Obesogenic behaviors during structured periods among children and adolescents with intellectual and developmental disabilities: a systematic review and meta-analysis

PubMed, PsycINFO, Embase, and Web of Science databases were used to identify relevant articles published before December 31st, 2024, no additional studies were identified after this date. The search strategy used the PICO framework, aligning search components with what is the population (P); the intervention/exposure (I); comparison (C), and outcome (O) [34]. Our search terms for the four PICO categories are as follows: (P) Youth with IDD (e.g., children and adolescents with intellectual and/or developmental disabilities), (I) Structure (e.g., weekday, school), (C) Less/more structure (e.g., weekend, summer), and (O) Behavior (e.g., physical activity, screen time, sleep, diet). Terms were truncated and combined using Boolean operators. An example of key term/MeSH terms is provided in supplementary Table 1. Additional manual searches were conducted using reference lists of previous reviews to identify any additional articles.

To be eligible for inclusion, studies met the following criteria: (i) observational design or report baseline data from an intervention trial. Observational studies were eligible because the paper sought to examine behaviors void of any intentional changes from an intervention. Interventional study designs were not excluded because, in theory, the baseline of intervention is analogous to a cross-sectional design; (ii) include participants who were children (aged 5–12 years) and/or adolescents (aged 12 to 18 years), or report a mean participant age that fell within these age ranges; (iii) participants diagnosed with at least one intellectual or developmental disability (defined in introduction); (iv) provide point estimates and variance of at least one obesogenic behavior across two timepoints with different amounts of structure (e.g., school vs. summer/vacation, weekday vs. weekend); and (v) published in the English language. Any studies that compared time within the same day (e.g., classroom versus recess) or reported time in a behavior as a proportion were excluded. Poster or conference presentations and dissertation work were not included. There were no geographical restrictions. The PRISMA flow diagram outlining the screening for eligible papers is pictured in Fig. 1.

All studies returned by the searches in each database were uploaded to Covidence Systematic Review Management Program, where duplicates were automatically removed (n = 1,607). Two authors (KK and MH) independently examined all potential studies by title and abstract to ensure they did not have any clear exclusion criteria (e.g., examined behaviors in only adults) before reaching a full-text screening. All articles that met the inclusion criteria after full-text screening were passed on to the extraction phase. In instances of disagreement concerning study inclusion, the authors (KK and MH) met to discuss the uncertainty and made a final decision together.

Key descriptors and outcomes measures were extracted from all eligible studies. Descriptors consisted of citation details (first author, year of publication), country, sample size, number and proportion of female participants, study region, and disability related information (e.g., diagnosis or disability inclusion criteria), type of comparison of structure (e.g., weekend vs. weekday or School vs. Summer), and outcome measured along with the measurement tool and procedures (e.g., accelerometer, self-report, PA thresholds). Sedentary behaviors consisted of both sedentary time and screen time. Extraction for screen time consisted of all day screen time and screen time after 8pm. PA consisted of moderate-to-vigorous PA (MVPA), energy expenditure, total PA, and steps. Extraction for sleep consisted of sleep duration (e.g., total sleep time), sleep onset, and sleep offset. Diet just consisted of a quantity of specified food categories (e.g., dairy, fast food, fruit, salty snacks, sugar sweetened beverages (SSB), sweets, and vegetables). For studies that reported median and interquartile range and/or range, an equation was used to transform them into mean and standard deviation [35]. When information was missing (e.g., sample size of subgroup in analysis) an attempt was made to contact the corresponding author to retrieve missing information before using the available data or eliminating the study from the analysis.

The risk of bias for each study was assessed using the NHLBI Study Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies [36]. This assessment consists of 14 questions that two separate raters (KK and MH) answer to assess the potential flaws in study methods that may lead to sources of bias. Each question could be answered with either met, did not meet, did not report, or not applicable. Any discrepancies regarding the risk of bias were resolved through discussion between the two reviewers (KK and MH), who reached a consensus decision. Risk of bias was used descriptively and not as an exclusion criterion or as a sensitivity analysis due to the limited number of available studies.

