A comprehensive search was conducted on the topic “effectiveness of exercise interventions on improving gross motor skills in children with Autism Spectrum Disorder” to identify all studies reporting changes in gross motor skills following exercise interventions in autistic children. Six databases were systematically searched: China National Knowledge Infrastructure (CNKI), Wanfang Data, PubMed, ProQuest, Web of Science, and EBSCO, from their inception to September 20, 2025. The search strategy employed a combination of keywords and Medical Subject Headings (MeSH) terms, constructed around four core concepts: Autism Spectrum Disorder (autism, ASD, etc.), Exercise Intervention (physical activity, sports, etc.), Gross Motor (motor skills, gross motor skills, etc.), and Children (child, preschool, etc.). Synonyms within each concept group were connected using the Boolean operator “OR,” and different concept groups were combined using “AND.” To ensure complete transparency and reproducibility, the reproducible search syntax executed across all six databases is provided as a Supplementary File (Table 1). Below is an example of the core keyword framework used to construct the search strategy. A detailed description of the search strategy is provided in the Appendix.

The Zotero software was used to screen the retrieved literature. Inclusion criteria were determined based on the PICOS principle. Participants (P) were children aged 3–12 years diagnosed with Autism Spectrum Disorder. Interventions (I) included land-based exercise training, aquatic exercise training, comprehensive exercise, and technology-assisted interventions (e.g., VR technology). Comparisons (C) were blank controls, routine exercise, or different exercise interventions from the experimental group. Outcome measures (O) were scores from relevant assessment tools, such as the gross motor sections of HAAR, BOTMP, PDMS-2, TGMD-2/3, BOT-2, MABC-2, etc. Study designs (S) included randomized controlled trials and non-randomized controlled trials.

Exclusion criteria were as follows: ① Participants were not diagnosed with ASD or had other severe neurological/musculoskeletal comorbidities; ② Interventions were non-exercise-based, or the comparison between experimental and control groups did not reflect the value of exercise intervention; ③ Non-Chinese or non-English literature; ④ Literature for which the full text was unavailable, had incomplete data, or was duplicate publications; ⑤ Studies that did not report complete data for the gross motor development assessment tools relevant to this study; ⑥ Non-controlled trials.

The primary outcome of this study was the change in standardized assessment scores for gross motor skills in children with ASD. The included studies utilized various internationally recognized assessment tools, mainly including: (1) Comprehensive motor development tests: Gross motor sections of the *Peabody Developmental Motor Scales-Second Edition (PDMS-2)*, and the *Bruininks-Oseretsky Test of Motor Proficiency (Second Edition/Revised) (BOT-2/BOTMP)*. These tools provide composite or subscale scores for dimensions such as balance, locomotion, and object control. (2) Fundamental motor skill-specific tests: *Test of Gross Motor Development-Second/Third Edition (TGMD-2/TGMD-3)*, specifically designed to assess locomotor and object control skills. (3) Other tools: Hong Kong Assessment of Attention for Autistic Children (HAAR), etc. Despite the diversity of tools, their core focuses revolve around the three major gross motor skill domains: balance, locomotion, and object control. For data synthesis, we extracted post-test group means, standard deviations, and sample sizes for these three dimensions or the most closely related dimensions from each study. If a study only reported total scores, it was included in the overall effect analysis but not in the skill-dimension subgroup analyses.

Studies were initially screened by title and abstract, followed by assessment of full-text eligibility. Discrepancies were resolved by consulting a third expert. Information extracted from the studies included participant characteristics (age, gender, country, diagnostic method, sample size), intervention components (intervention type, intensity, duration, frequency, etc.), outcome measures, and results (mean differences in gross motor changes in children with ASD after exercise intervention).

Two researchers independently assessed the methodological quality of the included randomized controlled trials and quasi-experimental studies using the Cochrane Risk of Bias tool. The assessment focused on: random sequence generation, allocation concealment, blinding of participants and personnel, completeness of outcome data, selective reporting, and other potential biases. Discrepancies in quality ratings were resolved through discussion with a third researcher to reach consensus (23).

This study used Review Manager (RevMan) 5.4 software, recommended by the Cochrane Collaboration, for data synthesis and analysis.

Figure 1 illustrates the study selection flow. After searching the six databases, a total of 450 records were identified. Following the removal of duplicates, 430 unique records remained. After assessing full-text articles for eligibility, 414 were excluded for the following reasons: excluded by title/abstract (n=386), invalid data (n=9), insufficient sample size (n=14), full text unavailable (n=5). Ultimately, 16 studies were included in the systematic review.

Table 2 presents the basic characteristics of the 16 included studies (total N = 493 children aged 3-12) conducted in multiple countries/regions from 2014 to 2025. All studies used standardized diagnostic tools. Intervention types included Land-based sports, Aquatic sports, and technology-assisted training. Specific doses and designs are detailed in Table 2.

