This study conducted a search for articles in January 2025 and updated the findings in February 2025 using the following electronic databases: PubMed, Web of Science, the Cochrane Library, Google Scholar, and the China National Knowledge Infrastructure (CNKI). The search terms included “table tennis,” “T cell,” “B cell,” and other relevant terms in both Chinese and English. The complete search strategy with detailed search strings, Boolean operators, MeSH terms, and filters for each database is provided in Supplementary File S1. The search terms included “table tennis,” “T cell,” “B cell,” and other relevant terms in both Chinese and English. There were no restrictions on time or language during the search process. Only eligible English and Chinese studies were considered for this review. This review was conducted in accordance with the guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA). Three independent authors reviewed the studies at each stage and managed them according to the PRISMA flowchart (Figure 1). They then determined whether inclusion was appropriate. Any disagreements among the three researchers were resolved by consensus to decide on the inclusion of the study.

The tests for CD3, CD4, CD8, CD4/CD8 ratio, IgA, IgG, IgM, TNF-α, IL-6, IFN-γ, CK, LDH, and BUN were evaluated as criteria for inclusion in the results. This review focused on studies investigating the effects of swinging exercise training. All types of swinging exercises, irrespective of intensity, frequency, or duration, were deemed eligible.

The inclusion criteria for this review were as follows: (1) cross-trial studies were included to draw evidence-based conclusions regarding immune system-related outcomes. (2) The study must categorize the included researchers into at least two groups: one group that engages in swinging exercises and another that serves as a non-intervention group. Additional control groups, such as a conventional exercise group, may be present; however, their data will not be included in the meta-analysis. (3) Exercise must be conducted without any other interventions.

Exclusion criteria were as follows: (1) non-original articles (including editorial letters, short reports, case studies, methods papers, review articles, and systematic reviews and meta-analyses); (2) animal studies; (3) studies without an exercise group; (4) studies without a control group; (5) studies that include both exercise and control groups; (6) dietary intervention studies; (7) studies that do not provide information on the association between the intervention and the selected outcomes; and (8) duplicate publications or sub-studies of included trials.

Any disagreements were discussed and resolved among all team members. The data features of each study were extracted as follows: (a) first author and year; (b) country; (c) participant characteristics, including sample size and age; (d) interventions; (e) experimental cycle; and (f) results, including CD3, CD4, CD8, CD4/CD8 ratio, IgA, IgG, IgM, TNF-α, IL-6, IFN-γ, CK, LDH, and BUN. Pre-test and post-test values (mean and standard deviation) were entered into the meta-analysis to generate forest plots. If the mean and standard deviation were not reported, they were calculated based on the standard error, median, range, and interquartile range.

The methodological quality of the included studies was assessed using the Cochrane Risk of Bias Tool (RoB 2) for randomized trials and the ROBINS-I tool for non-randomized studies. The assessment included the appropriate use of random sequences, allocation concealment, participant blinding, assessor blinding of results, and a clear description of losses and exclusions. A detailed summary of the risk of bias assessment for each study across all domains is provided in Supplementary File S2, including the rationale for each judgment. When these processes are documented in published reports, the studies are classified as “Criterion fulfilled.” Studies that do not meet these criteria are classified as “Criterion not fulfilled.” Studies that did not report these data were categorized as “not applicable.” The quality assessment was conducted independently by three reviewers, and any disagreements were resolved through consensus or by consulting a fourth reviewer.

All statistical analyses were performed using R software (version 4.5.1) with the meta, metafor, and dmetar packages. For continuous outcomes, the standardized mean difference (SMD) with 95% confidence intervals (CIs) was calculated as the summary effect measure. The random-effects model (DerSimonian–Laird method) was used to account for anticipated heterogeneity among studies. Heterogeneity was assessed using the I² statistic, where an I² value >50% indicated substantial heterogeneity. Univariate and multivariate meta-regression analyses were conducted to explore potential sources of heterogeneity, with age, exercise intensity (categorized into three levels based on typical daily practice durations in racket sports: Level 1: 1–1.5 h/day, Level 2: 2–3 h/day, and Level 3: >3 h/day), exercise duration (weeks), and exercise type (tennis, badminton, table tennis, and volleyball). Potential publication bias was evaluated visually using a funnel plot and statistically using Egger’s linear regression test. A leave-one-out sensitivity analysis was performed by sequentially removing each study to evaluate the robustness of the pooled results. For a subset of studies, original individual participant data were extracted from the included studies. Data preprocessing included normalization of all immune markers to z-scores. A random forest algorithm was implemented using the randomForest package, with the following parameters: ntree = 1000 and mtry = sqrt(p), where p is the number of features, and 10-fold cross-validation was used for model validation. Feature importance was calculated using mean decrease in accuracy. Unsupervised K-means clustering was applied to identify distinct responder profiles, with the optimal number of clusters determined by maximizing the average silhouette coefficient across k = 1 to 10 clusters. Cluster stability was assessed through 100 bootstrap replications. PCA was performed on standardized immune markers. A random forest algorithm was implemented using the randomForest package to rank the importance of immune markers (CD3, CD4, CD8, CD4/CD8, IgA, and IgG) in predicting exercise response. Unsupervised K-means clustering was applied to identify distinct responder profiles based on these immune markers, with the optimal number of clusters determined using the silhouette coefficient. Differences in immune marker levels and overall response between clusters were compared using t-tests. PCA was further used to visualize the separation between clusters in a reduced-dimensional space.