The standardized difference in means (SDM) and its standard error was calculated for each outcome within a study using Comprehensive Meta-Analysis (CMA, Version 4; Biostat, Englewood, NJ). The SDMs were coded in a manner that all positive effects supported the SDH (e.g., increased MVPA during periods with more structure) while negative effects did not support the SDH (e.g., decreased MVPA during periods with more structure). To synthesize findings across studies, we estimated pooled effects using both fixed- and random-effects models. The fixed-effects model assumes that all included studies estimate a common underlying effect size and that observed differences between studies arise solely from sampling error. In contrast, the random-effects model incorporates between-study variability by assuming that each study estimates a related true effect size. This approach accounts for heterogeneity attributable to differences in study populations, measurement protocols, or contextual factors. For this reason, random-effects models were considered primary due to expected heterogeneity across populations, diagnoses, and measurement tools. Presenting results from both models allows us to evaluate the robustness of the pooled estimates under varying assumptions about heterogeneity and provides a more comprehensive interpretation of the synthesized evidence. Effect sizes of ≤ 0.20 were considered small, effects from > 0.20–0.79 were considered medium, and effect sizes ≥ 0.80 were considered large [37]. Based on the guidance from the Cochrane Handbook for Systematic Reviews of Interventions we estimated and presented both fixed and random effects models [35]. The I² statistic was used to assess heterogeneity among effect estimates for each outcome. Thresholds of ≤ 30%, > 30%-60%, ≥ 60% were interpreted as low, moderate, and high heterogeneity, respectively [35]. A one-study-removed sensitivity analysis was employed to assess the influence of individual studies and identify potential outliers. Potential moderators were included in the Comprehensive Meta-Analysis data set, and a meta-regression was used to evaluate the influence of these potential moderators. Age was coded as children (5–12 years old), adolescent (≥ 13 years old), or both if the same included both age ranges. Diagnosis was coded, autism spectrum disorder, cerebral palsy, developmental coordination disorder, intellectual disability, Prader-Willi syndrome, and Down syndrome. Measurement tool was either objective, subjective, or both if the study used both (e.g., parent-reported sleep and accelerometry). Relevant information was exported to excel and forest plots were created for visual representation of effect sizes and 95% confidence intervals.

There was a moderate overall effect for PA (Random = 0.27, [95%CI: 0.13–0.40], p < 0.00; Fixed = 0.21, [95%CI: 0.17–0.25], p < 0.00) indicating PA was higher during periods with more structure. There was a high level of heterogeneity within the fixed-effects model (I² = 92.19%) and medium in the random-effects model (I² = 31.75%). MVPA effect size was positive and statistically significant with a moderate effect (Random = 0.32, [95%CI: 0.12–0.51], p < 0.00; Fixed = 0.31, [95%CI: 0.26–0.37], p < 0.00). Total activity (steps, counts, energy expenditure, etc.) also supported the SDH and was statistically significant with a small effect (Random = 0.21, [95%CI: 0.08–0.35], p < 0.00; Fixed = 0.14, [95%CI: 0.09–0.19], p < 0.00). Meta-regression analyses indicated that diagnosis, age, or measurement tool was not significantly associated with the pooled effect size. Supplemental Table 2. shows the effect sizes from the one-study-removed analysis which ranged from 0.16 to 0.27 (Fixed) and 0.24–0.29 (Random).

