Fig. 2). The remaining eight studies could not be included in the meta-analysis as they did not contain data required for calculations. One study reported individual scores. We contacted the remaining seven authors to collect the required data needed for the meta-analysis. Six authors did not respond, and one author no longer had access to the database. Effect sizes ranged from 0.033 to 1.898 in the pre-post intervention within-group analysis, as shown in Fig. 2. The overall intervention effect was small but significant (z(11) = 3.744, p < 0.001, Hedges' g = 0.439, 95% CI [0.209, 0.669]). The within group heterogeneity was not significant (Q(11 = 17.210, p = 0.070), and 41.9% of true variability (I 2 ) could be explained by individual study characteristics.
A small but significant between-group total effect size favored play-based interventions for children with ASD (z (8)  Following the subgroup analysis of intervention characteristics, a meta-regression analysis was performed on eight studies to further explain variability of the results (Chang et al. 2016, b;Goods et al. 2013, b;Kasari et al. 2006Kasari et al. , b, 2012Kasari et al. , b, 2014Kasari et al. , b, 2015Poslawsky et al. 2015, b;Quirmbach et al. 2009, b). The analysis of intervention characteristics indicated that intervention setting and group vs individual were not significant mediators of intervention effects (see Table 5). However, focus of the intervention (i.e., child, parent, peer or teacher) was found to be a significant mediator of play outcomes (Q(3) = 8.52, p = 0.036). Table 4 contains a description of the methodological quality and QualSyst ratings of the included studies. Two studies had adequate quality using the QualSyst checklist, and three studies had good quality. The remaining 14 studies had strong quality. Interrater agreement for overall scores of methodological quality of included studies was Kappa 0.884 (95% CI [0.755, 1.000]). Lego therapy A typical Lego therapy project would aim to build a Lego set with a social division of labor. In a group of 3 (which could be comprised of children with autism, peers and/or adults), one person is the "engineer," one the "supplier," and the other the "builder." Individuals communicate and follow social rules to complete the Lego build. The therapist's role was to highlight the presence of a social problem, and help children come up with their own solutions. Children started off building quick and simple models in pairs of threes with constant adult supervision and once proficient in a small group, they moved on to build more complex models over a few sessions. Eventually, children were

Methodological Quality
• Clinic • Group • Children attended therapy for 1 h per week for 18 weeks. Taking into account the holidays, the total duration of the study was 5.5 months  Collective establishment of a daily schedule to encourage cohesiveness among group. Activities included conversational exercises, structured games, free play, improvised storytelling, and music. Peers encouraged to lead their own groups with adult supervision as needed. Typically developing peers viewed as positive role models (2-3 classmates to each child with ASD) selected friendship survey results and teacher nominations.
The social group targeted peer engagement and acceptance using shared interests to provide the context for interactions. Activities were classroom and playground based. The group leader facilitated play as needed and faded as soon as the children played independently.   joint attention (4 sessions) into teachers' everyday classroom routines and activities. Includes an individualized approach where teachers could choose to implement activities for the whole class, in small groups, and/or in a one-on-one setting.
• Preschool classroom • Group • Weekly session including 10-15 min observation by the interventionist in the classroom followed by 1 h training session at a convenient time   Caregivers were asked to engage in free play with their child with autism as they normally would at home using a standard set of toys. Videos were coded for the percentage of time in engagement states. Child's activity was segmented into unengaged/other engagement, object-engagement, or joint engagement.
Child's play behaviors during the caregiver-child interaction were coded for types of functional and symbolic play acts. Functional play type refers to the total number of novel, child-initiated functional play acts. Symbolic play type refers to the total number of different novel, child-initiated symbolic play acts.
Children in the IT group engaged in significantly less object-focused play (F(3,34) = 4.45, p < 0.01) and significantly more joint-engagement (F(3,34) = 3.21,p < 0.05) compared to children in the WL group. There was no significant difference between groups for the category for unengaged/other-engagement.Children in the IT group also displayed significantly more types of functional play acts compared to the WL group. (F(3,34)    Child play behavior was videotaped (15 min free-play session) at the university hospital and at home by the researchers.
The children were provided with a standardized set of toys.
The parent was instructed to passively monitor while the child was playing. When the child was seeking contact or interaction, the parent was allowed to respond in a natural way. Video segments were coded by trained students for toy-preference and level of play category; (a) No significant intervention effect was found on children's play behavior. Children in both groups showed the same levels of play and variation in play.

