This systematic review and meta-analysis followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (Z. The protocol of this review was registered on PROSPERO (ID = CRD42024534097).

We systematically searched four databases—Web of Science, PubMed, Cochrane Library, and IEEE Xplore—from their inception until December 2, 2024. This review only included randomized controlled trials (RCTs) investigating the effects of video games on executive functions or motor skills in children and adolescents with NDDs. The search strategy utilized a combination of subject headings and keywords. This was supplemented by manual searches and reviewing the reference lists of included studies where necessary. Search terms included “video games,” “active video gaming,” “exergaming,” “computer Games,” “autism spectrum disorder,” “attention deficit disorder with hyperactivity,” “motor skills disorders,” “developmental coordination disorder,” “specific learning disorder,” “neurodevelopmental disorders,” “executive function,” “inhibitory control,” “cognitive flexibility,” “working memory,” “inhibition,” “shifting,” “updating,” “cognitive control,” “gross motor skills,” “fine motor skills,” and “motor skills,” etc. Please refer to Supplementary Material 1 for the specific search strategy.

This study followed the PICOS framework to define the literature screening criteria, as detailed below: Inclusion criteria: (1) P (Population): Participants must be children and adolescents aged 3–18 years with NDDs diagnosed using either the Diagnostic and Statistical Manual of Mental Disorders (DSM-5 or DSM-IV), the International Statistical Classification of Diseases and Related Health Problems, 10th Revision, or other disorder-specific diagnostic criteria; (2) I (Intervention): Studies employing standalone video game interventions or combined interventions were eligible. For combined interventions, the video game component had to constitute the primary element of the intervention, with the time dedicated to gaming being equal to or greater than that for other components; (3) C (Comparison): The control group should involve no video game intervention or utilize video games that do not exert any influence on the outcome indicators; (4) O (Outcomes): The outcome indicators include executive functions or motor skills; (5) S (Study design): Only RCTs are eligible.

Exclusion criteria: (1) Article published in languages other than Chinese and English; (2) Article with poor quality assessment; (3) Inability to directly or indirectly obtain the mean values and standard deviations of the required outcome indicators.

Following the initial search, MG, JR, and QW screened titles and abstracts. Potentially relevant records then underwent full-text review against the eligibility criteria. JR and QW independently screened articles against eligibility criteria. Discrepancies were resolved through discussion, with MG making final decisions if consensus was not reached. When necessary, AK contacted corresponding authors to request missing data. Studies with irrecoverable incomplete data were excluded.

For studies with multiple outcome measures, we applied a pre-specified prioritization hierarchy to select the primary outcome for extraction, as outlined in the study protocol. This hierarchy was based on the validation status, reliability, and frequency of use in the literature for each NDD population. The highest priority was given to validated and standardized tools with strong psychometric properties in the target population [e.g., Movement Assessment Battery for Children (MABC) for motor skills, Behavior Rating Inventory of Executive Function (BRIEF) for executive functions]; the next priority was assigned to other widely used, validated scales with published reliability and validity data. This pre-specified approach aimed to maximize methodological consistency and minimize measurement heterogeneity across studies.

MG, JR, and QW extracted data using a standardized form, including: (1) study characteristics (author, year, and sample size); (2) demographic characteristics (mean age, diagnosis); (3) intervention details (game name, platform, intervention parameters: duration, session length, frequency); (4) assessment methods. We extracted the mean and standard deviation for the primary outcome measures from both the experimental and control groups at post-test and follow-up time points.

We conducted a meta-analysis of the included continuous variables using Review Manager 5.3 software, calculating the Standardized Mean Differences (SMD) and 95% confidence intervals, with a significance level (α) set at 0.05. Effect sizes were categorized as small, medium, and large, corresponding to 0.2, 0.5, and 0.8, respectively (27). Heterogeneity was quantified using the I² statistic. In cases where I² was less than 50%, the studies were considered homogeneous, and a fixed-effects model was used for analysis; otherwise, a random-effects model was employed. Sensitivity analyses were performed for each indicator-covered study to verify the robustness of the results. To explore potential sources of heterogeneity, subgroup analyses were conducted using a random-effects model.

The quality of the included studies was assessed using the methods recommended by the Cochrane Handbook for Systematic Reviewers, and seven aspects of trial methods that may introduce bias were assessed (sequence generation, allocation concealment, blinding of participants and investigators, evaluators’ blinding, completeness of result data, selective reporting of results, and other biases). As the number of studies included in the meta-analysis for gross motor skills (n = 11) exceeded ten, we assessed publication bias for this outcome. Funnel plots were visually inspected, and Egger's test was performed using Stata 16.0 to assess the risk of publication bias.

