3.1 Search results
The initial search yielded 6,614 studies, and we observed that using different search terms to narrow them could overlook some of the of the most relevant research. Therefore, we adopted a more conservative approach. Identified records were filtered using keywords in titles and abstracts through the systematic review software Rayyan [42] (https://www.rayyan.ai/, accessed in March 2024). Additionally, reference screening was performed on the most relevant studies, and citation tracking was conducted using Research Rabbit software [27] (https://www.researchrabbit.ai/, accessed in March 2024), which yielded an additional 50 potentially useful studies. Our systematic review and meta-analysis ultimately included 27 studies. Of these, 11 studies reported changes in all three strength measures (MVCECC, MVCCON, and MVCISO) following both concentric-only and eccentric-only training. However, 16 studies did not report changes in MVCISO. In total, we gathered 162 study results (expressed as changes in percent units). This enabled us to calculate 71 standardized effect sizes for differences in improvements of MVCECC and MVCCON between both training groups (71 results in each group) and 20 standardized effect sizes for differences in improvements of MVCISO between training groups (20 results in each group). To aid interpretation of the findings, results of multiple studies within each training group and contraction mode were summarized and compared in separate meta-analyses. The stages of the search and study selection process are presented in Fig. 1.
3.2 Study characteristics
Individual study characteristics are presented in Table 1. Summary of study characteristics reporting MVCCON and MVCCON by categorical subgroup variables are presented in Table 2. The summary of study characteristics by categorical subgroup variables for the studies reporting MVCCON, MVCCON and additionally MVCISO are presented in Table 3. Additionally, in studies reporting MVCCON and MVCECC (Table 2), the number of repetitions per set ranged from 1 to 15 (mode = 10; mean ± SD = 9.6 ± 2.4), the number of sets ranged from 1 to 7 (mode = 3; mean ± SD = 4.7 ± 1.6), the number of training session per week ranged from 1 to 5 (mode = 3; mean ± SD = 2.8 ± 0.6) and training protocol duration ranged from 4 to 20 weeks (mode = 6; mean ± SD = 7.8 ± 3.8). The mean age of the participants was 25 years (SD = 4.7; range 20-38). The number of participants was 364 for the eccentric training group and 354 for the concentric training group. When summarizing multiple results from the same studies, the totals were 904 and 899, respectively.
** Table 1 around here **
** Table 2 around here **
In the studies where MVCISO was reported (Table 3), the number of repetitions per set ranged from 1 to 15 (mode = 10; mean ± SD = 9.0 ± 3.1), the number of sets ranged from 1 to 6 (mode = 3; mean ± SD = 3.7 ± 1.2), the number of training session per week ranged from 1 to 5 (mode = 3; mean ± SD = 2.9 ± 0.9) and training protocol duration ranged from 4 to 12 weeks (mode = 6; mean ± SD = 6.4 ± 2.0). The mean age of the participants was 23.3 years (SD = 1.3; range 21-28). The number of participants was 111 for the eccentric training group and 110 for the concentric training group. When summarizing multiple results from the same studies, the totals were 215 and 215, respectively.
** Table 3 around here **
3.3 Quality of evidence and completeness of reporting
PEDro scale values and completeness of reporting of the controlled randomized study items are presented for each particular study in Table 1. PEDro scores ranged from 5 to 9 (mode = 6; mean ± SD = 6.3 ± 1.0) indicating low to moderate risk of bias (Table S1 of the ESM). As shown in Figure 2, all studies reported the execution of the testing and training procedures, but only 19% reported who performed the training or testing protocol, 44% clarified where the training and testing was performed, 15% clarified if sample size was calculated, and 37% of the studies reported information regarding the dropout of the participants. None of the studies reported tailoring and/or modifications of training protocols (Table S2 of the ESM).
Altogether, due considerable heterogeneity (Q test results) and publication bias (Egger’s statistics; further reported in text following the results of the particular meta-analysis results and summarized in Table S3 of the ESM), quality of evidence were downgraded from high quality to low quality according to GRADE approach [36] for MVCECC results. Additionally, due to imprecision (confidence interval was crossed by a small effect size), MVCCON and MVCISO results were downgraded to very low quality.