Risk of Bias within Studies
The fail-safe N calculated during meta-analysis is 67, indicating a low-risk of publication bias. This means that we would need to locate and include 67 "null" studies for the combined 2-tailed p value to exceed 0.050.

Discussion
The aim of this study was to review and analyze the evidence for interventions to improve social play skills in children with ASD. A systematic review and meta-analysis of RCT studies were completed using the PRISMA and PRISMA-P statement as guides (Moher et al. 2009;Shamseer et al. 2015). The present study included 19 RCTs with a total of 1149 participants investigating the effectiveness of interventions to improve social play in children aged 2 to 12 years with ASD. When comparing individual child vs group interventions, the meta-analysis of 11 of these studies identified a small but significant effect size in favor of interventions focused on the individual child, as compared with group interventions.
In terms of the focus of the intervention, the meta-analysis demonstrated significantly better outcomes if the focus of the intervention was the child with ASD, as opposed to parents, peers, or teachers. The meta-analysis in this review showed that it is not one intervention characteristic, but the combination of different intervention components that lead to the development of improved play skills. This systematic review allows clinicians to identify combinations of intervention components that may be effective to use with children with ASD to improve play outcomes and provide recommendations for future research. However, the definition of play and how it is measured are inconsistent across different studies. This inconsistency of definition and reporting is a challenge for clinicians when attempting to identify effective play interventions for children with ASD. Similarly, further investigations require consistent understanding and clear reporting of what play is to allow researchers to develop and test the multimodal and active ingredients in effective play interventions. The findings of the meta-analysis show a small effect size which indicates that play interventions are feasible and achievable in clinical practice; however, there is a continued need to add to the evidence for play basedinterventions to further strengthen them.

Intervention Approaches
The most commonly used approaches to improve play skills across studies included: twelve studies created supportive environments and relationships by upskilling peers, parents or teachers; ten studies used coaching the child with ASD; ten studies identified and developed individualized play skills and  (Field et al. 2001, b); and one study used social stories (Quirmbach et al. 2009, b). It is difficult to identify which approaches are essential in improving play skills. One study that demonstrated significant large treatment effects utilized both coaching of the child with ASD and identifying and developing individual play skills (Kasari et al. 2006, b). The researchers did this by utilizing specific techniques using naturally occurring opportunities to prompt a particular treatment goal, such as imitating child's actions on toys and using the child's activity interests to develop play routines. Of the three studies that demonstrated significant moderate treatment effects, two included both supportive environment and relationships and development of individual play skills (Kasari et al. 2014(Kasari et al. , b, 2015. Both approaches in these studies used specific techniques to create opportunities of establishing jointly engaged play routines. The third study with moderate treatment effects included the approaches of supportive environment and relationships and coaching of the child with ASD (Corbett et al. 2016, b). Techniques included the use of video modeling and peer mediators (Corbett et al. 2016, b). Creating supportive relationships through the upskilling of parents, teachers, and peers in interventions may also provide support for generalization of play skills across environments and with other people. These relationships are frequently responsible for creating the social environment for interaction and transaction for the child with ASD. Parent, peer, and teacher mediated interventions show promise and require further development and investigation.

