Soccer is a high-intensity intermittent sport in which explosive actions such as sprinting, vertical jumping, and rapid changes of direction (COD) are critical to performance (Bloomfield et al., 2007; Gualtieri et al., 2023; Loturco et al., 2020). Match analyses suggest that short-distance sprinting is frequently involved in decisive situations, including goal-scoring actions, although these associations are influenced by tactical context and player role (Bradley et al., 2009; Chmura et al., 2018; Filter et al., 2023). Accordingly, training methods aimed at improving acceleration and maximal-velocity capacity are widely used to enhance match-relevant physical performance (Haugen et al., 2019b; Rumpf et al., 2016). However, conditioning practice in soccer is constrained by congested schedules, technical-tactical demands, and injury risk (Morgans et al., 2014).

For strength and conditioning practitioners, the practical challenge is rarely to choose a single training modality in isolation, because these components are typically embedded within concurrent training weeks alongside technical-tactical sessions and match-related demands. Rather, under time constraints, the key applied question is which training component should receive greater emphasis to maximize performance gains within the overall training week (Enright et al., 2015). While various interventions are commonly used within these concurrent training structures, their relative value when prioritizing one component over another remains a subject of ongoing debate. Therefore, a systematic evaluation using a network meta-analysis (NMA) is essential. Unlike a conventional pairwise meta-analysis, which is limited to direct head-to-head comparisons, an NMA can simultaneously integrate direct and indirect evidence across multiple training modalities and estimate their relative ranking. This approach is particularly valuable in soccer conditioning, where not all intervention types have been compared directly, yet practitioners must still decide which training component should be prioritized to optimize athletic development while balancing performance gains and physical load (Hutton et al., 2015).

Various training modalities, including strength, plyometric, and speed-oriented training, have been used to enhance sprint and jump performance in soccer players. However, an important evidence gap remains. Previous systematic reviews and conventional pairwise meta-analyses have mainly evaluated selected training modalities in isolation or through limited head-to-head comparisons, making it difficult to determine their relative effectiveness across the broader range of conditioning options available in practice (Marshall et al., 2021; Oliver et al., 2024). The present network meta-analysis extends prior pairwise syntheses by integrating both direct and indirect evidence across multiple intervention nodes within a single analytical framework, allowing simultaneous comparison and ranking of six pre-specified exercise modalities. To improve comparability, the included interventions were classified according to their dominant training stimulus into six categories: Traditional Strength Training, Plyometric Jump Training, Speed Training, Endurance Training, Flexibility Training, and Regular Training.

In this study, sprint performance and vertical jump height were pre-specified as co-primary outcomes because they capture the horizontal and vertical dimensions of explosive performance most directly relevant to soccer conditioning. Change-of-direction (COD) performance and maximal strength (1RM) were treated as secondary outcomes, included to provide mechanistic and applied context rather than to define the principal effectiveness claims of the review. Because multiple outcomes were analyzed, we pre-specified that the main inferential interpretation and overall conclusions would be based on the co-primary outcomes, whereas findings for secondary outcomes would be interpreted as supportive rather than primary evidence. Accordingly, this review aimed to compare the relative effects of six exercise modalities on sprint- and jump-related outcomes in soccer players using a frequentist NMA and to provide a more decision-relevant ranking of interventions for practitioners working under real-world competitive and logistical constraints.

This systematic review and network meta-analysis (NMA) was conducted in accordance with the Cochrane Handbook and reported following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 (PRISMA 2020) statement. The protocol was prospectively registered in PROSPERO (CRD42024608868).

A comprehensive literature search was conducted across PubMed, Embase, the Cochrane Library, Web of Science, and SPORTDiscus. To improve coverage and reduce retrieval bias, grey literature and trial registries, including ClinicalTrials.gov and the World Health Organization International Clinical Trials Registry Platform, were also searched. The search covered all records published from January 2000 to 30 September 2025. The strategy was developed according to the PICOS framework: Population (soccer players), Intervention (exercise modalities), Comparison (various training modalities or control groups), Outcome (sprint performance and jump height), and Study Design (randomized controlled trials, RCTs). The full PubMed search strategy is presented in Table 1, and the remaining strategies are provided in Supplementary Table S1.

To improve consistency of intervention classification and reduce within-node heterogeneity in the network meta-analysis, we refined the taxonomy using pre-specified operational definitions and decision rules based on the dominant training stimulus rather than broad conceptual labels alone. Interventions were classified into six pre-specified categories: Traditional Strength Training, Plyometric Jump Training, Speed Training, Endurance Training, Flexibility Training, and Regular Training.

