Physical activity, sedentary behavior and their correlates in children with Autism Spectrum Disorder: A systematic review

Autism Spectrum Disorder affects up to 2.5% of children and is associated with harmful health outcomes (e.g. obesity). Low levels of physical activity and high levels of sedentary behaviors may contribute to harmful health outcomes. To systematically review the prevalence and correlates of physical activity and sedentary behaviors in children with Autism Spectrum Disorder, electronic databases (PsycINFO, SPORTDiscus, EMBASE, Medline) were searched from inception to November 2015. The review was registered with PROSPERO (CRD42014013849). Peer-reviewed, English language studies were included. Two reviewers screened potentially relevant articles. Outcomes of interest were physical activity and sedentary behaviour levels and their potential correlates. Data were collected and analysed in 2015. Of 35 included studies, 15 reported physical activity prevalence, 10 reported physical activity correlates, 18 reported sedentary behavior prevalence, and 10 reported sedentary behavior correlates. Estimates of children’s physical activity (34–166 mins/day, average 86 mins/day) and sedentary behavior (126–558 mins/day in screen time, average 271 mins/day; 428–750 mins/day in total sedentary behavior, average 479 mins/day) varied across studies. Age was consistently inversely associated, and sex inconsistently associated with physical activity. Age and sex were inconsistently associated with sedentary behavior. Sample sizes were small. All but one of the studies were classified as having high risk of bias. Few correlates have been reported in sufficient studies to provide overall estimates of associations. Potential correlates in the physical environment remain largely unexamined. This review highlights varying levels of physical activity and sedentary behavior in children with Autism Spectrum Disorder. Research is needed to consistently identify the correlates of these behaviors. There is a critical need for interventions to support healthy levels of these behaviors.

Introduction Low levels of physical activity (PA) and high levels of sedentary behaviors (SB) are associated with short-and long-term health consequences from early childhood, through childhood and adolescence. Consequences include cardio-metabolic risk factors, impaired psychosocial wellbeing and cognitive functioning and poorer weight status [1][2][3][4][5][6][7]. However, across the world few children or adolescents (hereafter referred to as children) participate in at least 60 minutes per day of moderate-to vigorous-intensity physical activity (MVPA) as is recommended [8][9][10][11][12][13]. There are particular sub-groups whose risk of inactivity may be even higher, such as children (0-18 years) diagnosed with an Autism Spectrum Disorder (ASD).
ASD is clinically defined by impairments in social, communication and reciprocal interaction, with repetitive, restricted, and stereotypical behavioral patterns [14] and is now thought to affect up to 2.5% of children [15][16][17]. Social impairment and restricted interests combined with high rates of motor problems [18] commonly present in this disorder may limit the opportunities for engagement in PA and put children with ASD at risk of increased SB and other harmful health outcomes including and obesity [19]. Engaging in PA may offer opportunities for socialization with peers, broaden a child's range of interests, increase motor skills and thus have positive impacts on a range of outcomes including physical health, social-emotional and developmental functioning [20]. Therefore, similar to typically developing (TD) children [21], children with ASD could potentially benefit from lifestyle modifications that promote increased PA and reduced SB. To date, there is no clear evidence documenting the prevalence of PA and SB in children with ASD, nor comprehensive evidence of the correlates of these behaviors. Understanding prevalence and correlates will inform future PA and SB interventions and identify potential strategies. Therefore, the aim of this review was to systematically examine reported prevalence of PA and SB, along with their correlates, in children with ASD.

Methods
The guidelines outlined in the Preferred Reporting Items for Systematic Reviews and Metaanalyses (PRISMA) statement [22] were followed (S1 Table. Completed PRISMA checklist). The review was registered with PROSPERO (CRD42014013849); no protocol exists. A literature search was conducted using four databases: PsycINFO, SPORTDiscus with Full Text, EMBASE and Medline. Search strategies included key words in four categories: (1) population, (2) PA, (3) SB and (4) Autism Spectrum Disorder. Table 1 provides the full search strategy for Table 1. Search strategy used in PsycINFO.