Intervention Dosage
Intervention dosage refers to the quantity of treatment provided and can be reported as total hours or over a set period of time, such as 1-h session per week (Linstead et al. 2017). Ten of the 19 studies reviewed reported either one or two sessions per week and four of the five effective interventions with the  . 2 Within-group pre-post intervention meta-analysis largest effect sizes, utilized either daily or twice weekly sessions over multiple weeks (six to 12 weeks; Corbett et al. 2016, b;Kasari et al. 2006Kasari et al. , b, 2014Kasari et al. , b, 2015. Multiple opportunities over time are needed to allow for practice of social play skills from joint engagement to initiating play to joining in with peers who are already playing. This is similar to findings from the 2005 review of play therapy that identified the efficacy of treatment delivered by a therapist increases with the number of sessions (up to a range of between 30 and 35 sessions; Bratton et al. 2005). The session duration for the majority of interventions in this review was between 30 min and an hour. Play interventions in this review were less time intensive when compared to weekly social skills training interventions. Social skills training intervention session duration ranged from 1 to 3 h across eight studies, and in another review focusing only on group interventions, session duration ranged from 1 to 1.5 h across five RCT studies (Rao et al. 2008;Reichow et al. 2013). This difference in time may be reflective of the age of participants in the play interventions (ranging from 2 to 12 years). Using shorter session for younger participants is developmentally more appropriate to support engagement and learning, compared with the older participants in the social skills interventions (ranging from 6 to 18 years). Regardless, the play intervention session duration range in this review appears to be feasible.
In considering what the optimal dosage may be, the current review identified that three of the five interventions with large effect sizes involved sessions of between 30 min and an hour with multiple sessions per week and a total number of intervention hours ranging from 10 to 15 h (Kasari et al., 2006(Kasari et al., , b, 2014(Kasari et al., , b, 2015. Kasari et al. (2006, b) compared daily 30 min sessions in a preschool setting (focusing on symbolic play, as compared to a joint attention intervention of the same duration and a no treatment control group), whereas Kasari et al. (2015, b) compared a twice weekly 30-min play session with a weekly 60-min parent only psychoeducational intervention. Kasari et al. (2014, b) compared a twice-weekly 60-min play session with the child and parent in the home with a weekly 2h parent only education group program. Authors of a 2017 review of behavioral interventions for children with ASD in a clinical setting found a linear relationship between treatment intensity and treatment outcomes (Linstead et al., 2017). Linstead et al. (2017) examined results of 726 children with a mean age of 7.1 years and found the intensity of the intervention accounted for 35% of the variance in treatment outcomes. Multiple sessions over time allow for complex skills to be developed, reviewed, and assimilated, supporting possible generalization of play skills to other environments and with other social partners. Importantly, as social play interactions become more complex across early and middle childhood, intervention components need to change and meet the demands of the increasingly complex contexts and skills required for successful engagement (Del Giudice, 2014).

Setting
The current review found that the play setting did not seem to favor the effectiveness of the interventions. It may be helpful to consider implementing interventions across various naturalistic settings to reinforce treatment principles and promote generalization of treatment effects. A naturalistic play environment provides the opportunity to develop play skills and interests, assisting with skill generalization across contexts and outside the intervention context. These results are consistent with the results of a previous review of school-based social skills interventions for children with ASD (Bellini, Peters, Benner, & Hopf, 2007). Using meta-analysis analysis of 55 single subject design studies, Bellini et al. (2007) recommended educators in school settings select interventions that could be implemented in naturalistic settings, as opposed to removing children from the classroom or playground for the intervention. Bellini et al. (2007) suggested that the familiarity and inclusion in real social situations had a positive effect on treatment outcomes. Further research is required to investigate contextual factors that influence outcomes. As such, clinicians and educators should not limit their choice of interventions to improve play skills in children with ASD based on setting. Unfortunately, reporting of generalization of play skills across environments has been neglected in the studies included in this review. The lack of reporting of generalization of skills is consistent with other psychosocial interventions for children with ASD (Rao et al., 2008;Reichow et al., 2013). Interventions that provide opportunities to develop skills in real social situations and across different contexts need to be balanced with what is feasible and practical for families, clinicians, and researchers.