Figure 2 illustrates the risk of bias across the included studies. Across the five domains, some concerns were noted in the 16 studies, primarily related to selection bias and performance bias (Figure 3). Regarding randomization process, detection bias, attrition bias, and reporting bias, the included studies demonstrated low risk of bias.

Through subgroup analysis, this study systematically compared, for the first time, the effects of different exercise types and intervention doses on the three core dimensions of gross motor skills in children with ASD. The meta-analysis, synthesizing data from 16 studies involving 493 participants, indicates that exercise interventions have a significant and positive overall effect on improving gross motor skills in children with ASD (SMD = 0.81–0.86), providing key evidence for exercise as an effective supplementary rehabilitation approach for children with ASD (28).

The core value of this study lies in revealing heterogeneity of effects through subgroup analysis. Exercise type is a key factor contributing to heterogeneity. Land-based sports demonstrated the most consistent and largest positive effects across all three skill dimensions, making them the most recommendable form of exercise intervention in the current evidence base. In contrast, Aquatic sports only showed significant improvement in locomotor skills, while technology-assisted interventions (e.g., VR training) did not show stable significant benefits in this analysis.

Regarding intervention dose, this study found a non-linear dose-response relationship. The medium-dose regimen with a total duration ≤ 1440 minutes (approx. 8–12 weeks) produced significant improvements in balance, locomotor, and object control skills, suggesting this range may be the “efficient window period” for generating broad benefits. Although the high-dose regimen (> 1440 minutes) maintained a significant improvement in object control skills, its effects on balance and locomotor skills did not reach statistical significance. This may be related to the small number of studies and sample size in the high-dose group rather than fully reflecting dose ineffectiveness.

The above comparison of different types of exercises is based on subgroup analysis across studies. This indirect comparison across studies is significantly less powerful than a direct randomized controlled comparison. Therefore, the current conclusions regarding the differences in the effects of different types of exercises should be interpreted with caution. Future verification of these differences is needed through head-to-head randomized controlled trials.

Although this study conducted a systematic review and Meta-analysis of the included literature, it still has the following limitations. Firstly, in the research methods section, some of the included studies were non-randomized controlled trials, and there was a high risk of bias in aspects such as the generation of randomization sequences, the concealment of allocation, and the implementation of blinding for subjects and intervention implementers (60). These methodological limitations may have overestimated the overall intervention effect to some extent. Secondly, in terms of outcome measurement, as mentioned earlier, the mainstream gross motor assessment tools used in this study (such as TGMD, BOT-2) mainly target motor skills in the terrestrial environment. This measurement characteristic may give a natural advantage to terrestrial movement interventions, while failing to fully capture the unique benefits of water-based or technology-assisted movements, thereby affecting the fairness of comparisons between different types of movements. Thirdly, in the analysis section, the intervention dose threshold set in this study (1440 minutes) was pre-defined based on the distribution of existing studies and clinical experience, but it still has a certain degree of subjectivity and exploratory nature. Its universality needs to be further verified in different populations and environments. Although subgroup analysis was conducted based on movement type and total dose, and lower heterogeneity was observed in some results (such as balance and manipulation skills), there was still moderate heterogeneity in the displacement skill results (I² = 46%), indicating that there are other potential influencing factors, such as the age of the subjects, the severity of symptoms, family participation, and the intervention implementation environment, which have not been adequately controlled. Fourthly, in the evidence base section, in some subgroups, such as water-based movements, technology-assisted movements, and high-dose intervention groups, the number of included studies was limited, resulting in low statistical power and insufficient stability of the corresponding results. The conclusions need to be verified by more high-quality studies. Fifthly, the included studies lacked long-term follow-up data, making it impossible to assess the persistence of the movement effects. Finally, although a publication bias test was conducted, the asymmetry of the funnel plot suggests the possibility of unpublished negative results, which may overestimate the overall effect size.

Based on the findings of this study, structured land-based exercise interventions should be prioritized in future clinical rehabilitation and special education practices, as they have been demonstrated to significantly and comprehensively enhance balance, locomotor, and object control skills in children with ASD. Concerning intervention dosage, it is recommended to limit the total duration to within 1440 minutes, as this range has been identified as a key window for achieving optimal benefits. For Aquatic sports and technology-assisted training, current evidence does not support their widespread adoption as core intervention programs. However, they may be considered as personalized supplementary approaches when resources permit, provided that attention is given to program standardization and individual adaptability (61).

Regarding future research directions, it is recommended to conduct additional high-quality, large-sample randomized controlled trials. Efforts should focus on strengthening the evidence base in areas currently supported by limited data, such as Aquatic sports and technology-assisted interventions. Furthermore, research should further explore the adaptive relationship between different intervention dosages and variables such as child age and symptom severity, to identify the “optimal dose” for specific populations. The incorporation of neuroimaging techniques, such as fMRI and fNIRS, is encouraged to elucidate the neural mechanisms underlying the improvement of gross motor skills through exercise interventions (62). Additionally, future studies are advised to employ prospective trial registration, use standardized gross motor assessment tools, and actively publish negative or neutral findings to enhance the transparency and comparability of the evidence.