In the search strategy, a total of 6,274 articles were identified through the online database. During title and abstract screening, along with full-text review, 6,147 records were manually marked as duplicates. After removing these duplicates, 704 articles remained. Following the title and abstract screening, 573 articles were excluded, leaving 131 articles for further consideration. Upon reviewing the full text, 14 articles—comprising 8 foreign studies and 6 domestic studies—were ultimately included for analysis. The inclusion process is illustrated in Figure 1.

Among the 14 studies included in this review, three focused on badminton, seven on tennis, two on volleyball, and two on table tennis. The studies primarily involved participants from China, Japan, the United States, Brazil, and Serbia. The research spanned from 2004 to 2022 and included a total of 440 participants, comprising 211 in the experimental group and 229 in the control group. In terms of evaluation results, seven studies examined changes in CD3 levels; thirteen studies investigated the levels of CD4 and CD8; and ten studies analyzed the changes in the CD4/CD8 ratio. Additionally, three studies assessed changes in IgA, IgG, and IgM levels, while three others focused on changes in TNF-α and IL-6. Two studies also examined CK, BUN, and LDH. Table 1 provides detailed information on the exercise patterns and characteristics involved in each study.

Research quality and sports training research reports were evaluated by three independent reviewers. Among the 14 included studies (Zheng et al., 2015; Schafer et al., 2014; Yang and Hu, 2022; Xing et al., 2013; Li and Zhang, 2013; Kell, 2011; Suzui et al., 2004; Michalickova et al., 2016; Wenzhuang et al., 2007; Ma JinGe and Fang Lei, 2017; Kun, 2019; Yanmin et al., 2011; Wenli and Yanmin, 2011; Haibing, 2007), the literature was published between 2004 and 2022. Of these, 12 studies (92.86%) used a random sequence method, 10 studies (90.91%) implemented allocation concealment, 9 studies (69.23%) utilized a blind design, 13 studies (92.86%) conducted blind evaluations of the results, 13 studies (92.86%) provided complete data, and 8 studies (72.73%) were identified as selective reports (Figure 2).

The meta-analysis results (Figure 3A) showed that, contrary to the inhibitory effects (SMD = −1.24, 95% CI: −1.58 to −0.90) observed for T-cell markers (CD3, CD4, CD8, and CD4/CD8 ratio) (SMD = 0.86, 95% CI: 0.42–1.30), B-cell immunoglobulin (IgA, IgG, and IgM) and cardiorenal function markers (CK, LDH, and BUN) exhibited significant enhancing effects (SMD = 0.94, 95% CI: 0.55–1.33) (p < 0.05 or p < 0.01). The effect of swinging exercise on inflammatory markers (TNF -α, IL-6, and IFN -γ) is not significantly different (p > 0.05). The distribution of the effect size (Figure 3B) further confirms this patterned difference. These results suggest that under corresponding experimental or clinical conditions, changes in B-cell and cardiac renal function may be closely related to the inhibitory state of T cells. We applied the GRADE framework to assess the certainty of evidence for each outcome (Supplementary File S3). The evidence for T-cell markers was rated as moderate certainty, while evidence for B-cell immunoglobulins and inflammatory markers was rated as low certainty due to imprecision and heterogeneity. The evidence for cardiorenal markers was rated as very low certainty due to limited studies and high heterogeneity.

