The review will focus on patients of any age diagnosed with ASD using standardized criteria (eg, the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition; International Classification of Diseases, 10th Revision; or their latest revisions, the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, Text Revision and International Classification of Diseases, 11th Revision). Given that the critical window for cognitive development in ASD is childhood and intervention responses vary significantly across developmental stages (eg, infants vs adolescents vs adults), age will be treated as a primary factor in our analytical framework.

To minimize confounding effects on cognitive assessment, studies involving participants with specific conditions will be excluded. These exclusions comprise individuals with severe neurological comorbidities, such as active epilepsy defined as 2 or more seizures in the previous 3 months, cerebral palsy, or known structural brain abnormalities. We will also exclude participants with severe intellectual disability, characterized by an IQ below 50 as measured using standardized scales such as the Wechsler Intelligence Scales.

Furthermore, individuals with major psychiatric disorders, including schizophrenia or bipolar disorder, and those with significant sensory impairments, such as severe visual or hearing loss that would preclude valid cognitive testing, will be ineligible.

Finally, the review will exclude patients with severe physical illnesses or unstable health conditions, such as uncontrolled cardiovascular or pulmonary disease, liver or kidney failure, malignant tumors, or the active phase of severe autoimmune diseases.

Textbox 1 summarizes the specific inclusion and exclusion criteria for participants designed to ensure a focused assessment of massage therapy’s effects on cognitive function.

The presence of these excluded conditions, such as severe neurological comorbidities and intellectual disability, may independently and profoundly impact cognitive function, thereby obscuring the specific intervention effect of tuina and massage therapy.

For the experimental group, massage therapy is defined as any systematic manual manipulation of soft tissues, including tuina, acupressure, or Swedish massage, performed by a trained professional for therapeutic purposes to provide a comprehensive evidence map of manual therapies for ASD cognitive symptoms. To ensure consistency and clinical relevance, studies must meet minimum operational criteria for tuina as recommended by the technical specifications for pediatric tuina in medical institutions [32].

Specifically, the intervention must be described as “tuina” or “pediatric tuina” and incorporate fundamental maneuvers defined in the specifications, such as kneading, pressing, pushing, and rubbing performed by a trained practitioner (eg, tuina therapist or physician) on specific body regions (eg, head, back, and limbs) or acupoints (eg, Baihui DU20 and Sishencong EX-HN1).

Furthermore, each treatment session should typically last between 15 and 30 minutes to align with standard clinical practice, and studies must clearly report the treatment frequency and total intervention period to facilitate subgroup analysis of the dose-effect relationship.

Other massage therapies (eg, Swedish massage, Thai massage, shiatsu massage, and acupressure) will also be included provided they offer clear descriptions of the techniques applied. These interventions will be categorized into TCM based (eg, tuina and acupressure) and Western based (eg, Swedish and myofascial release) for subsequent analysis. Studies lacking a detailed description of the manual procedures will be categorized as insufficiently described tuina for the purpose of subgroup analysis.

Massage therapy could be administered as a stand-alone treatment or as an adjunct to conventional care (eg, behavioral therapy and special education). Cointerventions must be reported and will be accounted for in the analysis.

The control groups will include any of the following: no treatment, waitlist, placebo or sham interventions (eg, light touch or simulated massage without therapeutic techniques), or usual care or standard of care (eg, routine behavioral therapy and special education).

Studies using other active interventions (eg, pharmacological treatments) as a comparator will also be included provided the experimental group receives massage therapy as an adjunct to the same standard care.

Studies in which massage therapy is administered to the control group will be excluded.

The comprehensive literature search was conducted from database inception to September 30, 2025, across 5 English-language databases and 4 Chinese-language databases. English-language resources included PubMed, Web of Science, Scopus, Cochrane Library, and Embase, whereas Chinese-language resources encompassed China National Knowledge Infrastructure (CNKI), Wanfang Data, VIP Chinese Scientific Journal Database, and the Chinese Biomedical Database.