Figure 1 illustrates the screening process of the included studies. During the full-text review stage, the primary reasons for exclusion were missing data (n = 10), irrelevant outcomes (n = 19), and non-RCTs (n = 10, including within-subjects designs (28, 29) and pretest-posttest designs (30–32). After rigorous screening, we ultimately included 20 RCTs.

The specific characteristics of the included studies are presented in Table 1.

The results of the literature quality evaluation are presented in Figures 2, 3. Among the 20 included articles, several studies failed to implement double-blinding due to practical difficulties. Figure 4 displays the assessment results of publication bias for studies related to gross motor skills, indicating the presence of publication bias risk (t = 2.28; p < 0.05; 95% CI: 0.024–7.061). Subsequently, we conducted a trim-and-fill analysis to adjust for this bias, and the outcomes are presented in Figure 5. Following the trim-and-fill analysis, the effect size for the improvement in gross motor skills was adjusted from the original g = 0.502 (95% CI: 0.047–0.957) to g = 0.710 (95% CI: 0.234–1.186). This adjustment underscores that despite the presence of publication bias, the positive impact of video game training on gross motor skills remains robust.

Disrupted integration and dynamic instability within the Prefrontal cortex–parietal–subcortical network (involving structures like the caudate nucleus and cerebellum) constitute the core neural basis of EF deficits in children with NDDs. For example, children with ADHD show reduced activation in key nodes of this network (e.g., the inferior frontal gyrus) during task performance, which is directly linked to impairments in response inhibition and WM (51). Children with ASD exhibit abnormalities in prefrontal-cerebellar connectivity and DMN dysfunction, affecting cognitive flexibility (52). Building on this neural mechanism, video game-based interventions can enhance EFs such as attention and inhibitory control by boosting the reward system (e.g., dopaminergic pathways) and promoting plasticity within the aforementioned neural networks (53–55).

Empirical studies show that video game-based interventions outperform control conditions in children and adolescents with ADHD and ASD, a finding supported by systematic reviews. For example, Khan (56) and Ikezawa (57) observed improvements across multiple EF domains in NDDs children in reviews of internet-based interventions including video games. Among 12 related trials, only one reported no significant improvement (38), likely due to low adherence, as only 66.5% of participants completed more than 20 training sessions. Overall, the evidence supports video games as an intervention to improve EF in children with NDDs.

In terms of motor skills, functional deficits in children with NDDs primarily stem from developmental abnormalities in the sensorimotor cortex-cerebellum-basal ganglia circuitry, which impair motor planning, execution, and feedback integration, affecting both gross and fine motor coordination (58, 59). Specifically, children with ASD commonly exhibit cerebellar structural abnormalities and compensatory activation of fronto-striatal circuits (58), whereas children with DCD and ADHD display disrupted interhemispheric connectivity within motor networks, with DCD characterized by a loss of right putamen connectivity dominance (59). These neural alterations form the core basis of motor impairments in NDDs.

Based on this, AVGs exert a positive influence on gross motor skills through full-body motion feedback. Multiple studies support their effectiveness in improving gross motor abilities in children and adolescents. Although neuroimaging evidence specifically for active video games remains limited, other research (e.g., fMRI studies in children with cerebral palsy) indicates that motor training can modulate activation in the fronto-parietal and sensorimotor regions. These neuroplastic changes are closely correlated with behavioral improvements (60). Given that AVG training similarly emphasizes repetition, full-body engagement, and task-oriented practice, it is reasonable to hypothesize that its effects also depend on the plasticity of motor control networks.

In contrast, the efficacy of AVGs in improving fine motor skills appears limited, with only one out of six studies reporting superior performance in hand agility training compared to conventional interventions (36). This limitation may stem from the spatial precision constraints of current mainstream motion-sensing technologies, which hinder their ability to capture fine hand movements with high fidelity. Research indicates that commonly used devices such as Kinect and commercial VR controllers are designed primarily for large-scale limb interactions and exhibit insufficient accuracy in detecting finger-level motions (61, 62). Consequently, these technological limitations likely restrict the potential of AVGs for fine motor training tasks such as grasping and pinching.