3.4 Meta-analyses results
3.4.1 Effect of different resistance training modes on eccentric strength gain (MVCECC)
A meta-analysis of 27 studies with 71 comparisons showed statistically significantly beneficial effects of eccentric-only in comparison to concentric-only strength training for improvement of MVCECC (Hedge’s g = 2.03, 95% CI: 0.74 to 3.32; p < 0.01; very large effect) (Figure S1 of the Electronic Supplementary Material [ESM]). A small sample adjustment to the robust meta-analysis results expanded the confidence intervals of Hedge’s to 0.69 to 3.39. Five individual effect sizes from two studies were identified as highly influential. However, the overall results were robust to their exclusion from the model as the interpretation of the model did not change. Overall Hedge’s g decreased to 1.51 (95% CI: 0.59 to 2.42; p < 0.01; large effect), with the small sample adjustment for the 95% CI being 0.55 to 2.47 (Fig. 3).
Egger’s test results indicated publication bias for the MVCECC (p<0.01) meta-analysis indicating smaller studies showing higher benefits in favour of eccentric or concentric training, respectively. Moreover, statistically significant overall heterogeneity among the studies was found (Q = 647.9, df = 65; p < 0.01; I2 = 96.5%). Within study effect size variability (Level 2) was low to moderate (29%), while between study variability (Level 3) was moderate to high (68%).
3.4.2 Effect of resistance training modes on concentric strength gain (MVCCON)
A meta-analysis of 27 studies with 71 comparisons did not show statistically significantly different benefits of eccentric-only and concentric-only strength training for improvement of MVCCON (Hedge’s g = –0.71, 95% CI –1.65 to 0.22; p = 0.13; small effect) (Figure S2 of the ESM). A small sample adjustment to the robust meta-analysis results expanded the confidence intervals of Hedge’s to –1.69 to 0.27. Five individual effect sizes from three studies were identified as highly influential. However, the overall results were robust to their exclusion from the model as the interpretation of the model did not change. Overall Hedge’s g decreased to trivial, i.e. –0.10 (95% CI: –0.69 to 0.48) with small sample adjustment 95% CI ranging from –0.72 to 0.51 (Fig. 4).
Egger’s test results indicated publication bias for the MVCCON (p<0.01) meta-analysis indicating smaller studies showing higher benefits in favour of eccentric or concentric training, respectively. Moreover, statistically significant overall heterogeneity among the studies was found (Q = 358.4, df = 65; p<0.01; I2 = 92%). Within study effect size variability (Level 2) was low (21%), while between study variability (Level 3) was moderate to high (71%).
3.4.3 Effect of resistance training modes on isometric strength gain (MVCISO)
A meta-analysis of 11 studies with 20 comparisons did not show statistically significantly different effects between eccentric and concentric resistance training protocols for improvement of MVCISO (Hedge’s g = –0.31, 95% CI –2.40 to 1.75; p = 0.77; small effect) (Figure S3 of the ESM). A small sample adjustment to the robust meta-analysis results expanded the confidence interval (Hedge’s g = –0.31, 95% CI –2.65 to 2.02). Two individual effect sizes from two studies were identified as highly influential. After removing them from the analysis, the overall effect size changed sign and narrowed the confidence intervals from negative (favouring concentring training) to positive trivial (Hedge’s g: 0.04 with 95% CI: –0.82 to 0.90), favouring eccentric training (Fig. 5). Confidence intervals (95%) for robust meta-analysis with small samples adjustment ranged from –0.96 to 1.05.
Egger’s test results indicated publication bias for the MVCISO (p<0.01) meta-analysis indicating smaller studies showing higher benefits in favour of eccentric or concentric training, respectively. Moreover, statistically significant overall heterogeneity among the studies was found (Q = 120.7, df = 25; p<0.01; I2 = 89%). Within study effect size variability (Level 2) was low (11%), while between study variability (Level 3) was moderate to high (71%).