Traditional Strength Training was defined as lower-body resistance training primarily intended to develop maximal force, typically using heavy external loads (generally ≥ 80% 1RM or ≤ 6RM) in exercises such as squats, deadlifts, split squats, or leg press variations. Plyometric Jump Training comprised interventions in which fast stretch-shortening cycle actions, such as drop jumps, countermovement jumps, hurdle jumps, and bounds, constituted the main overload stimulus. Speed Training referred to sprint-specific interventions targeting acceleration or maximal-velocity mechanics, including assisted, resisted, or unresisted sprinting performed at maximal or near-maximal intensity with full or near-full recovery. Endurance Training included interventions primarily targeting aerobic or anaerobic metabolic conditioning rather than neuromuscular overload. Flexibility Training included interventions in which static, dynamic, ballistic, or proprioceptive neuromuscular facilitation stretching formed the main additional training content. Regular Training referred to comparison conditions in which participants continued their usual soccer practice without any additional structured physical intervention specifically designed to overload strength, plyometric, speed, endurance, or flexibility qualities; on this basis, it was treated as a relatively low-stimulus active comparator rather than a high-load training condition.

To minimize misclassification bias, a study arm was assigned to a single node only when one modality clearly represented the dominant training stimulus, defined a priori as accounting for at least 60% of the prescribed additional training content, based on session time, number of drills, or number of sets when such information was available. When a program combined two or more modalities without a clearly dominant component, it was classified as multicomponent and was not merged into a single-modality node. For incompletely reported protocols, classification was based on the totality of prescription details, including load intensity, repetition range, exercise selection, contraction intent, recovery structure, and the authors’ stated training objective. These refinements were implemented to improve within-node homogeneity and strengthen the plausibility of the transitivity assumption, while acknowledging that some residual heterogeneity is unavoidable in applied training studies.

Studies were included if they met the following PICOS-based criteria: (1) the experimental group comprised soccer players receiving a specific exercise intervention, with sex and competitive level not used as exclusion criteria; (2) the control group received a different training modality or regular training only; (3) the study design was a randomized controlled trial; (4) at least one co-primary outcome was reported, namely sprint performance at any pre-specified distance or vertical jump height (SJ or CMJ); secondary outcomes of interest were COD performance and maximal strength (1RM), which were extracted when available and synthesized to provide supportive evidence regarding training-specific adaptations; and (5) the full text was available for data extraction. No language restrictions were applied during the initial selection process.

Studies were excluded if they were non-randomized or quasi-randomized controlled trials, secondary research or other non-original articles, or trials with incomplete, missing, or irrecoverable critical data required for the network meta-analysis.

No language restrictions were applied. For non-English articles identified during screening, full texts were processed using a standardized translation workflow, with initial machine translation followed by manual verification by two bilingual researchers (Y.W. and Z.L.). For articles written in languages not mastered by the team, professional translation services or native-speaking academic colleagues were consulted when necessary to confirm eligibility and extract critical data. This procedure was used to implement the no-language-restriction policy consistently and to minimize potential selection bias. In addition, intervention classification was independently reviewed by two investigators during data extraction. Any disagreements regarding node assignment, particularly for borderline or multicomponent programs, were resolved through discussion and consensus.

Risk of bias for the included RCTs was independently assessed by two authors (Y.W. and Z.L.) using the Cochrane Risk of Bias tool version 2.0 (RoB 2, 2019). Five domains were evaluated: (1) bias arising from the randomization process, (2) bias due to deviations from intended interventions, (3) bias due to missing outcome data, (4) bias in measurement of the outcome, and (5) bias in selection of the reported result. Each domain was rated as low risk, some concerns, or high risk according to RoB 2 guidance. Overall risk was judged as high risk if any domain was rated high, some concerns if at least one domain raised concerns but none was rated high, and low risk if all domains were rated low. Disagreements were resolved through discussion and consensus. Inter-rater agreement was quantified using Cohen’s kappa, which was 0.83, indicating substantial agreement.

Outcomes were pre-specified hierarchically. The co-primary outcomes were sprint performance (5, 20, 30, and 40 m) and vertical jump height (SJ and CMJ), as these outcomes most directly reflect the sport-relevant explosive capacities targeted by the included interventions. Secondary outcomes were change-of-direction (COD) performance and maximal strength (1RM), which were included to complement the interpretation of neuromuscular adaptation and training transfer.

Because the review synthesized multiple related outcomes, we pre-specified that the primary interpretation of comparative effectiveness would be based on the co-primary outcomes only. Secondary outcomes were considered supportive rather than primary evidence. No formal multiplicity adjustment across outcomes was applied, because the outcomes were analyzed in separate outcome-specific network meta-analyses and the review was primarily intended to compare patterns of effectiveness across performance domains rather than to test a single family-wise confirmatory hypothesis. Accordingly, findings for COD and 1RM were interpreted cautiously and were not used to make primary effectiveness claims.

