This systematic review and meta-analysis was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines. The study protocol was prospectively registered with the International Prospective Register of Systematic Reviews (PROSPERO) prior to data collection and literature retrieval (registration number: CRD420251007058).

This study adopted a systematic review and meta-analysis design aimed at comprehensively evaluating the effects of exercise and nutritional supplementation on physical performance in taekwondo athletes.

The literature search covered studies published from the inception of each database up to February 27, 2025. A comprehensive search was conducted in PubMed, Scopus, and Web of Science (WoS). Keywords related to the study objectives were applied to titles, abstracts, and full texts to maximize the retrieval of potentially eligible studies (detailed search strategies are provided in Appendix 1).

To enhance both accuracy and efficiency, the search syntax was customized for each database: [Title/Abstract] in PubMed, TITLE-ABS-KEY in Scopus, and TS (Topic) in WoS. Boolean operators (AND, OR, NOT) were employed to refine keyword combinations. Studies involving patient or clinical populations were excluded to ensure that only those focusing on taekwondo athletes were retained.

The search was confined to peer-reviewed English-language journal articles. Grey literature—including theses, dissertations, conference proceedings, and non–peer-reviewed reports—was not considered in order to maintain methodological consistency and ensure data transparency across studies.

To ensure the methodological rigor of this study, a double-blind literature screening procedure was adopted and carried out independently by two reviewers. In the initial screening phase, the two reviewers independently screened the titles and abstracts to identify studies that either met the inclusion criteria or required further evaluation. Studies that were potentially eligible or presented uncertainties were forwarded to the full-text review stage. Subsequently, the selected full texts were independently assessed by two reviewers (CX and WZ) to determine whether they met the inclusion criteria for the systematic review and meta-analysis. In the event of disagreement between the two reviewers (CX and WZ) during the screening process, consensus was sought through discussion; if consensus could not be reached, a third reviewer (LL) was consulted to make the final decision.

The data extraction process was also conducted independently by two reviewers (CX and WZ) using a pre-designed Excel extraction sheet to ensure standardized documentation. Extracted data included: first author, year of publication, study design, participant characteristics (e.g., age, body weight, training experience, and sex), sample size, intervention type (exercise modality and nutritional supplement), duration and frequency of intervention, and reported performance outcomes.

Outcome data were extracted from each study, including TSAT (seconds), FSKT (repetitions), CMJ (centimeters), VO₂max (mL/kg/min), and HR (beats per minute, bpm), along with corresponding means and standard deviations (SD), standard errors (SE), or 95% confidence intervals (95% CI). If a study reported only SE without SD, SD was calculated using the formula SD = SE ×  n (for paired/crossover designs, within-subject correlations were accounted for where available).

For studies with incomplete outcome reporting or graphical-only data, the original authors were contacted via email to request raw data. All extracted data were cross-verified by both reviewers to ensure accuracy and reliability.

The Cochrane Risk of Bias tool (RoB 2.0), as recommended by the Cochrane Collaboration, was used to assess the methodological quality of the included studies. The assessment was independently performed by two reviewers (CX and WZ). RoB 2.0 domains include: randomization process; deviations from intended interventions; missing outcome data; measurement of the outcome; selection of the reported result. Each domain was categorized as having a “low risk,” “high risk,” or “some concerns” regarding bias. For each outcome, an overall risk of bias judgment was made following the RoB 2.0 algorithm: “low risk” if all domains were judged as low risk, “high risk” if at least one domain was judged as high risk, and “some concerns” if at least one domain raised some concerns but no domain was at high risk. Any discrepancies were resolved through discussion, with a third reviewer (LL) consulted when necessary.

The meta-analysis was performed using Review Manager (RevMan) and Stata 16.0 statistical software. The meta-analysis prioritized primary outcomes (TSAT, FSKT-10s, FSKT-mult); secondary outcomes (CMJ, VO₂max, HR indices) were synthesized to contextualize sport-specific effects. Given the variation in measurement tools and units across the included studies, standardized mean difference (SMD) and corresponding 95% confidence intervals (95% CI) were adopted as the effect size metrics to facilitate comparability. SMD was calculated by dividing the difference in means between the intervention and control groups by the pooled standard deviation. A significance threshold of p < 0.05 was applied. The magnitude of the SMD was interpreted according to Cohen’s classification—pre-specified in the analysis plan—as trivial (SMD < 0.2), small (0.2 ≤ SMD < 0.5), moderate (0.5 ≤ SMD < 0.8), and large (SMD ≥ 0.8).

