The study strictly followed the PRISMA Statement (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) (30), ensuring a rigorous and standardized reporting methodology. The included literature was thoroughly compiled and analyzed in accordance with the standards outlined in the International Systematic Review Writing Guidelines (31).

A comprehensive literature search was conducted across multiple databases, including Embase, PubMed, Web of Science (WOS), The Cochrane Library, Wanfang Data, VIP Information, and China National Knowledge Infrastructure (CNKI). The search encompassed all publications from the inception of each database through July 31, 2024. Two researchers (YS and SJ) independently performed the literature retrieval. In instances of disagreement, a third researcher was consulted to help resolve the issue and reach a consensus. To ensure a thorough and comprehensive search, the included literature and relevant review references were also examined. The detailed search strategy is provided in Supplementary File S1.

The inclusion criteria were as follows: (1) Studies involving patients aged ≤18 years and diagnosed with ADHD by professionals based on recognized criteria (Diagnostic and Statistical Manual of Mental Disorders or other relevant ADHD diagnostic scales) (32, 33) were included. No restrictions were set on the subtype of ADHD, gender, family background, economic status, level of intelligence, country, or ethnicity of origin of ADHD. (2) The experimental group underwent only exercise intervention, which could include either single or combined interventions of exercise. The control group received usual care and health education, with no additional interventions. (3) Only RCTs were included in the analysis. (4) The mean and standard deviation of anxiety, depression, and emotional regulation were obtainable directly or indirectly in the experimental and control groups before and after the intervention. (5) The baseline data exhibited no significant differences between the experimental and control groups.

The exclusion criteria were as follows: (1) Studies involving patients with autism, Tourette syndrome, epileptic disorders, or other comorbid conditions, (2) studies in which the intervention that did not involve a singular or combined exercise program, (3) studies with incomplete or unextractable data, (4) reviews, commentaries, case studies, cohort studies, book chapters, or duplicate publications, (5) studies with poor methodological quality [Physiotherapy Evidence Database (PEDro) scores < 4] (34), and (6) non-randomized controlled trials.

Anxiety is an emotional disorder primarily characterized by excessive worry, restlessness, fear, and physical symptoms, such as palpitations, sweating, and insomnia (35). Assessment instruments for anxiety include the Behavioral Assessment System for Children-T (BASC-T), Child Behavior Checklist (CBCL), Conners Rating Scales (CRS), Youth Self-Report Scales (YSRS), and Beck Anxiety Inventory (BAI).

Depression is characterized by a persistent feeling of sadness and/or a diminished interest or pleasure in activities. In children with ADHD, depression frequently manifests as irritability rather than low mood. It often co-occurs with anxiety, presenting more physical discomfort, and may include psychiatric symptoms such as hallucinations (36). Several assessment tools were utilized to evaluate depression in children, including the CRS, BASC-T, CBCL, and Children's Depression Inventory (CDI).

Emotional dysregulation symptoms in ADHD are frequently characterized by irritability, frustration, and rapid mood swings. The intensity or frequency of these symptoms often appears disproportionate to the individual's environment, age, and developmental stage (37). Various assessment tools were utilized to evaluate emotional dysregulation, including the CRS, Profile of Mood States (POMS), Short Mood and Feelings Questionnaire (SMFQ), Kinder Lebensqualität Fragebogen (KINDL), and Pediatric Quality of Life Inventory (PedsQL).

Two authors (YS and SJ) independently screened the studies and extracted data based on predefined inclusion and exclusion criteria. In cases of disagreement, the authors discussed and reached a consensus. If consensus could not be achieved, a third author (XW), blinded to the screening process, made the final decision on study inclusion. The following data were extracted: basic study characteristics (author, publication year, country), subject information (sample size, age), intervention details (specific measures for both experimental and control groups), intervention dosage (time, frequency, duration), and main outcome indicators. When relevant statistics were not reported, the means and standard deviations were estimated based on the sample size, median, range (minimum and maximum), and interquartile range (38). All studies included in this research were published in peer-reviewed journals. In cases where original data were presented only in visual formats (e.g., pictures) or remained unavailable despite contacting the original authors via email, the data were extracted using a chart extraction tool. If a single study employed two or more assessment methods for outcome indicators, only the results of the most commonly used method were extracted, or subgroup data were combined to minimize the risk of double counting and reduce anomaly weighting. The Cohen's kappa value for both researchers was 0.94, indicating a high level of agreement.

Two assessors (YS and MQ) evaluated the quality of the included studies utilizing the PEDro scale (39) and assessed the risk of bias in the studies employing the Cochrane risk-of-bias tool (latest version, ROB-2) (40).

The PEDro scale an 11-item assessment tool used to evaluate the methodological design and statistical reliability of clinical studies. The “eligibility criteria” item is excluded from the scoring system. Each item is scored with 1 point if it meets the specified criterion, and0 points if it does not. The total score was used to categorize studies into quality levels: scores below 4 points indicated poor quality, scores between 4 and 5 represented fair quality, scores between 6 and 8 were deemed good quality, and scores between 9 and 10 were classified as excellent quality. The Cochrane ROB-2 tool complemented the PEDro scale by assessing biases in the design and outcomes of RCTs. This tool categorizes studies as having “low risk,” “some concern,” or “high risk” of bias.

The GRADE profiler system (41) was employed to assess the quality of evidence for the outcome indicators. This assessment incorporates five downgrading factors: publication bias, inconsistency, imprecision, indirectness, and risk of bias. The evidence was classified into four levels: high (no downgrading), moderate (downgraded by one level), low (downgraded by two levels), and very low (downgraded by three levels). Two researchers independently conducted the quality assessment. In cases of discrepancy, a third researcher was consulted to reach a consensus through discussion.