For quantitative synthesis, post-intervention means and standard deviations were extracted for all outcomes. Outcome units were standardized before analysis, and the units used for each outcome together with the interpretation of effect direction are summarized in Table 2. Sprint performance was analyzed as time in seconds for the 5, 20, 30, and 40-m tests, with lower values indicating better performance; therefore, negative MDs favored the intervention group. Vertical jump performance was analyzed as jump height in centimeters for SJ and CMJ, with higher values indicating better performance; therefore, positive MDs favored the intervention group. COD performance was analyzed as movement speed in meters per second, with higher values indicating better performance; therefore, positive MDs favored the intervention group. Maximal strength was analyzed as 1RM in kilograms, with higher values indicating better performance; therefore, positive MDs favored the intervention group. To maximize comparability across intervention nodes and study designs, outcome data were extracted from assessments conducted immediately after completion of the intervention. This time point was prioritized because it was consistently available across the included randomized controlled trials and most directly reflected the short-term effect of the assigned training modality. No longer-term follow-up data were available in the included trials; therefore, the present network meta-analysis was restricted to immediate post-intervention effects and could not evaluate the durability of training-induced adaptations. Because the validity of the network depends partly on the comparability of interventions within each node, the intervention taxonomy and node assignments were finalized before model fitting using the pre-specified operational criteria described in Section 2.2. When only change-from-baseline scores were reported, post-intervention values were calculated according to the Cochrane Handbook, and when SDs were missing, they were estimated from standard errors or confidence intervals. All randomized controlled trials (RCTs) included in the network meta-analysis had independent control groups. Therefore, each contrast was treated as statistically independent, and no adjustment for shared comparators was required.

A frequentist random-effects network meta-analysis was conducted using the Stata network suite (version 15.1), with standard errors and 95% confidence intervals estimated for all comparisons. Between-study heterogeneity was estimated separately for each outcome-specific network, assuming a common between-study variance (τ²) across treatment comparisons within each network and estimating it using restricted maximum likelihood (REML). All performance metrics were converted to standardized native units: seconds (s) for sprint outcomes, centimeters (cm) for SJ and CMJ, meters per second (m/s) for COD, and kilograms (kg) for 1RM. Because sprint, jump, COD, and 1RM outcomes were reported in interpretable native units, MDs were retained as the primary effect measure. However, for sprint outcomes, identical units do not necessarily imply full measurement commensurability across studies, because timing method, start procedure, and test surface varied. Therefore, sprint testing characteristics were extracted for the studies contributing sprint outcomes and summarized in Supplementary Table S32, and an additional timing-method sensitivity analysis was conducted to examine the robustness of the sprint findings (Supplementary Table S33).

To improve interpretation of between-study heterogeneity and the likely range of effects in future studies, we additionally report the estimated τ² for each outcome-specific network and, where informative, 95% prediction intervals for key contrasts (Supplementary Table S31). To evaluate the potential impact of effect modifiers and to further examine the plausibility of the transitivity assumption, we conducted pre-specified network meta-regression analyses for several participant characteristics, including sex, age, maturation, playing standard, season phase, and baseline ability. These analyses are reported in Supplementary Table S28. All meta-regression analyses were performed as separate univariable models, with one covariate entered at a time. No multivariable meta-regression models were fitted. Therefore, the analyses did not involve simultaneous inclusion of multiple covariates in the same model, which reduced concerns regarding collinearity. Because each model included only one covariate, this strategy was also adopted to minimize the risk of overfitting, particularly in sparse outcome networks. In addition, all covariate data used in the meta-regression analyses were complete; therefore, no imputation procedures were required and no studies were excluded because of missing covariate data. Given the limited number of studies in some outcome networks, these meta-regression findings were interpreted cautiously as exploratory rather than confirmatory. Consistency was assessed using both global and local approaches. Global inconsistency was examined with the design-by-treatment interaction model, whereas local inconsistency was assessed using node-splitting (Supplementary Table S30). Because a non-significant node-splitting p-value does not establish consistency, particularly in sparse loops with limited statistical power, these tests were interpreted together with the magnitude of inconsistency factors and the width of their 95% confidence intervals. Therefore, consistency was judged cautiously as the absence of important detected disagreement between direct and indirect evidence, rather than as proof of exact agreement. Interventions were ranked using Surface Under the Cumulative Ranking (SUCRA) probabilities (0-100%). Small-study effects and potential publication bias were assessed at the outcome-network level rather than using comparison-adjusted funnel plots. Because several treatment loops in the present network were sparse, comparison-adjusted funnel plots were not prespecified for the primary assessment. Instead, for outcome networks including at least 10 studies, conventional funnel-plot asymmetry and Egger’s linear regression test were examined; Begg’s rank correlation test was additionally reported as a supportive analysis. For outcome networks with fewer than 10 studies, no formal statistical test was interpreted because such tests have limited power and may be misleading in sparse evidence structures. To prespecify how these signals informed certainty of evidence, suspected small-study effects or publication bias were considered in the CINeMA reporting-bias domain and could contribute to downgrading certainty for the affected outcome/comparisons. In sparse networks, absence of statistical significance was not interpreted as evidence of absence of publication bias. In addition, sensitivity analyses were performed for all outcome-specific networks using leave-one-study-out influence diagnostics and outlier checks to examine the robustness of the pooled estimates. Where small-study effects were suspected, these findings were also considered when interpreting the stability of the results and when judging certainty of evidence.