To identify ongoing or recently completed trials, we systematically searched clinical trial registries, specifically ClinicalTrials.gov and the Chinese Clinical Trial Registry.

The search identified varying numbers of records across these platforms, including 8 results from PubMed, 37 from Web of Science, 24 from Embase, 161 from the Cochrane Library, 49 from ClinicalTrials.gov, 100 from the Chinese Biomedical Database, 51 from CNKI, 78 from Wanfang Data, and 86 from VIP Chinese Scientific Journal Database. A detailed representative search strategy for PubMed is presented in Table 1 for reference.

The search terms integrated 3 primary concepts, including ASD-related terms, massage-related interventions such as tuina and manual therapy, and cognitive-related outcomes, combined with specific filters for RCTs.

Furthermore, to minimize publication bias and identify ongoing or recently completed studies, we conducted supplementary searches for gray literature, including conference proceedings from major international and domestic conferences in the fields of autism research and complementary medicine, such as the International Meeting for Autism Research and key Chinese academic conferences on tuina or manual therapy. Academic dissertations were searched within the thesis repositories of the aforementioned Chinese and English databases, including CNKI, Wanfang Data, and ProQuest Dissertations and Theses Global. The reference lists of all included studies and relevant systematic reviews were manually scanned for additional eligible studies.

The study selection process will begin with a pilot-testing phase conducted to ensure consistent interpretation of the eligibility criteria among reviewers and refine the screening criteria prior to formal screening. During this phase, a random sample of 20 citations from the search results will be independently screened by 2 reviewers based on titles and abstracts. The interrater agreement will be calculated using the Cohen κ statistic.

A κ value of 0.75 or higher will be considered indicative of substantial agreement and sufficient for proceeding to formal screening. If the κ value is below 0.75, the reviewers will discuss the discrepancies, receive additional training on the eligibility criteria, and refine the screening guide if necessary. This pilot process will be repeated with a new sample until the satisfactory agreement level is achieved.

Following the successful pilot test, the study selection process will be performed independently by 2 reviewers. They will first screen the titles and abstracts of all retrieved records against the eligibility criteria using the Rayyan (Qatar Computing Research Institute) or EndNote (Clarivate Analytics) software. The full texts of potentially relevant studies will then be retrieved and assessed for final inclusion. Any discrepancies at any stage will be resolved through discussion or, if necessary, adjudication by a third senior reviewer.

Data extraction will be performed using a standardized, prepiloted data extraction form that will be developed in Microsoft Excel. Two reviewers will independently extract data from the included studies, with any discrepancies resolved through consensus or consultation with a third reviewer. The extracted information will encompass study characteristics such as the first author, publication year, country, sample size, and study design, as well as participant details (age, gender, diagnostic criteria, baseline severity, and comorbidities).

Regarding intervention and comparator details, the reviewers will record the type of massage or tuina, specific techniques, acupoints, treatment duration, session frequency, total sessions, and cointerventions (including the systematic recording of the type, frequency, and intensity of all concurrent interventions reported in each study).

Outcome data extraction will involve collecting means, SDs, or change scores from baseline for all primary and secondary outcomes at all reported time points, as well as the number of events for dichotomous outcomes. Furthermore, the specific cognitive assessment tool or specialized scale used for each relevant outcome in each study will be systematically recorded to ensure a nuanced evaluation of cognitive domains.

Finally, safety data concerning the type and incidence of adverse events will be documented, and in instances in which critical data are missing or unclear, the corresponding authors will be contacted via email twice over a 2-week period to request the information.

Statistical analyses will be performed using RevMan (version 5.4; The Cochrane Collaboration) and Stata (version 17; StataCorp).

For continuous outcomes (eg, ABC and CARS scores), the mean difference or standardized mean difference with 95% CIs will be calculated [33]. We prefer using change-from-baseline scores (with SDs of the change) for continuous outcomes; if unavailable, posttreatment scores will be used, and a sensitivity analysis will be performed to assess the impact of this choice.