Stata 18 software was employed for effect size combination, subgroup analysis, sensitivity analysis, and publication bias testing. The standardized mean difference (SMD) and 95% confidence interval (CI) were utilized to combine effect sizes, accounting for variations in testing methods and measurement units. SMD values were categorized as fair (<0.2), small (0.2–<0.5), moderate (0.5–<0.8), and large (≥0.8). Heterogeneity was assessed using Higgins' I² statistic, with I² ≤ 25% indicating no heterogeneity, 25% < I² ≤ 50% low heterogeneity, 50% < I² ≤ 75% moderate heterogeneity, and I² > 75% high heterogeneity. Studies were considered homogeneous and analyzed using a fixed-effect model if I2 < 50% and p > 0.1. Further determination of heterogeneity sources was necessary if I2 ≥ 50% and p < 0.1. Sensitivity analyses were conducted when I² exceeded 50% to assess the robustness of merged results. Publication bias was evaluated using Stata 18.0 software, employing funnel plots and Egger's test. The “trim and fill” method was used to further evaluate the stability of meta-analysis results if publication bias was indicated. Publication bias was not assessed for fewer than five studies due to low test efficacy. The significance level for the meta-analysis was set at p ≤ 0.05.

A comprehensive literature search yielded 7,743 articles from various databases: Embase (n = 706), PubMed (n = 1,151), WOS (n = 2,142), The Cochrane Library (n = 3,113), Wanfang Data (n = 271), VIP Information (n = 41), and CNKI (n = 319). An additional 25 articles were manually identified. Following the initial screening of titles and abstracts, 65 papers were selected for further evaluation. The subsequent full-text review resulted in the exclusion of 47 articles due to various reasons: Twelve lacked relevant endpoint data, two did not meet endpoint criteria, nine employed non-conforming intervention methods, four were duplicates, two were conference publications, six were review articles, ten did not meet age-related inclusion criteria, and two did not meet population-related inclusion criteria. Ultimately, eighteen articles were included in the final analysis. Figure 1 provides a detailed illustration of the literature screening process.

The analysis included 18 RCTs (27–29, 42–56), encompassing 830 participants. The subjects were predominantly children diagnosed with ADHD, ranging from 5 to 14 years of age, with studies published between 2004 and 2023. All included studies implemented an exercise intervention for the experimental group, while the control group received conventional treatments or reading interventions. The physical exercise interventions were primarily classified into two categories: monotypic exercise interventions (55.6%) and mixed exercise interventions (44.4%). Monotypic exercise modes involve a single form of physical activity, whereas mixed exercise modes incorporate multiple forms. The control group underwent conventional rehabilitation, medication, social-behavioral interventions, and a structured educational approach combined with auditory integration training. Intervention frequency varied from 1 to 7 sessions per week, spanning 1 to 20 weeks, with individual sessions lasting between 10 and 90 min. Exercise intensities were categorized as low intensity (44.4%), moderate intensity (33.3%), and high intensity (22.3%). No adverse events were reported in any of the included studies. Additional details are provided in Table 1.

All 18 RCTs included in this review met the following criteria: “participant selection criteria specified,” “random assignment of participants to groups,” “baseline similarity,” “participation rate >85%,” “intention-to-treat analysis,” “analysis of statistical results between groups,” and “the study reported point and variability measures for at least one main outcome,” as detailed in Table 2. Among these, seven studies (27, 45, 47–50, 52) achieved “allocation concealment,” one study implemented blinding for outcome assessment, and none reported a “participation rate ≤15%.” Regarding the PEDro scale, the 18 studies scored between 6 and 8, with a mean score of 6.4. No studies of low methodological quality were identified, suggesting that the overall methodological quality was satisfactory.

Based on the ROB-2 assessment tool, seven studies clearly delineated the randomization process and were categorized as low risk for bias. Six studies adequately described the intended intervention and were consequently deemed at low risk of bias. Eighteen studies demonstrated no bias due to missing outcome data, fifteen studies showed no bias in outcome measurement, and fourteen studies exhibited no bias in outcome reporting selection. The final quality ratings for the included literature were as follows: B for eighteen documents, with no documents receiving A or C ratings (Figure 2). Two independent reviewers (SY and SJ) independently assessed the quality of the literature during the evaluation process.

To investigate the source of heterogeneity, a sensitivity analysis was performed using Stata 18.0. As illustrated in Figure 4, the combined effect was analyzed by sequentially excluding individual studies. After removing individual studies, the results for anxiety revealed SMD (−0.66 to −0.49), I² (53.9%–63.9%), and p < 0.05; for depression, SMD (−0.40 to −0.66), I² (0.0%–61.4%), and p < 0.05; and for emotion regulation, SMD (0.67 to 1.24), I² (85.3%–91.7%), and p < 0.05. The analysis demonstrated no significant changes before and after the removal of individual studies. The low sensitivity of the study data indicates that the results exhibit a considerable level of reliability and stability. These findings suggest that the combined analysis did not result in substantial alterations, thus demonstrating a degree of robustness.

Egger's test revealed no significant differences in anxiety (Z = 1.39, p > |t| = 0.197 > 0.05), depression (Z = −1.11, p > |t| = 0.317 > 0.05), or emotion regulation (Z = 2.54, p > |t| = 0.052 > 0.05). These results indicated that the findings were relatively robust (Figure 5).

The quality of evidence was evaluated utilizing the GRADE profiler methodology. Factors such as publication bias, inconsistency, imprecision, indirectness, and risk of bias were not downgraded. The assessment of the quality of evidence regarding the impact of physical activity on ameliorating anxiety, depression, and emotion regulation in children with ADHD was determined to be of high level, as illustrated in Figure 6.