A systematic search across the electronic databases PubMed, Embase, the Cochrane Library, Web of Science, and SPORTDiscus yielded 6,546 records. SPORTDiscus was included to improve retrieval of sports-specific literature not consistently indexed in broader biomedical databases. After removal of 1,012 duplicates, 5,534 records underwent title and abstract screening, of which 5,217 were excluded. Of the 317 reports sought for retrieval, 3 were unavailable. The remaining 314 full-text articles were assessed for eligibility, and 283 reports were excluded because of absence of a control group (n = 108), missing critical outcome data (n = 52), unmatched intervention protocols (n = 58), or ineligible publication formats such as reviews, case reports, or conference abstracts (n = 65). Ultimately, 31 randomized controlled trials (RCTs) met the eligibility criteria and were included in the final network meta-analysis (Figure 1).

Risk of bias was assessed using the Cochrane RoB 2 tool. Of the 31 included studies, 10 were judged as low risk, 18 as some concerns, and 3 as high risk. These ratings informed the within-study bias domain in the subsequent certainty-of-evidence assessment. Given the nature of exercise intervention research, complete blinding of participants and personnel was often not feasible, largely because informed consent and direct participation in specialized training programs are inherently difficult to mask. A detailed summary of the risk-of-bias assessments is shown in Supplementary Figure S1, and comprehensive evidence profiles are provided in Supplementary Table S20, Table S21, Table S22, Table S23, Table S24, Table S25, Table S26 and Table S27 .

The 31 included RCTs involved soccer players as the primary participant population. The evidence base was composed predominantly of male professional or academy-level players, whereas women’s soccer and other playing levels were not meaningfully represented in the eligible trials. These studies compared six distinct training modalities: Traditional Strength Training, Plyometric Jump Training, Speed Training, Endurance Training, Flexibility Training, and Regular Training. The evaluated outcomes focused on linear sprint performance (5 m to 40 m distance), vertical jump height (Squat Jump and Countermovement Jump), maximal strength (1RM), and change-of-direction (COD) performance. The intervention durations across the studies ranged from 1 to 16 weeks. All included studies were published in English. Further study characteristics are summarized in Supplementary Table S2. For studies contributing sprint outcomes, testing characteristics relevant to measurement comparability were additionally summarized in Supplementary Table S32. Most studies used electronic/photoelectric timing systems, whereas a minority used video/app-based methods or did not clearly report the timing method; start procedures and running surfaces also varied across studies.

Consistency was evaluated using both local node-splitting analyses and the global design-by-treatment interaction model. Local inconsistency factors were generally small across outcome-specific loops, although 95% confidence intervals were often wide in sparse parts of the network ( Supplementary Table S3, Table S4, Table S5, Table S6, Table S7, Table S8, Table S9 and Table S10 ). Accordingly, node-splitting P-values were not interpreted in isolation; instead, consistency was judged cautiously by considering the magnitude and precision of local inconsistency estimates together with the global inconsistency results. Several local loops were sparse, particularly in the 30-m and 40-m sprint networks and in comparisons involving Flexibility Training or Endurance Training. At the global level, the design-by-treatment interaction model did not detect important inconsistency, but this finding should be interpreted as supportive rather than confirmatory. Exploratory network meta-regression analyses based on separate univariable models (Supplementary Table S28) suggested that baseline ability and maturation modified effects in several outcomes (P < 0.05), whereas playing standard and season phase did not significantly modify effects in the 5-m and 20-m sprint networks. In addition, the network meta-regression analysis of training dosage modifiers (Supplementary Table S29) showed no significant interaction for training frequency, session duration, or intervention duration across the eight performance outcomes (all P ≥ 0.05). Overall, these findings support, but do not establish, the plausibility of transitivity and should be interpreted cautiously, particularly in sparse outcome networks. Additional information on between-study heterogeneity and prediction intervals is provided in Supplementary Table S31.