Population
"school age*" or "young child*" or youth or adolesc* or teen* or preschool or early childhood or child* or infant or boy* or girl* or female* or male* 2. Physical activity "physical* activ*" or exercis* or "motor activit*" or "locomotor activit*" or sport* or recreation* PsycINFO, modified where necessary for the remaining databases. Searches were completed from inception of the databases to November 2015. Data were collected and analysed in 2015/ 2016. Papers were included if they were: (1) peer reviewed, written in English and available in full text; (2) included children aged 0-18 years with ASD; and (3) reported on the amount of time children spent in habitual PA/SB; or (4) reported a measure of association between PA/SB and another variable (e.g., sex). Case studies and studies that reported only qualitative data or measures of fitness and energy expenditure were excluded. Exclusion criteria were case studies and studies that reported fitness or aerobic fitness or energy expenditure. No sample characteristics, such as low or high functioning Autism, nor the presence of comorbidity, were considered as exclusion criteria to allow for the greatest number of papers to be considered for inclusion.
The following process was followed to ensure that all relevant papers were retrieved and included: (1) two authors initially screened titles (KD and RJ); (2) abstracts and full papers of those that were deemed to be relevant were then screened by the same two authors; and (3) data were extracted from included studies (KD). For steps one and two, where authors did not agree (<5%) resolution was achieved through discussion or a third author (TH) was consulted. For analyses focused on correlates, only one study reported results of multiple variable analyses with all studies reporting bivariable and unadjusted models. Therefore, results of unadjusted models are reported. Notations are made in tables where differences in unadjusted and adjusted models exist.
Prevalence of PA included average daily minutes of MVPA, total PA (counts per minute, CPM) or number of days being active. Where available, the percentage of children meeting the 60min MVPA international recommendations [23] was extracted. For SB, prevalence was reported as average daily minutes of screen time (TV watching, video gaming, etc.) or total sedentary time. Reported associations between behaviors and potential correlates were coded as positive (+), negative (-) or null (0). The number of studies supporting each association is given as a percentage of the total studies for that variable. These codes were then analyzed and given a summary code based upon the percentage of studies and direction of association as follows: 0-33% of studies reporting an association in a given direction were coded as 0 (no association); 34-59% of studies supporting an association in a given direction were coded as? (inconsistent association); and 60-100% of studies supporting an association in a given direction were coded as + or-(positive/negative association). Variables investigated four or more times were coded ++ (strong positive association),-(strong negative association), or 00 (strong null association) [24][25][26]. Only the variables that are reported four or more time are discussed. If variables are reported fewer than four times, the evidence was considered insufficient to make an assumption about the direction of association. Correlates were divided into subgroups: child, familial and environmental variables, aligning with the Ecological Model [27]. A meta-analysis was not completed due to too few studies investigating the same correlate and outcome.
Risk of bias was determined using a published six-component rating scale [28] and was assessed on: selection bias (e.g., representativeness), study design (e.g., longitudinal), confounders (e.g., education), blinding (e.g., outcome assessor aware of group allocation), data collection methods (validity, reliability), and withdrawals/dropouts (e.g., were they reported, percent of complete data: weak <60%; strong !80%). A risk of bias score for each subsection was calculated as per the standardized protocol (1 = low risk, 2 = moderate risk, 3 = high risk). If a component was not described it was rated as high risk of bias. Once individual components were assessed, a global score for each paper was calculated (1 = low risk [3 low risk ratings, no high risk ratings], 2 = moderate risk [<3 low risk ratings, 1 high risk rating], 3 = high risk [>1 high risk rating]). The data extraction tables are provided in the supporting information (S2 Table. Data extraction table for the physical activity papers; S3 Table. Data extraction tables for the sedentary behavior papers)
Sixteen of the 17 PA studies were classified as having high risk of bias; only Memari et al [30] had moderate risk of bias. The only risk of bias subsection that consistently received a low rating was the data analyses procedures (S4 Table).
Five studies reported PA prevalence for samples of ASD and typically developing (TD) children [29,36,[38][39][40]; two studies [29,38] did not report statistical assessment of difference between those samples for accelerometry data. However, Bandini et al. [29] reported both accelerometry and parent-report PA data. They found that there was no association between the two for either children with ASD or TD children but that TD children participated in significantly more types of activities, and more hours per year of activities, than children with ASD when measured by parent-report. Dreyer-Gillette [39] found that a significantly higher percent of children with ASD (18.8%) spent no days being physically active (20 mins or more in vigorous PA) compared with their TD counterparts (10.2%). Similarly, a significantly lower percent of children with ASD (31.3%) spent 4-6 days per week being physically active compared with TD children (39.8%). However, there were no differences in the percent of ASD and TD children who spent 1-3 or 7 days being active [39]. Tatsumi [40] reported that children with ASD had significantly lower CPM (630.5) on weekend day mornings compared with TD children (746.0); CPM did not vary for other times of the week or other periods of weekend days (afternoon, evening). The final study [36] reported no statistical difference in CPM between ASD and TD children. Seven studies reported prevalence of compliance with the 60 minutes of MVPA recommendation [29,31,33,34,36,37,43]: these varied between 21% and 100% for children with ASD [29,31,33,34,36,37,43] and 43% and 92% for TD children [29,36].