For multi-arm studies, we addressed potential dependencies between effect sizes through the following approaches: (1) extracting independent comparisons where each intervention group was separately compared with the control group; (2) using shared control group methods with sample size adjustments when appropriate to avoid double-counting of participants; (3) conducting sensitivity analyses excluding multi-arm studies to verify the robustness of results. Random-effects models were prespecified for primary analyses; fixed-effects models were used only in sensitivity analyses when heterogeneity was low (p ≥ 0.05 and I² < 50%).

Heterogeneity among studies was assessed using the I² statistic and τ². When p ≥ 0.05 and I² < 50%, results were considered to show low heterogeneity; otherwise, substantial heterogeneity (p < 0.05 or I² ≥ 50%) prompted subgroup/sensitivity analyses. Subgroup variables included intervention duration (≤1 week vs. >1 week), intervention type, sex, study design (RCT vs. crossover), weight category (≤65 kg vs. >65 kg), and training experience (≤5 years vs. >5 years). Subgroup analyses were pre-specified and interpreted cautiously. For subgroups with fewer than five studies (or fewer than 150 total participants), estimates were reported descriptively without emphasizing p-values, and such findings were regarded as hypothesis-generating rather than confirmatory.

Random-effects meta-regression (REML) was planned only for outcomes with at least 10 effect sizes, using prespecified moderators including intervention type, duration, sex, study design, weight class, and training experience. Because most outcomes involved fewer than 10 contributing studies and CMJ exhibited negligible heterogeneity (I² ≈ 0%), meta-regression was not performed or was limited to exploratory analyses that were not intended for confirmatory inference.

To characterize between-study variability, a priori subgroup analyses, τ² estimates, and 95% prediction intervals (PI) were applied. When multiple effects were reported within a single study, potential dependency was minimized through shared-control adjustments and sensitivity analyses. Consideration was also given to the use of robust variance estimation, but this approach was not implemented due to the limited number of available studies.

In addition, when at least 10 studies were included for a given outcome, publication bias was preliminarily assessed by visual inspection of funnel plots generated using Stata 16.0, and by Egger’s test where appropriate; for outcomes with fewer than 10 studies, small-study effects were interpreted qualitatively due to limited power. Sensitivity analysis was also conducted by sequentially removing individual studies to observe changes in the pooled effect size, thereby verifying the robustness and stability of the results and strengthening the evidence supporting the conclusions.

A total of 993 relevant records were identified through a systematic search of three major electronic databases: Web of Science (WoS), PubMed, and Scopus. Specifically, 538 records were retrieved from WoS, 154 from PubMed, and 301 from Scopus. Prior to the initial screening, 376 duplicate records were removed, resulting in 617 unique records eligible for title and abstract screening.

During this initial screening stage, 535 records were excluded based on titles and abstracts due to irrelevance to the research topic, leaving 82 articles for full-text assessment.

After full-text review, 51 articles were excluded for the following reasons: 19 did not report relevant outcome measures; 4 had unavailable full texts; 22 did not meet the required study design criteria; 3 provided incomplete post-intervention data; and 3 were non-English publications.

Following a rigorous and systematic screening process, a total of 31 studies met the inclusion criteria and were included in the subsequent analyses. A detailed flow diagram of the literature selection process is presented in Figure 1.

According to the RoB 2.0 risk of bias assessment, the overall methodological quality of the 31 randomized controlled trials included in this study was relatively high. Among the six core evaluation domains, the randomization process performed best, with approximately 90% of studies judged as low risk and only 10% rated as having some concerns. Missing outcome data and selective reporting were also well controlled, with more than 85% of studies classified as low risk. By contrast, the measurement of outcome domain was relatively weaker, with about 75% rated as low risk and 25% as having some concerns. For deviations from intended interventions, 80% of studies were rated low risk, 15% had some concerns, and 5% were rated high risk. Overall, 80% of the included studies were judged as having low risk of bias, 20% as having some concerns, and none as high risk.