In the 5 m sprint test, expressed as time in seconds (s), Traditional Strength Training ranked highest by SUCRA (99.8%; Figure 2B) and showed the most favorable effect for enhancing initial acceleration. Speed Training ranked second (SUCRA = 66.5%) and showed a statistically significant improvement over Regular Training (MD = -0.07 s, 95% CI: -0.10 to -0.04; Supplementary Table S11). Traditional Strength Training also demonstrated a statistically significant improvement over Regular Training (MD = -0.09 s, 95% CI: -0.11 to -0.07; Supplementary Table S11). Plyometric Jump Training ranked third (SUCRA = 33.4%), whereas Regular Training ranked lowest by SUCRA (0.3%).

Network results for the 20 m sprint, expressed as time in seconds (s), revealed that Traditional Strength Training had the highest probability of effectiveness (SUCRA = 89.9%; Figure 3B). Traditional Strength Training demonstrated a statistically significant improvement over Regular Training (MD = -0.13 s, 95% CI: -0.20 to -0.06; Supplementary Table S12). Speed Training ranked second (SUCRA = 73.2%) and also showed a favorable effect compared with Regular Training (MD = -0.10 s, 95% CI: -0.16 to -0.03; Supplementary Table S12). Endurance Training (SUCRA = 49.9%) and Flexibility Training (SUCRA = 37.9%) ranked lower, and Regular Training remained the least effective intervention (SUCRA = 11.7%).

For the 30 m sprint test, expressed as time in seconds (s), the hierarchy of effectiveness shifted toward velocity- specific and reactive modalities (Figure 4B). Speed Training achieved the highest SUCRA (74.2%), followed by Plyometric Jump Training (65.2%) and Traditional Strength Training (64.8%). Endurance Training also showed substantial efficacy (SUCRA = 63.0%). In contrast, Regular Training (SUCRA = 18.7%) and Flexibility Training (SUCRA = 14.1%) were positioned at the bottom of the ranking. However, the principal contrast between Speed Training and Regular Training did not reach statistical significance (MD = -0.16 s, 95% CI: -0.34 to 0.01; Supplementary Table S13), so this ranking should not be interpreted as definitive evidence of superiority.

Regarding the 40 m sprint, expressed as time in seconds (s), no training intervention demonstrated a statistically significant improvement compared with Regular Training. Although Endurance Training achieved the highest SUCRA (74.1%), the principal contrast with Regular Training was not statistically significant (MD = -0.21 s, 95% CI: -0.45 to 0.03; Supplementary Table S14). Speed Training (SUCRA = 69.6%) and Traditional Strength Training (SUCRA = 51.5%) ranked next, but the cumulative ranking curves (Figure 5B) should be interpreted with caution. In the absence of clear between-intervention differences, these rankings reflect substantial statistical uncertainty rather than a definitive hierarchy of effectiveness. Regular Training was again the least effective modality (SUCRA = 4.9%).

In the SJ test, expressed as jump height in centimeters (cm), Traditional Strength Training demonstrated the most favorable outcomes (SUCRA = 86.2%; Figure 6B) and showed a significant improvement over Regular Training (MD = 4.40 cm, 95% CI: 2.07 to 6.74; Supplementary Table S15). Speed Training (SUCRA = 56.3%) and Plyometric Jump Training (SUCRA = 55.2%) ranked next and showed broadly similar SUCRA profiles, whereas Regular Training ranked lowest by SUCRA (2.3%).

In the CMJ test, expressed as jump height in centimeters (cm), Traditional Strength Training maintained its superiority and ranked highest by SUCRA (73.0%; Figure 7B). This was followed by Plyometric Jump Training (SUCRA = 67.8%) and Endurance Training (SUCRA = 61.4%). Traditional Strength Training showed the most beneficial effect compared with Regular Training (MD = 3.58 cm, 95% CI: 2.00 to 5.15; Supplementary Table S16). Regular Training ranked lowest by SUCRA (0.3%).

In the COD test, expressed as movement speed in meters per second (m/s), Traditional Strength Training (SUCRA = 83.8%) and Plyometric Jump Training (SUCRA = 81.4%) ranked highest by SUCRA, as illustrated in Figure 8B. Traditional Strength Training demonstrated superior results compared with Regular Training (MD = 0.13 m/s, 95% CI: 0.08 to 0.18; Supplementary Table S17). Conversely, Regular Training (SUCRA = 1.9%) and Flexibility Training (SUCRA = 24.3%) were the least effective for enhancing COD performance. For this outcome, positive MD values indicate better performance.