Sedentary behavior-Study characteristics
All 18 studies reporting SB outcomes were classified as having high risk of bias. All studies reporting SB outcomes used an appropriate type of analysis for the study design (S5 Table).

Sedentary behavior-Prevalence
Twelve studies reported SB as screen time which incorporated activities such as TV watching, video gaming, computer and social media use [32,[46][47][48][49][50][51][52][54][55][56][57]. Two studies reported the use of smart phone technology [39,55]; one reported on digital tablet use [55]. However, those studies only reported time using those devices as part of a variable which included time using other devices (e.g. laptop) also. Four studies reported total sedentary time (i.e. measured objectively) [37,38,43]. Two studies reported sedentary time as a percentage of the whole day, thus these data could not be compared with other studies [39,53]. The average number of minutes/ day of total screen time ranged from 86 to 430 (average 231 min/day) and the average number of minutes/day of total sedentary time ranged from 312 to 789 (average 514 min/day).
Six studies [39,46,49,51,55,56] reported the prevalence of SB for both ASD and TD samples; only two studies reported significant differences in SB between ASD and TD children. Chonchaiya et al. [46] found that children with ASD spent significantly more time in screen time (4.6 vs. 2.6 hours/day) and fewer complied with screen time recommendations of less    Physical activity, sedentary behavior in children with Autism than two hours per day (6% vs 44%) compared to their TD counterparts. In contrast, Dreyer Gillette et al. [39] found that a significantly higher percent of children with ASD (19.8%) never used a computer/cell phone/hand held video games/other electronic devices compared with their TD counterparts (8.7%). However, that study found no difference in children's use of TV/video/video games, nor in the number of hours children used computer/cell phone/hand held video games/other electronic devices.

Sedentary behavior-Correlates
Eleven studies reported potential correlates of SB (Table 5), with 23 different correlates examined: 15 were child variables; eight were familial variables; no physical environment variables were reported. Only two variables were reported in sufficient studies to determine an overall association. Age and sex showed an inconsistent association with SB across studies.