The overall effect for sedentary behaviors (sedentary and screen time) was small and did not support the SDH (Random = -0.01, [95%CI: -0.38-0.36], p = 0.95; Fixed = -0.05, [95%CI: -0.11-0.00], p = 0.06). There was a high level of heterogeneity within the fixed-effects model (I² = 97.55%) and no heterogeneity within the random-effects model (I² = 0.0%). The indication of no heterogeneity by the I² in the random effects model is likely due to the low number of studies and how the model estimates variance. The overall effects for sedentary time did not support the SDH (Random = -0.20, [95%CI: -0.82-0.43], p = -0.53; Fixed = -0.31, [95%CI: -0.40- -0.23], p < 0.00) demonstrating no difference in sedentary time during periods of more structure compared to periods of less structure. Screen time had a small-moderate effect size (Random = 0.23, [95%CI: -0.12-0.58], p = 0.19; Fixed = 0.16, [95%CI: 0.08–0.23], p < 0.00) and showed that amount of screen time did not differ during periods of more structure compared to periods with less. Meta-regression analyses indicated that diagnosis was significantly associated with the pooled effect size (β = 0.3, 95% CI [0.04–0.56], p = 0.02), suggesting that studies examining only children and adolescents with ASD were more likely to report less sedentary behaviors during periods with more structure, compared to other IDD. Age and measurement tool was not associated with the pooled effect size. Supplemental Table 3 shows the effect sizes from the one-study-removed analysis which ranged from − 0.07 to 0.14 (Fixed) and − 0.05–0.15 (Random).

The overall effect for sleep was small-moderate but was not statistically significant (Random = -0.01, [95%CI: -0.16-0.14], p = 0.88; Fixed = -0.06, [95%CI: -0.11- -0.01], p = 0.03). There was a high level of heterogeneity within the fixed-effects model (I² = 86.42%) and small in the random-effects model (I² = 5.24%). A moderate effect was seen for total sleep duration and did not support the SDH (Random = -0.15, [95%CI: -0.32-0.03], p = 0.10; Fixed = -0.19, [95%CI: -0.26- -0.13], p < 0.00) suggesting sleep duration was shorter during periods with more structure. Sleep onset (Random = 0.12, [95%CI: -0.01-0.26], p = 0.08; Fixed = 0.12, [95%CI: 0.01–0.23], p = 0.04) and sleep offset (Random = 0.25, [95%CI: -0.01-0.51], p = 0.06; Fixed = 0.22, [95%CI: 0.09–0.34], p < 0.00) were not statistically significant. However, the directionality of the effects shows sleep onset and offset were later during periods with less structure. Meta-regression analyses indicated that diagnosis was significantly associated with the pooled effect size (β = -0.33, 95% CI [-0.49- -0.16], p < 0.01), suggesting that studies examining only children and adolescents with cerebral palsy were more likely to report worse sleep during periods of more structure, compared to other IDD. Supplemental Table 4. shows the effect sizes from the one-study-removed analysis which ranged from − 0.05 − 0.03.

While there was only one study that examined diet, it included two separate comparisons (e.g., summer camp weekdays vs. no summer camp weekdays and summer camp weekdays vs. summer weekend days) so we ran a single fixed model to examine the overall effects. The effects of diet were small and supported the SDH (0.16, [95%CI: 0.03–0.29], p = 0.02). All measures of diet were positive except consumption of sweets but not statistically significant, and dairy was the only statistically significant effect observed (0.36, [95%CI: 0.01–0.71], p = 0.04). A meta-regression and one-study-removed were not run for this study because there was only a single study analyzed.

The shortcomings of this review highlight the gaps in literature examining obesogenic behaviors in children and adolescents with IDD. A notable limitation is not only the small pool of studies but the lack of studies across all behaviors. As stated earlier, PA was the most studied behavior with 23 studies examining various aspects of PA; however, there were only 5 studies examining screen time. Relying on a small collection of studies impacts the ability to generalize the pooled estimate from these studies, an issue made more complex by the variability in study characteristics among these studies (e.g., measurement tool, sample size, disability). The findings of this paper may offer early evidence that the SDH framework may function similar in children and adolescents with IDD, yet more research is needed to form a well-supported conclusion.