In the 1RM test, expressed in kilograms (kg), Traditional Strength Training ranked highest for maximal strength (SUCRA = 99.2%; Figure 9B). It showed clear advantages over Speed Training (MD = 15.12 kg, 95% CI: 0.22 to 30.02) and Regular Training (MD = 20.86 kg, 95% CI: 10.73 to 30.98; Supplementary Table S18). Speed Training (SUCRA = 54.3%) and Regular Training (SUCRA = 28.2%) ranked lower, while Flexibility Training was ranked last (SUCRA = 18.3%).

To evaluate potential small-study effects and publication bias, we assessed outcome-specific funnel-plot asymmetry rather than using comparison-adjusted funnel plots, because several treatment loops in the network were sparse. For outcome networks with at least 10 studies, conventional funnel plots were visually inspected and Egger’s linear regression test was applied; Begg’s rank correlation test was reported as a supportive analysis. As shown in Figure 10 and Supplementary Table S19, no statistically significant small-study effects were detected for most outcomes. However, for 1RM strength, Egger’s test suggested possible small-study effects (P = 0.009). For the 40-m sprint outcome (8 studies), formal interpretation of funnel-based statistical testing was considered limited because of the small number of included studies. These signals were incorporated cautiously into the certainty-of-evidence assessment and were not interpreted in isolation from imprecision and network sparsity.

Sensitivity analyses were conducted for all eight outcomes using leave-one-study-out influence diagnostics and outlier checks (Supplementary Figure S10, Figure S11, Figure S12, Figure S13, Figure S14, Figure S15, Figure S16 and Figure S17). Overall, omission of any single study did not materially alter the direction or general pattern of the pooled estimates across the outcome networks. However, the magnitude of the pooled effect appeared less stable for 1RM than for the co-primary sprint and jump outcomes, indicating that this secondary outcome should be interpreted more cautiously. This interpretation was further supported by the finding of possible small-study effects for 1RM in Egger’s test. For sprint outcomes, an additional timing-method sensitivity analysis was performed by excluding studies that used video/app-based timing methods, while retaining studies with unclear timing-method reporting to avoid excessive fragmentation of sparse networks (Supplementary Table S33). The key findings for the 5-m and 20-m sprint outcomes were materially unchanged after this restriction, whereas the 30-m and 40-m results remained directionally similar but imprecise. These results suggest that the main sprint conclusions were not driven solely by the inclusion of non-photoelectric/electronic timing methods.

This network meta-analysis evaluated the comparative effectiveness of six training modalities on the athletic performance of soccer players, acknowledging that in practice these methods are typically embedded within concurrent and multi-modal training systems (Enright et al., 2015; Morgans et al., 2014). Our findings suggest a performance-specific pattern of comparative effectiveness across outcomes, although the hierarchy for longer sprint distances, particularly 40 m, remained uncertain (Marshall et al., 2021; Oliver et al., 2024).

In the analysis of 5 m sprint performance, which represents the initial acceleration phase, Traditional Strength Training demonstrated a statistically significant benefit compared with Regular Training (MD = -0.09 s, 95% CI: -0.11 to -0.07) and ranked highest across the network. With a SUCRA value of 99.8%, it was identified as the most effective intervention for enhancing initial acceleration (Seitz et al., 2014; Skratek et al., 2024). This finding aligns with the mechanical imperatives of starting from a stationary position, where overcoming inertia requires maximal force production (Cormie et al., 2011a; Loturco et al., 2020). High-load resistance training is theorized to facilitate these improvements, potentially by enhancing the rate of force development (RFD), which is critical during the explosive push-off phase of a sprint start (Brumitt and Cuddeford, 2015; Suchomel et al., 2016). Galantine et al. (2025) found that take-off velocity in ballistic lower limb tests strongly correlates with sprint acceleration, further supporting the link between force capacity and early-stage sprinting. Furthermore, the neural adaptations typically associated with strength training—including enhanced motor unit recruitment and firing frequency—may provide a plausible physiological foundation for the predominant concentric muscle actions required during the first few steps (Kim et al., 2019; Santos et al., 2023).

Notably, the present network meta-analysis reveals that Traditional Strength Training maintains its top-ranked position for 20 m sprint performance (SUCRA = 89.9%), significantly outperforming Regular Training (MD = -0.13 s, 95% CI: -0.20 to -0.06). This result suggests that within the context of soccer-specific distances, the capacity to generate maximal ground reaction forces (GRF) likely remains a decisive driver of performance throughout the mid-acceleration phase (Hunter et al., 2005; Nagahara et al., 2018). Traditional Strength Training not only provides a structural basis via potential increases in muscle cross-sectional area but may also optimize neuromechanical efficiency, allowing athletes to apply higher impulses within extremely short ground contact times (Blazevich et al., 2007; Cormie et al., 2011b). This interpretation is also consistent with the broader resistance-training literature showing that program structure and exercise mode can influence strength, hypertrophy, and sprint/jump transfer (De Souza et al., 2018; Egan and Sharples, 2023; Ferrete et al., 2014; García-Valverde et al., 2022; Zhang et al., 2022). Consequently, for soccer players aiming to enhance their explosive break-away ability over the first 20 meters, establishing a foundation of maximal lower-body strength appears to be a primary focus of physical development (Keiner et al., 2022; Loturco et al., 2015).