Discussion
Prevalence rates and correlates of PA and SB among children diagnosed with ASD, and how these potentially differ from TD children, largely remains unknown. In this review, estimates of time in MVPA and screen time per day, and compliance with MVPA and SB recommendations, varied for children with ASD and typically developing children between studies. Despite the substantial variations in estimates of time spent in PA and screen time for ASD and TD children, children with ASD seem to spend more time in screen time (and in turn total SB) and less time in PA than TD children (26-430 mins/day in screen time for children with ASD, compared with 72-318 mins/day in screen time for typically developing children [54,57,59,64]). It is well documented that estimates of behaviors such as PA and SB, and subsequently estimates of compliance with recommendations, vary, based on method of data collection (e.g. survey, accelerometry), data management methods [58][59][60] and other factors such as season [61]. Nonetheless, PA appears low and SB high among children with ASD (when mean minutes per day is used to compare children) and targeted interventions may be warranted once potential correlates are identified. Of the 47 potential correlates examined in the studies in this review (24 for PA and 23 for SB), most (53%) were classified as child variables. Only two child variables, age and sex, were reported sufficient times across studies to allow overall assumptions to be made. Consistent with the literature among TD children, this review found negative associations between PA and age or school level/grade [62,63]. That is, as children's age/school level/grade increases, their level of PA decreases. This suggests that PA interventions for children with ASD should be implemented as early as possible to maximize PA while children are young and curb potential age-associated decreases. In contrast to current literature for TD children, this review found an overall inconsistent association of sex with PA among children with ASD. Among typically developing populations, boys are consistently more active than girls [26,[64][65][66]. The null association reported may have been influenced by the demographics of the participants in the included papers. The two studies which reported that boys were more active than girls had participants, on average, younger than 10 years old. However, for the other two studies where no association between sex and physical activity was identified, the average age of the participants was several years older. As children's age increases, it is feasible that the influence of sex is less defined among children with ASD. Nonetheless, the small number of studies reporting sex as a potential correlate is likely to limit the opportunity to identify any individual variable as a correlate and further research is needed to more comprehensively understand whether or not boys with ASD are more active than girls with ASD.
In terms of SB, in TD children mixed associations have been reported for age and sex, some reporting strong, consistent associations [67] and others inconsistent associations [24,64,68,69]. In the current review, inconsistent associations of age and sex with SB were reported [70][71][72][73]. The relatively small number of studies included in this review may have contributed to the inconsistent associations (six studies reported associations between SB and age and five studies between SB and sex). Furthermore, the small sample sizes for most of the studies included in this review may mean studies are not powered to detect associations. Some studies included only a narrow age range for their samples (for instance, 2.6 years ±0.7 years [46]) and other characteristics may similarly be relatively homogenous given the small samples. Therefore, it is possible that sample characteristics may have insufficient diversity for associations to be identified. A number of child-related 'ASD specific' potential correlates such as ASD symptom severity [37,45,47,57], maladaptive behaviors [32], and psychotropic medication (taking or not taking) [30] were investigated; none were investigated in four or more studies for either PA or SB. Although these variables were not strongly associated, they still may be important correlates to consider in the design of future interventions. There is a growing awareness that ASD is a heterogeneous disorder, with many complexities and comorbidities [74] which give rise to large variations in health and well-being: it is often the type and nature of associated comorbidities and health-related factors which may influence children's outcomes, possibly including reduced opportunity for PA [75]. It is therefore essential to report such comorbidities in studies to determine their relative contribution to health behaviors as well as the contribution PA and SB may have in supporting positive health outcomes. Given the prevalence of ASD comorbidities and other 'ASD specific' potential correlates, it may be that specialized facilitators are the most appropriate personnel to implement interventions. Such personnel would likely have a greater understanding of the complexities of ASD and how different comorbidities may affect PA and SB levels.
The correlates of PA and SB for TD children have been thoroughly evaluated and reported, for instance: [24,25,64,67,[76][77][78][79][80]. While some reviews include a broad range of correlates, such as individual, social and physical environment correlates [25,26,64,67,81], others have focused specifically on one domain or construct (e.g. physical environment, neighborhood safety [82,83]). Reviews in TD children also typically focus on a narrower age range than this review does (e.g., preschoolers [24,80], adolescents [78,84]), and correlates vary across reviews. As in this review, reviews of the correlates of TD children's PA and SB typically identify many variables which have been investigated in only one or two studies. It is therefore difficult to provide a clear and synthesized overview of correlates in TD children against which to compare the correlates in children with ASD. However, across reviews in TD children, boys appear to be more active than girls [24,64], younger children are more active than older children [62,63], children with active parents are more active [24,64,85], and children who spend more time outdoors are more active [24,64]. Parental rules [85], family TV viewing [67] and age [73] may be associated with SB in TD children.
Given the dearth of PA and SB interventions for children with ASD [18], interventions specific for children with ASD may be needed. The content and behavior change strategies to be included in such interventions require further research. Studies investigating prevalence of SB in children with ASD, which include the full range of electronic devices (e.g. computer, digital tablets) are required. In this review no studies of the time children with ASD use digital tablets met the inclusion criteria. With increasing use of these devices in children with ASD for entertainment and education [86][87][88], and strong acceptance by parents and educators [89], studies investigating this behavior are warranted.