Outcome Measures
Play is frequently used to improve other developmental areas, rather than being the focus of the study (Wong et al., 2015). Many studies using a play-based intervention to improve communication and social skills in children with ASD did not use a play outcome measure, resulting in their exclusion from this analysis. These excluded studies typically reported on aspects of social communication, such as joint attention, but not a comprehensive measure that captures the complex skills involved in play.
Play as an independent outcome may have diminished social validity as it is not researched as much as other related skills, such as language and general social skills. Social validity is the significance of the intervention strategies and treatment objectives and refers to the perceived social importance of the intervention results (Foster & Mash, 1999). Certainly, social interactions with peers have demonstrated social validity but this is not necessarily associated with play skills (Watkins et al., 2015). Furthermore, reduced social validity is often related to reduced treatment fidelity, which, in turn, may influence treatment effects (Callahan et al., 2017). Therefore, there is a need to educate parents, teachers, clinicians, and researchers on the importance of improved play outcomes in and of itself. Clinicians and researchers should consider the feasibility of including additional education and resources for parents and teachers on the importance of improving play as an outcome of the intervention.
Even when studies met the inclusion criteria for this review, play outcomes were not necessarily the primary focus of the intervention. This may be due to reduced social validity of play with clinicians and researchers. An alternative explanation may be that the outcomes focused on foundation level social skills that are easier to observe and therefore measure. For example, joint attention, behavior, and communication outcomes are frequently the outcomes that were measured in play-based interventions. However, it is difficult to say if improvements in these foundation skills contribute to the development of more complex play skills without therapists and researchers also reporting on play outcomes. Using outcome measures that report on play will support the social validity of play and encourage researchers, clinicians, and families to take play seriously (Bundy, 1993). Reporting on both play and social outcomes allows researchers to develop interventions that are more closely aligned with outcomes that families' value and which will impact on peer engagement.
Observation was the most frequent means of measuring play in this review; however, not all observations were reported using validated measures with published psychometric properties. An example of an appropriate norm-referenced standardized assessment is the Child-Initiated Pretend Play Assessment (CHIPPA; Stagnitti, 2007;Uren & Stagnitti, 2009). The CHIPPA measures the complexity of a child's play skills, their ability to use symbols, and being reliant on someone else for play ideas. A possible explanation for why researchers are creating outcome measures specific to the study and not using preexisting outcome measures with proven psychometric properties is because of the difficulty of measuring play in a natural setting, given the complexity and intrinsic motivation inherent to play (Bundy, 1993(Bundy, , 2011. As such, measuring playfulness may provide a consistent, valid, and reliable alternative (Bundy, 1993(Bundy, , 2011. Playfulness is defined as a disposition to engage in play and has been shown to be responsive to change following intervention (Bundy, 2010). The Test of Playfulness is an appropriate outcome measure for observing play in natural settings with robust psychometric properties (Bundy, 2010;Skard & Bundy, 2008).

Recommendations
The continued development of play interventions for children with ASD using RCTs is important. Researchers conducting RCTs need to clearly report the intervention components following the CONSORT statement, so play-based intervention research can be advanced and potentially be adapted to different settings.
Consistent and comprehensive reporting of play outcomes using valid and reliable measures when investigating a playbased intervention is needed. The use of play outcomes by researchers and clinicians will support the social validity of play and allow for balanced comparisons between interventions. Future research should also consider identifying and comparing the active ingredients within an intervention. Specifically, further investigation is recommended into the use of peers and how they could be more effectively utilized to support the child with ASD to improve their play.
Finally, while the current review included children aged 2 to 12 years, there was significant variability in the inclusion criteria of participants of the studies, including developmental ability of the study participants. We recommend that future investigations include descriptive information of participants' language and social skills to enable clinicians to determine if the intervention would be appropriate to their client's needs.

Limitations
The inclusion criteria requiring the studies to report on play outcomes were necessary to be able to compare across studies; however, they were potentially restrictive, and effective play intervention studies may have been missed in this review because they did not explicitly report on play outcomes. While similarities of participant demographics between intervention and comparison groups remained similar across the different studies, there was some variability in the type of comparison groups. Seven of the studies used a wait-list, no treatment control group, while the remaining studies used an alternative treatment comparison group. Due to the differences in these comparison group types, and to make balanced comparisons between the studies, we included only alternative treatment comparison groups' studies used in meta-regression. This ensured homogeneity between comparison groups and outcomes but limited the number of combinations that could be assessed due to collinearity. As a result, significant relationships between study components may not have been identified.

Conclusion
The results of this systematic review and meta-analysis suggest that play-based interventions produced small to medium treatment effects between 0.083 and 0.586 for children with