At the study level, the vast majority of trials demonstrated low risk (green) across key methodological domains, indicating generally acceptable design and implementation quality. The primary methodological concerns were concentrated in the domains of outcome measurement and deviations from intended interventions, where several studies were rated as “unclear risk” (yellow) due to insufficient objectivity in measurement methods or incomplete blinding implementation. Notably, no study was rated as high risk (red) in any domain, which strengthens the credibility of the meta-analysis findings and reflects the strict inclusion criteria applied, effectively minimizing the entry of low-quality studies.

The results of the RoB 2.0 assessment are illustrated in Figures 2, 3.

Given the substantial heterogeneity identified in the overall analysis, subgroup analyses were conducted to explore potential sources of heterogeneity. Subgroups were categorized based on intervention duration, type of exercise intervention, sex, study design, body weight, and training experience.

To assess the robustness of the meta-analytic findings, sensitivity analyses were performed for all performance-related outcomes in taekwondo athletes. The analyses involved sequentially removing each individual study to examine whether the pooled effect size remained stable—that is, whether the revised estimate after exclusion stayed within the 95% CI of the overall pooled effect.

The results indicated that the exclusion of any single study did not substantially alter the direction or magnitude of the overall effects, suggesting that the findings of this meta-analysis are robust and not unduly influenced by any single trial. Sensitivity analysis results for exercise training are presented in Figures 16–21, and those for nutritional supplementation are shown in Figures 22–27.

To further assess the potential presence of publication bias, Egger’s regression test and Begg’s rank correlation test were employed for quantitative evaluation. The Egger’s test results for each outcome are summarized in Table 7.

Overall, all Egger’s test p-values were greater than 0.05—except for TSAT (p = 0.049) and VO₂max (p = 0.032) in the exercise intervention subgroup—indicating no significant evidence of publication bias for most outcomes.

For the two potentially biased outcomes (TSAT and VO₂max), the trim-and-fill method was applied for adjustment. The results revealed that no missing studies were imputed, and the adjusted pooled estimates changed from TSAT: SMD = −0.82 (95% CI: −1.43 to −0.21; p = 0.009) to SMD = 0.44 (95% CI: 0.24 to 0.81; p = 0.009), and from VO₂max: SMD = 1.54 (95% CI: 0.58 to 2.49; p = 0.002) to SMD = 4.65 (95% CI: 1.79 to 12.08; p = 0.002).

The consistency of statistical significance and direction before and after adjustment suggests that the potential publication bias for TSAT and VO₂max is minimal and has negligible impact on the final conclusions.

Additionally, Begg’s funnel plots were generated to visually inspect publication bias. Most studies were symmetrically distributed within the 95% CI range, and no substantial asymmetry was detected in the funnel plots, confirming the absence of major bias. Detailed results are presented in Appendix 5.

This meta-analysis represents the first comprehensive synthesis of both exercise and nutritional interventions targeting taekwondo athletes. It integrated six key physiological and performance indicators, thereby offering an evidence-based framework for optimizing training and supplementation strategies in combat sports.

However, several limitations should be acknowledged. First, considerable heterogeneity in study design, sample size, and measurement protocols may affect the generalizability of findings. Second, most studies investigated single interventions with short-term follow-up, precluding conclusions about long-term or combined effects. Third, several subgroup analyses were based on small samples (k = 2–3), limiting statistical power and precision; these results should therefore be interpreted as exploratory. Fourth, although publication bias was generally low, the restriction to English-language, peer-reviewed studies may still have introduced selection bias.

Future research should focus on large-scale, multicenter randomized controlled trials employing standardized testing and longitudinal tracking. Integrating physiological, biochemical, and psychological indicators will yield a more comprehensive understanding of performance adaptations. Moreover, AI-driven analytics and wearable technology could enable real-time performance monitoring, predictive modeling, and precision-tailored training or nutrition interventions, ultimately advancing the scientific basis for elite taekwondo performance optimization.