3.4.4 Analyses of moderators
Despite high statistical heterogeneity among the studies comparing the magnitude of increase of MVCECC between eccentric and concentric training, no statistically significant differences were found within subgroups of sex (Chi2 = 3.2; p = 0.20; pseudo R2 = 2.7%), muscle (Chi2 = 0.59; p = 0.44; pseudo R2 = 6%), and training status (Chi2 = 0.89; p = 0.83; pseudo R2 = 18%). Effect of eccentric-only training was more superior when testing was performed at the same velocity as training in comparison to when testing was performed at the lower velocity as the training (difference of 1.53 in Hedge’s g, p<0.05), nevertheless training to testing isokinetic velocity subgroups together could not explain the variability of the effect (Chi2 = 7.2; p = 0.07; pseudo R2 = 1%). Participant-related continuous moderators analyses showed no effect of age (Chi2 = 3.0; p = 0.08; pseudo R2 = 12.7%), initial MVCECC (Chi2 = 0.16; p = 0.69, pseudo R2 = 10%), initial MVCCON (Chi2 = 0.55; p = 0.46, pseudo R2 = 12%) and MVCECC/MVCCON ratio (Chi2 = 2.05; p = 0.15; pseudo R2 = 12%). Moreover, training related variables included in multiple meta-regression (number of repetitions, sets, frequency and duration of training), could not statistically significantly explain the variability (Chi2 = 8.2; p = 0.08; pseudo R2 = 19.5%). Within training-related factors, only duration of the training protocol had shown statistically significant influence to the effect size (β = 0.25 [95% CI: 0.05-0.45]; SE = 0.10; z = 2.4; p < 0.05). Thus, the longer the training protocol, the more beneficial the eccentric training was over concentric training for improving MVCECC when controlling for the rest of training-related factors (Table S4 of the ESM).
Comparing the magnitude of increase of MVCCON between eccentric and concentric training, no statistically significant differences were found between subgroups of sex (Chi2 = 3.90; p = 0.14; pseudo R2 = 4.4%), muscle (Chi2 = 0.00; p = 0.98; pseudo R2 = 10%) and training status (Chi2 = 1.89; p = 0.60; pseudo R2 = 21%). Statistically significant differences were found within subgroups of training to testing isokinetic velocity (Chi2 = 8.4; p < 0.05; pseudo R2 = 49%). A statistically significant lower effect of eccentric training compared to concentric training on the improvement of MVCCON was observed when the eccentric training was performed at higher velocities than those used in the MVCCON testing in comparison to the effect of the same training and testing velocity (Hedge’s g = –0.99 [95% CI from –1.75 to –0.23]; p < 0.05). Participant-related continuous moderators showed no effect of age (Chi2 = 0.68; p = 0.41; pseudo R2 = 13%), initial MVCECC (Chi2 = 0.08; p = 0.77; pseudo R2 = 14.5%), initial MVCCON (Chi2 = 0.19; p = 0.66; pseudo R2 < 19.7%) and MVCECC/MVCCON ratio (Chi2 = 1.84; p = 0.17; pseudo R2 = 5.2%). Moreover, training related variables (number of repetitions, sets, frequency and duration of training), could not statistically significantly explain the variability (Chi2 = 5.7; p = 0.22; pseudo R2 = 40.7 %) (Table S5 of the ESM).
Among the studies comparing the magnitude of increase of MVCISO between eccentric and concentric training, no statistically significant differences were found within subgroups of sex (Chi2 = 0.02; p = 0.99; pseudo R2 = 47%), muscle (Chi2 = 0.20; p = 0.66; pseudo R2 = 18%) and training status (Chi2 = 2.85; p = 0.24; pseudo R2 = 9%). Participant-related continuous moderators showed no effect of age (Chi2 = 0.3; p = 0.60; pseudo R2 = 14.7%), initial MVCECC (Chi2 = 0.10; p = 0.75, pseudo R2 = 3.6%), initial MVCCON (Chi2 = 0.5; p = 0.47, pseudo R2 = 29%) and MVCECC/MVCCON ratio (Chi2 = 0.17; p = 0.68; pseudo R2 = 31%). Moreover, training related variables (number of repetitions, sets, frequency and duration of training), could not statistically significantly explain the variability (Chi2 = 0.65; p = 0.96; pseudo R2 = 14%) (Table S6 of the ESM).