At 30 m, Speed Training ranked highest by SUCRA; however, the corresponding comparison with Regular Training was imprecise and crossed the null. Therefore, these findings are more appropriately interpreted as a trend rather than definitive evidence of superiority. Any interpretation beyond the observed ranking pattern should be considered an extrapolation rather than a direct finding of the present synthesis. Although sprint performance over this distance may plausibly reflect the transition from acceleration toward higher running velocities, the current evidence does not establish a clear between-intervention advantage. In addition, this outcome should be interpreted in light of methodological variation across studies, including differences in starting stance, timing systems, and running surfaces, all of which may influence whether a given sprint distance reflects primarily acceleration or top-end speed (Haugen et al., 2019a).

At 40 m, Endurance Training had the highest SUCRA value; however, no intervention was statistically superior to Regular Training. Therefore, the apparent 40 m ranking should not be interpreted as a reliable hierarchy of comparative effectiveness. Any interpretation beyond this result should be considered an extrapolation rather than a finding directly supported by the present synthesis. In addition, the Endurance Training node included heterogeneous interventions primarily targeting metabolic conditioning, rather than a uniform set of methods specifically designed to improve isolated single-bout 40 m sprint performance. In particular, phosphocreatine resynthesis is more relevant to recovery between repeated sprint bouts than to the time achieved in a single maximal 30-40 m sprint.

Accordingly, any apparent advantage of endurance-oriented interventions for 40 m performance should be interpreted as indirect, modest, and hypothesis-generating rather than as evidence of a primary mechanistic role. Although 40 m sprint performance may plausibly reflect the transition toward maximal velocity, the current evidence does not demonstrate that endurance-oriented training improves this outcome in soccer players. At most, this pattern may be regarded as hypothesis-generating and should be interpreted cautiously, particularly because medium- to long-distance sprint performance reflects an interaction of maximal-velocity mechanics and speed-endurance qualities rather than a simple extension of early acceleration alone (Brocherie et al., 2016; Nicholson et al., 2022; Rakovic et al., 2018; Shalfawi et al., 2012; Maciel et al., 2024).

Traditional Strength Training was also the most effective modality for both the SJ and CMJ, reinforcing the importance of maximal-force capacity for vertical power production. This finding is consistent with evidence that high-load resistance training improves force output through both neural and morphological adaptations, including enhanced motor unit recruitment and changes in muscle architecture that may support concentric impulse generation (Blazevich et al., 2007; Gouveia et al., 2023; Mirzayev, 2017). At the same time, Plyometric Jump Training ranked particularly well for the CMJ, which is mechanically plausible because the CMJ relies more strongly on effective stretch-shortening cycle utilization than the SJ. The near-equivalent efficacy of strength- and plyometric-oriented approaches for CMJ performance suggests that both maximal-force development and reactive-strength enhancement contribute meaningfully to vertical explosive capacity in soccer players (Lamas et al., 2012; Ramírez-delaCruz et al., 2022; Rønnestad et al., 2008).

As pre-specified secondary outcomes, COD and 1RM were interpreted as supportive rather than primary evidence and were not used to define the principal comparative conclusions of this review. Traditional Strength Training (MD = 0.13 m/s compared with Regular Training, 95% CI: 0.08 to 0.18) and Plyometric Jump Training (MD = 0.13 m/s compared with Regular Training, 95% CI: 0.05 to 0.21) demonstrated significant and comparable efficacy for enhancing COD performance (Nygaard Falch et al., 2019; Pardos et al., 2024). This pattern is reasonable because COD performance depends on both braking capacity and rapid re-acceleration, which may be supported by force-oriented and reactive-strength adaptations. However, the included COD tests were predominantly planned, closed-skill tasks, so the findings are more likely to reflect improvements in mechanical change-of-direction capacity than in full reactive agility.