In addition to the aforementioned technical limitations of the intervention itself, this study also has certain methodological limitations. The wide age range of participants (3–18 years) increases generalizability but also introduces substantial developmental heterogeneity. According to hierarchical neurodevelopmental theory, basic sensorimotor networks mature relatively early, whereas frontal circuits responsible for higher-order cognitive-motor integration continue developing until late adolescence (63). This aligns with the nonlinear developmental trajectory of the prefrontal cortex: executive functions develop rapidly between middle childhood and mid-adolescence, accompanied by synaptic pruning that optimizes network efficiency and stabilizes by late adolescence (64, 65). Therefore, pooling individuals at different developmental stages in a single analysis may obscure stage-specific intervention effects. Future studies should consider stratification by developmental stage or combine neurodevelopmental features for subgroup analyses to more precisely elucidate intervention outcomes.

Moreover, the majority of the meta-analyses summarized above had small sample sizes. This makes the pooled estimates particularly susceptible to small-study effects and potential publication bias, meaning the observed effect sizes may be inflated. Therefore, while promising, these results should be considered preliminary and require confirmation in larger, more robustly designed trials.

Given the substantial heterogeneity among the included studies in terms of diagnosis, intervention protocols, and other factors, this subgroup analysis aimed to explore the influence of key variables (such as video game type, dosage, and frequency) on effect size variability, rather than to identify a single “optimal” protocol. It should be emphasized that this analysis was an exploratory post-hoc assessment; each subgroup included only 3–6 studies and involved multiple comparisons, resulting in limited statistical power and an increased risk of false-positive findings. Therefore, the results should be interpreted with caution.

Specifically, for cognitive flexibility, subgroup analyses based on game type, intervention frequency, and session duration generally reduced within-group heterogeneity, with AVGs appearing to yield more pronounced improvements than SVGs. This indicates that this outcome may show a certain degree of consistency across studies, and its effect size could be influenced by intervention characteristics. This may be because AVGs, by integrating physical movement with cognitive tasks, co-activate the sensorimotor cortex, prefrontal executive control network, and cerebellar–basal ganglia circuitry. Such multi-system engagement is particularly beneficial for cognitive flexibility, which depends on the dynamic interaction between the dorsolateral prefrontal cortex and the dorsal anterior cingulate cortex (66). Moreover, the high metabolic demands of the prefrontal cortex may explain why lower-dose AVG interventions (≤3 sessions/week, ≤30 min/session) still produced considerable effects in our pooled data.

In contrast, inhibitory control exhibited moderate overall heterogeneity and was insensitive to the examined dosage and format variables, with within-subgroup heterogeneity not systematically decreasing. This may reflect its neural substrate, which is primarily governed by the structurally stable right inferior frontal gyrus, conferring relatively limited plasticity in response to variations in training parameters (67).

The analysis of working memory and gross motor skills revealed more complex patterns. For working memory, subgroup analysis indicated that SVGs and longer single-session durations (>30 min/session) were associated with improved outcomes, aligning with the mechanism whereby SVGs primarily engage the dorsolateral prefrontal–parietal working memory network and rely on deep, sustained engagement in processes such as rule memorization and strategic planning (68, 69). For gross motor skills, short-term interventions (≤6 weeks) showed higher effect sizes, possibly reflecting the rapid progression phase in the early stages of motor skill acquisition.

However, residual within-subgroup heterogeneity remained for both outcomes, suggesting that effect variability may also be influenced by other methodological and contextual factors, such as differences in sample characteristics, inconsistencies in measurement tools, and specific details of intervention implementation. Therefore, a single clinical variable is insufficient to fully account for these effect patterns, and future research should further control for these potential sources of variability.

In summary, for outcomes with relatively high consistency, such as cognitive flexibility, the findings provide limited yet meaningful support for cross-diagnostic data pooling under a random-effects model. However, substantial residual heterogeneity remains for outcomes such as working memory and gross motor skills, making it difficult to directly generalize the pooled effects to specific diagnostic subtypes or intervention protocols. Therefore, the pooled estimates presented in this report should be interpreted as reflecting overall trends rather than serving as direct clinical prescriptive recommendations.

Several limitations of this study warrant consideration when interpreting the results: 1) a limited number of included studies with generally small sample sizes, and potential publication bias across all outcome domains; 2) the heterogeneity in diagnoses and intervention protocols among the included studies limits the specificity and direct clinical generalizability of the pooled results; 3) absence of data on the critical variable of exercise intensity; 4) a notable lack of long-term follow-up data, which prevents the assessment of effect sustainability and stability; 5) Due to insufficient data, stratified analyses based on specific NDD subtypes and developmental stages could not be conducted. Future research should focus on standardizing intervention dosage, enhancing intensity monitoring, expanding sample sizes, employing more consistent assessment tools, and prioritizing the collection of long-term follow-up data. This will allow for a deeper investigation into the persistence and stability of intervention effects, as well as their transfer to real-world situations and related cognitive tasks.