To investigate potential differences among specific swinging sports, we conducted subgroup analyses for CD4+/CD8+ ratio, CD4⁺, and CD8⁺ T cells. The CD4+/CD8+ ratio analysis included 11 studies across four sport types: tennis (five studies), table tennis (three studies), badminton (two studies), and volleyball (two studies). Tennis demonstrated the most pronounced effect on CD4+/CD8+ ratio reduction, with studies such as Yang WL (2011) (SMD = −1.39) and Yang YM (2022) (SMD = −1.39) showing substantial decreases. Volleyball studies (Xing JQ, 2013: SMD = −1.18; Xing J Q, 2013: SMD = −1.20) also showed significant reductions, while table tennis and badminton exhibited more variable responses (Supplementary Figure S1).

For CD4⁺ T cells, the analysis of 14 studies revealed that badminton (Ma JG, 2017: SMD = −6.99) and tennis (Yang WL, 2011: SMD = −5.32; Kell H, 2014: SMD = −5.84) produced the most substantial decreases, whereas table tennis showed mixed effects with both decreases (LiH, 2013: SMD = −0.99) and increases (Zheng QS, 2015: SMD = 2.04) (Supplementary Figure S2).

The CD8⁺ T-cell analysis of 13 studies demonstrated considerable heterogeneity across sports. Tennis showed the most consistent pattern, with multiple studies indicating decreases (Kell H, 2014: SMD = −5.84; Schafer M, 2014: SMD = −5.12), while table tennis exhibited both increases (Yang WL, 2011: SMD = 2.30; Yang YM, 2022: SMD = 2.36) and decreases (Yang HB, 2007: SMD = −2.54) (Supplementary Figure S3).

The results of the meta-regression analysis showed that age (Figure 4A) (β = 0.011, p = 0.839), exercise intensity (Figure 4B) (β = −1.737, p = 0.131), and duration (Figure 4C) (β = −0.006, p = 0.494) did not have a statistically significant moderating effect on the effect size. In the multivariate meta-regression model, the inclusion of these three moderating variables did not significantly explain the heterogeneity between studies (R² = 0%, p = 0.529). These results suggest that the response of CD4+ T cells to swing motion may be influenced by other unmeasured factors or that the small sample size limits the statistical power of detecting regulatory effects. The publication bias assessment showed that (Figure 4D) Egger’s test results were not statistically significant (z = −1.142, p = 0.254), and the funnel plot had good symmetry, indicating a low possibility of publication bias. Sensitivity analysis determines the robustness of the results by removing each study one by one. The results show that the estimated values of the combined effect size fluctuate between −1.72 and −0.91, and the confidence intervals of all estimates contain zero, further confirming the stability of the main analysis results.

This study used machine learning methods to conduct an in-depth analysis of immune marker data from swing-type exercise participants, aiming to identify the combination of immune markers most sensitive to exercise responses and distinguish between responders and non-responders. Through the random forest algorithm and cluster analysis, we have made the following important findings: the importance analysis of random forest variables (Figure 5A) shows significant differences in the contribution of six immune markers to predicting motor response ability. Among them, the importance score of the CD4/CD8 ratio is the highest. The importance of IgA and IgG immunoglobulins follows closely behind. The importance of CD4 and CD3 is at a moderate level, while the importance of CD8 is relatively low. This importance ranking pattern suggests that the impact of swinging exercise on the immune system may mainly be achieved by regulating the balance of T-cell subsets and enhancing humoral immunity, rather than simply changing the number of specific T-cell subsets. The clustering analysis results (Figure 5B) further confirmed the above findings. K-means clustering based on all six immune markers clearly divided the samples into two clusters, and contour coefficient analysis confirmed that the optimal number of clusters was 2. The clustering visualization (Figure 5C) shows that the two clusters are well separated in the immune marker space, with only a small amount of overlap. Cluster feature analysis revealed that the responder cluster (assuming cluster 1) exhibited significantly higher CD4/CD8 ratios (feature importance = 0.24), IgA (0.19) levels, and IgG (0.18) levels, while the non-responder cluster (assuming cluster 2) showed relatively lower values for these indicators. This patterned difference was further validated by the immune marker level boxplot (Figure 5D), and the response level comparison boxplot confirmed that the response level in the responder category was significantly higher than that of the non-responder category (t-test, p < 0.05), thereby verifying the effectiveness of the classification system (Figure 5E). Principal component analysis provides another perspective for understanding the relationship between immune markers and motor response. The PCA plot (Figure 5F) shows that there is some separation between responders and non-responders in the principal component space, but the degree of separation may be lower than that in the clustering results, indicating that although immune markers can distinguish response types, there may be some overlapping areas. These results reveal that the CD4/CD8 ratio, IgA, and IgG are the most important immune markers for predicting swing motion response. These biomarkers not only effectively distinguish responders from non-responders,but also show high predictive value for long-term survival outcomes.