Therefore, the observed effectiveness of strength and plyometric training likely stems from improvements in foundational mechanical factors—such as eccentric leg stiffness, braking capacity, and reactive strength—which are paramount for managing high approach velocities during sharp turns. However, the extent to which these gains transfer to reactive agility should not be assumed to be complete. In applied soccer settings, reactive agility requires not only the physical ability to decelerate and re-accelerate, but also the perceptual-cognitive capacity to detect, interpret, and respond to evolving external cues such as opponent movement, ball trajectory, and game context. Accordingly, improvements in planned COD performance may provide an important physical foundation for reactive agility, particularly by increasing the movement options available once a decision has been made, but they may not fully translate to match-play agility where integrated perceptual-decision training or sport-specific open-skill practice is required. This distinction is important when interpreting the present COD findings, because the observed benefits are likely to reflect enhancement of the mechanical component of directional change rather than the full reactive agility construct (Secomb et al., 2015; Spiteri et al., 2014).

Regarding maximal strength, Traditional Strength Training ranked highest and showed a favorable effect compared with Regular Training. However, because 1RM was a pre-specified secondary outcome and the sensitivity analyses suggested less stable effect estimates than for the co-primary outcomes, this finding should be interpreted as supportive rather than definitive. In addition, possible small-study effects were detected for 1RM, further reducing confidence in the apparent magnitude of benefit.

Based on the SUCRA hierarchies and effect sizes established in this study, several practical implications may be considered. However, it is imperative to acknowledge that in applied soccer environments, these training modalities are rarely implemented in isolation. Instead, they are typically integrated into a concurrent training framework alongside technical-tactical sessions and varying team training loads (Enright et al., 2015; Morgans et al., 2014). Thus, the observed effects should be interpreted as the added contribution of a given training emphasis rather than as stand-alone prescriptions for team-based practice.

Practitioners may use the present findings to guide which training component receives greater emphasis within this broader training context. When short-distance acceleration, braking capacity, and aerial or force-dominant actions are prioritized, greater emphasis on Traditional Strength Training and, in some cases, Plyometric Jump Training may be warranted. When longer sprint exposure is a specific priority, speed-oriented methods may reasonably receive greater emphasis, although the current 30-40 m evidence remains uncertain. In many practical settings, a balanced multi-modal approach will remain the most appropriate strategy, with final decisions shaped by training age, competitive schedule, current readiness, and positional or tactical demands. Where match-relevant agility is a priority, improvements in planned COD ability should ideally be complemented by open-skill drills involving perceptual and decision-making demands, because gains in closed-skill directional change tests may not fully capture reactive agility performance in competition.

Despite the comprehensive nature of this NMA, several limitations must be acknowledged. First, although we refined the intervention taxonomy using explicit operational criteria and a priori decision rules, some residual heterogeneity within nodes may remain. Second, while neither the local node-splitting analyses nor the global design-by-treatment interaction model identified statistically significant inconsistency, these findings should be interpreted as supportive rather than conclusive, particularly because several loops were sparse. Third, the included population was restricted mainly to male professional or academy-level players; therefore, generalizability to women’s soccer and lower or different competitive levels remains uncertain. Fourth, the present network evaluated exercise modalities as stand-alone intervention categories and does not directly inform optimal sequencing or periodization within longer-term training plans. Fifth, outcome assessment was limited to immediate post-intervention effects, so the durability of adaptation could not be determined. Finally, for sprint outcomes, some residual measurement heterogeneity related to timing technology, start procedure, and test surface cannot be excluded, especially in the sparser 30 m and 40 m networks.

To strengthen the practical relevance of future evidence, randomized trials should move beyond isolated modality comparisons and more directly evaluate commonly used modality combinations and training sequences. At a minimum, future studies should report concurrent soccer load, adherence, injury or adverse-event data, and clearer outcome timing relative to the intervention and competitive schedule. Greater attention should also be given to women’s soccer, developmental cohorts, and position-specific reporting. In addition, duration-stratified or adequately powered moderator analyses are needed to test whether shorter and longer programs differ meaningfully in their comparative effects.

This systematic review and network meta-analysis suggests that, among the pre-specified co-primary outcomes, Traditional Strength Training may provide the most consistent benefits for short-distance acceleration (5-20 m) and vertical jump performance in soccer players, although certainty varied across comparisons. Evidence for longer sprint distances (30-40 m) was more uncertain, and no intervention was statistically superior to Regular Training at 40 m. Secondary outcomes (COD and 1RM) were interpreted as supportive rather than primary evidence, and the 1RM finding should be viewed cautiously because of less stable sensitivity results and possible small-study effects. Overall, the present findings support cautious consideration of strength-oriented training within a broader multi-modal conditioning approach when short-sprint acceleration and vertical jump performance are prioritized; however, the conclusions should be applied cautiously beyond predominantly male professional or academy-level samples and should not be taken as direct evidence for the optimal sequencing or periodization of strength, plyometric, and speed-oriented methods.