For dichotomous outcomes (eg, efficacy rates), the odds ratio with 95% CIs will be used.

The decision to perform a meta-analysis will depend on both clinical and statistical heterogeneity. We will first evaluate the clinical similarity of the included studies. A quantitative synthesis (meta-analysis) will only be conducted if the clinical heterogeneity is deemed acceptable. If the interventions are considered too diverse (eg, vastly different techniques or contexts) or if the I² value remains high despite subgroup investigations, a descriptive qualitative synthesis will be provided instead of a pooled effect size. The fixed-effects model will be considered only if there is minimal statistical heterogeneity.

If substantial heterogeneity (I²≥50%) is detected, we will perform subgroup analyses based on several prespecified factors to explore the potential sources of variability. These factors will include participant age tiers, specifically early childhood (<6 years), school age (6-11 years), and adolescence or adulthood (≥12 years) [34]. Additionally, we will examine the type of massage therapy by comparing TCM-based vs Western-based manual therapies and categorize outcomes by specific treatment durations. Outcomes will also be analyzed separately according to time point intervals, specifically immediately after treatment, short-term follow-up (≤3 months after treatment), and medium- or long-term follow-up (>3 months after treatment). Finally, the analyses will account for the comparator type, such as usual care vs placebo, and the intervention context, identifying whether massage therapy was administered as a stand-alone treatment or as an add-on therapy to usual care.

If significant heterogeneity (I²≥50%) persists after the aforementioned subgroup analyses, we will conduct univariate random-effects meta-regression to quantitatively explore the sources of heterogeneity. The following categorical variables, derived from the subgroup analyses, will be tested as potential moderators: patient age, categorized as less than 6 years vs 6 years and above; the total treatment course, categorized as less than 8 weeks vs 8 weeks and above; and the type of control intervention, such as usual care vs placebo or sham therapy. Additionally, the proportion of participants receiving a specific concurrent therapy will be evaluated as a moderator to further investigate the robustness of the findings.

Statistical heterogeneity will be assessed using the I² statistic and the Cochran Q test (with a significance level of P<.10). Given the inherent clinical diversity in massage techniques (eg, TCM based vs Western based), treatment durations, and participant characteristics across the included trials, a random-effects model will be applied as the primary analytical approach for all outcomes. This model is preferred as it accounts for both within-study and between-study variability, providing more conservative and generalizable estimates. The fixed-effects model will only be used as a sensitivity analysis to test the robustness of the results when statistical heterogeneity is absent (I²=0%) and the studies are deemed sufficiently similar in their population, intervention, comparator, and outcome elements.

Subgroup, sensitivity, and meta-regression analyses will be performed to explore heterogeneity. Following Cochrane recommendations, subgroup analyses and meta-regression will only be performed if at least 10 studies are available for the specific outcome to avoid spurious findings. Results from analyses with fewer than 10 studies will be interpreted with extreme caution, focusing primarily on descriptive presentation.

To test the robustness of the primary meta-analysis results, sensitivity analyses will be conducted by sequentially excluding studies based on specific criteria and comparing the pooled effect sizes. Specifically, we will exclude studies judged to have a high risk of overall bias, those with a single-group sample size of fewer than 20 participants, and those with a follow-up loss rate greater than 15%. Additionally, we will examine the robustness of the overall findings by including only studies that use the most frequently reported massage modality, such as tuina only or Swedish massage only, depending on the identified evidence base. Finally, studies involving adult populations (aged ≥18 years) will be excluded to assess whether the findings remain robust specifically for the pediatric and adolescent ASD population.

If 10 or more studies are included in a meta-analysis, publication bias will be assessed visually using funnel plots and statistically using the Egger linear regression test.

If funnel plot asymmetry is suggested and the Egger test indicates statistical significance (P<.05), the nonparametric trim-and-fill method will be used to estimate the number of potentially missing studies and compute an adjusted effect size. The original and adjusted effect estimates will be compared to evaluate the potential impact of publication bias on the results.