Limitations of existing data and subsequent recommendations
This is the first review to summarize prevalence of PA and SB among children with ASD and to report on PA and SB correlates. Results and recommendations should be considered in light of the following limitations: 1. Sample sizes: Studies in this review generally included small samples. Mean sample sizes for PA and SB studies were 41 and 82, respectively, after the one very large study was removed as an outlier. SB studies, on average, included twice as many participants with ASD as PA studies and all studies involved more boys than girls (consistent with more boys (4.3:1) being diagnosed with ASD than girls [90,91]). Ensuring samples are sufficient to provide meaningful results is essential. It is therefore recommended that future studies include a power calculation to ensure they recruit a sample of sufficient size to be able to identify associations should they exist and to provide meaningful results. Furthermore, investigating prevalence rates and correlates for different age groups and both sexes would be helpful to inform future interventions.
2. Age of participants: many of the studies included in this review utilized samples with large ranges in participant ages and reported aggregated data across all children. This may be necessary with small sample sizes. Nonetheless, it is recommended that future studies focus on a narrower age range where children are typically at the same or similar developmental stage (e.g. 3-5 years, 5-11 years, 12-18 years), and exposed to similar environments (e.g. preschool, primary school, secondary school) to allow more meaningful and potentially useful results to be reported.
3. Number of correlates: In this review, few correlates have been investigated in sufficient studies to draw consistent conclusions. It is recommended that future studies endeavor to measure as broad a range of potential correlates as possible, as studies in TD children show that correlates exist across all domains of the ecological model [24,25,64,[92][93][94][95]. This may also be the case among children with ASD. Including a broad range of correlates allows for the relative strength of associations to be examined when other influences are controlled for and aids future intervention development to ensure efficacious outcomes. Although variables included will be specific to each study to meet the intended aims, suggestions include: age, sex, ASD symptom severity, perceived parental barriers, parental self-efficacy, and characteristics of the neighborhood environment which may be supportive of physical activity. Further, a number of potentially important correlates remain to be examined: no studies examined motor competence [96], preference for sedentary/active opportunities [97], access to facilities or neighborhood environment characteristics [98,99], which may support or constrain behaviors.
4. Consistent reporting of outcomes: In this review, reporting of data within studies was inconsistent. For example, different types of SB were reported in each study: some reported time watching television [46] or video-gaming [49,50]; others reported television viewing and video-gaming or computer use [32,47]. No studies were identified which included specific SB other than screen behaviors, such as reading or doing craft. Reporting varied between studies, making comparisons difficult: some reported prevalence rates over seven days [33,34,38,43]; others only reported weekday and/or weekend prevalence [40,42]. Most studies reporting SB outcomes used proxy-report questionnaires: only five reported validity and reliability data for those questionnaires, raising concerns regarding instrument accuracy [32,38,43,44,52]. It is recommended that future studies, where feasible, assess physical activity and sedentary behavior/screen time using reliable and validated instruments, ideally using objective measures and reporting over a seven-day period.
5. ASD diagnosis: The majority of studies did not confirm ASD diagnosis of the children and relied on prior or parent-report. DSM criteria were seldom reported as confirmation of clinical diagnosis. Many studies included children with comorbid ID. It is unclear whether PA and SB are impacted by ASD symptom severity, having comorbid ID or compounded by the presence of both. Thus, it is recommended that future studies consistently report children's diagnosis of ASD, in line with DSM diagnostic criteria and take into consideration common co-morbidities for this clinical populations.
6. Risk of bias: The risk of bias of the studies in this review was overwhelmingly high; this is important as it aids in comparison [100] and should be considered in future research in this population. It is possible that studies included in this review were methodologically robust but were rated as having high risk of bias due to inadequate reporting. It is recommended that reporting details of studies in line with existing statements (e.g. STROBE [101], CON-SORT [102]) should be adhered to so study risk of bias, rather than poor reporting, can be accurately assessed.

Conclusions
This review highlights generally lower levels of PA and higher levels of SB among children with ASD compared to TD children [103]. Furthermore, this review highlights that age is consistently inversely associated, and sex inconsistently associated with physical activity. Age and sex are inconsistently associated with sedentary behavior. For optimal physical activity and sedentary behavior levels, this review supports the notion that children with ASD could potentially benefit from lifestyle modifications that promote increased physical activity and sedentary behavior. Additionally, given the complexity of ASD, studies investigating additional correlates of ASD would be beneficial. Cohort studies would be beneficial in providing a stronger level of evidence and showing temporal associations.
Supporting information S1  Table. Risk of bias a for papers reporting physical activity outcomes. a 1 = strong quality/ low risk of bias, 2 = moderate quality/risk of bias, 3 = weak quality/high risk of bias. (DOCX) S5 Table. Risk of bias a for papers reporting sedentary behavior outcomes. a 1 = strong quality/low risk of bias, 2 = moderate quality/risk of bias, 3 = weak quality/high risk of bias.