The research on school versus summer is very limited and the only study that used objective measures from school to summer only had 6 participants. Further research is needed evaluating the difference in these obesogenic behaviors in the school year compared to summer. When examining weekend behaviors, it is logical to presume a caregiver would, to some extent, be able to keep routines, activities, and overall structure of the day consistent for these 48 h. However, when days with less structure are extended, like during summer break, caregivers may be limited in their ability to keep similar routines and behaviors consistent due to constraints in resources (e.g., time of work, money for day camps, availability of developmentally appropriate camps). This could have a two-fold negative impact, worsening of obesogenic behaviors of the child and/or increased burden on caregiver to provide structure that is unable to be captured during the weekend. While examining seasonal differences in BMI gains and obesogenic behaviors in children with IDD, these studies could explore secondary outcomes related to household stress and other indirect outcomes that may be negatively impacted during the summer.

Second, only two studies examined sleep, physical activity, and sedentary behaviors, together, allowing for a more granular view of changes in these behaviors in the context of a 24-hour framework. For example, Brazendale et al. (2023) showed that children with autism spectrum disorder had slightly more sleep on school days which resulted in a shift of less sedentary time and light PA and more MVPA compared to summer days [80]. Similarly, Smit et al. (2020) measured sleep duration, sedentary time, and active time, enabling an examination of how time allocated to these behaviors differed between weekdays and weekends [83]. While it is important to understand how structure impacts obesogenic behaviors, a more complete approach to understanding structure’s impact on these behaviors is to examine how time is allocated across these three behavior categories. For instance, a day in which sleep duration is shortened and replaced with more sedentary time is less favorable than a day in which that time is filled with some form of PA. Building upon this 24-hour movement behaviors perspective, future research can use time-use methodology to identify where and with whom these individuals accumulate increases in PA versus sedentary time. Ultimately, allowing researchers to integrate time-use methodology with 24-hour movement behavior measures to examine the context (e.g., where and with whom) that optimal behaviors occur to inform future interventions targeting obesogenic behaviors in children and adolescents with IDD [84].

This review provides important implications for the prevention and treatment of obesity in children and adolescents with IDD. The SDH has been instrumental in our understanding of the etiology of obesity and our ability to treat and prevent it in typically developing children. The extension of this framework to children and adolescents with IDD could provide additional paths of addressing obesogenic behaviors and improving BMI in this population that mirror those of typically developing children. Similar to the literature of typically developing children, MVPA is significantly improved during periods with more structure, suggesting that interventions intended to improve PA in this population may want to target periods of reduced structure, opposed to adaptive current forms of structure. Conversely, diet and sleep intervention may have more difficulty materializing meaningful outcomes in this population as our review shows these behaviors varied minimally across varying amounts of structure. Demonstrating that structure can improve behaviors and health outcomes in children and adolescents with IDD may have additional implications for clinical practices. Structure may act as a low-level but persistent exposure that works in conjunction to ongoing interventional therapies. Clinicians may also benefit from monitoring seasonal periods of increased risk and modify current standard of care and provide contextual recommendations. The findings of this paper suggest that extending the SDH to children and adolescents with IDD may provide clinically relevant guidance for monitoring periods of increased health risk and integrating structure as a complementary component of standard care.

Future research should aim to address the weaknesses addressed previously while continuing to build the pool of literature examining obesogenic behaviors during periods of structure among children and adolescents with IDD. Future research should prioritize larger sample sizes with consistent methodologies, examining multiple obesogenic behaviors across extended periods of reduced structure, such as the summer. Further, it is important that research begins examining the relationship between changes in obesogenic behaviors and changes in BMI and begins to identify population specific factors (e.g., individual- and family-level) that may explain the variation in outcomes, allowing for better generalizability and mechanistic understanding. As this line of research grows and our understanding deepens, structure’s benefits may be used to inform decision-making efforts of practitioners and policy development.

This review has not been registered. There is no review protocol published for this manuscript.

Not applicable.

Not applicable.

The authors declare no competing interests.

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.