3.4.5 Changes in strength within training group
As shown in Fig. 6, results from individual studies indicate that eccentric-only training resulted in an improvement of MVCECC by 27.3% (95% CI: 19.4-35.2%; p<0.05; robust 95% CI: 18.2-36.4%), MVCCON by 12.6% (95% CI: 8.6-16.7%; p<0.05; robust 95% CI: 8.0-17.3%), and MVCISO by 18.2% (95% CI: 11.7-24.7%; p<0.05; robust 95% CI: 10.2-26.2%) (Table S7 of the ESM).
Improvements after concentric-only training were observed as follows: MVCECC increased by 10.2% (95% CI: 8.2-12.3%; p<0.05; robust 95% CI: 7.2-13.2%), MVCCON by 13.8% (95% CI: 10.0-17.5%; p<0.05; robust 95% CI: 9.7-17.9%), and MVCISO by 16.9% (95% CI: 9.5-24.2%; p<0.05; robust 95% CI: 7.8-25.9%) (Table S7 of the ESM).
No influential individual results were identified, and Egger’s test results showed no publication bias, with p-values ranging from 0.08 for MVCCON after concentric training to 0.797 for MVCISO after eccentric training. Statistically significant heterogeneity was confirmed by Q-test (p<0.05) in all cases, with I2 values exceeding 98% in all cases. Specifically, level 2 I2 varied from 0% for MVCISO after concentric training to 27% for MVCCON after concentric training, while level 3 I2 ranged from 72% for MVCCON after concentric training to 99% for MVCISO after concentric training (Table S7 of the ESM).
3.4.5 Differences in strength changes within training group and between groups
Pairwise comparisons revealed differences in strength improvement among testing contraction modes following eccentric-only training: MVCECC versus MVCCON showed an 11.5% difference (95% CI: 11.2-11.7; robust 95% CI: –1.9 to 24.8; p<0.05), illustrated in Fig. 6. The difference between MVCECC and MVCISO was 8.6% (95% CI: 8.0-9.3; robust 95% CI: –44.4 to 61.7; p<0.05), and between MVCCON and MVCISO was 3.8% (95% CI: 3.3-4.3; robust 95% CI: –5.7 to 13.2; p<0.05) (Table S8 of the ESM).
For concentric-only training, the respective differences were also notable: MVCECC versus MVCCON at 1.7% (95% CI: 1.5-1.9; robust 95% CI: 4.1-7.5; p<0.05), MVCECC versus MVCISO at 6.2% (95% CI: 5.7-6.7; robust 95% CI: 1.4-11.0; p<0.05), and MVCCON versus MVCISO at 2.1% (95% CI: 1.7-2.6; robust 95% CI: –7.0 to 11.2; p<0.05) (Table S8 of the ESM).
No influential individual results sizes were detected. Egger’s test revealed publication bias (p<0.10) in all comparisons except for MVCECC versus MVCISO after concentric training (p=0.44), and statistically significant heterogeneity was confirmed in all cases with Q-test’s Chi2 statistics (p<0.05). I2 values exceeded 90% for all comparisons. Specifically, level 2 (within-study) I2 ranged from 38.6% for the MVCCON and MVCISO comparison for concentric training to 66.4% for the MVCECC and MVCISO comparison post concentric training. Level 3 (between-study) I2 ranged from 32.3% for the MVCECC vs MVCISO comparison post concentric training to 62% for the MVCECC vs MVCCON comparison post eccentric training (Table S8 of the ESM).