Abstract
Objective
This study aimed to determine (1) the effects of resistance training combined with creatine supplementation on muscle strength (upper- and lower-body strength) and body composition (lean mass, fat mass, body-fat percentage, and body mass), and (2) whether these effects were moderated by participant characteristics (age, gender, and training experience) and intervention protocols (intervention duration, weekly frequency, dosing strategy, daily dose, and supplementation protocol).
Methods
We conducted a systematic search of PubMed, Web of Science, Cochrane Library, Embase, and EBSCO from database inception through April 21 2025, and performed backward citation tracking to identify additional relevant trials. Hedge's g (g) served as the summary effect size for each outcome. Meta-analysis was performed using a three-level random-effects model in the metafor package (R). Potential moderators were explored via subgroup analyses and meta-regression.
Results
Sixty-three studies (n = 1489) were included. Compared with resistance training alone (RT), resistance training plus creatine (RT + CR) produced significant increases in upper-body strength (g = 0.54, p < 0.001), lower-body strength (g = 0.55, p < 0.001), lean mass (g = 0.30, p < 0.001), and body mass (g = 0.12, p = 0.04), and a significant reduction in body-fat percentage (g = −0.14, p = 0.02). No significant difference emerged for fat mass (g = −0.10, p = 0.14). Exploratory subgroup analyses suggested that age may be associated with strength gains, with statistically significant between-subgroup differences observed for both upper- and lower-body strength. These findings should be interpreted cautiously because multiple moderator analyses were conducted and the age subgroups were unevenly represented. Exploratory meta-regression also suggested an inverse association between age and strength effects, with larger effects observed in studies involving younger participants. Neither subgroup nor meta-regression analyses identified consistent moderators for body composition outcomes.
Conclusion
Compared with resistance training alone, resistance training combined with creatine supplementation significantly enhanced muscle strength and lean body mass, without a significant effect on fat mass. The modest increase in total body weight was primarily attributable to gains in lean body mass, contributing to a slight reduction in body fat percentage. Among participant characteristics, age appeared to be associated with the magnitude of strength effects, although this exploratory finding should be interpreted cautiously given the uneven representation of age subgroups. Intervention-protocol variables did not significantly influence the outcomes. These conclusions should also be interpreted within the context of heterogeneous and incompletely reported nutritional intake and training backgrounds across the included trials.
Keywords: Resistance training, creatine, strength, body composition, meta-analysis
1. Introduction
Resistance training (RT) is widely recognized as an effective strategy for enhancing muscle strength and improving body composition, and it is commonly employed to counteract sarcopenia and boost athletic performance. In recent years, nutritional interventions designed to amplify training adaptations have garnered considerable interest, with creatine supplementation standing out for its potent ergogenic effects [1]. Creatine is a nitrogenous organic compound stored predominantly in skeletal muscle (approximately 95% of total body creatine, of which about two-thirds exists as phosphocreatine [PCr]) [2,3]. Its primary mechanism involves increasing intramuscular PCr stores, thereby facilitating rapid adenosine triphosphate (ATP) resynthesis during short bouts of high‐intensity exercise. This enables individuals to sustain higher power outputs and complete additional repetitions, ultimately translating into greater training volume and more pronounced gains in strength and hypertrophy [4,5]. Moreover, creatine's cell‐hydration effect activates anabolic signaling pathways—such as upregulation of insulin‐like growth factor-1 (IGF-1) and satellite cell proliferation—further promoting muscle protein synthesis and fiber growth [6,7]. On the basis of extensive evidence, the International Society of Sports Nutrition (ISSN) endorses creatine as one of the most effective and safest dietary supplements for enhancing anaerobic performance and supporting skeletal muscle adaptations [8].
A substantial body of literature has demonstrated that RT combined with creatine supplementation (RT+CR) yields significantly greater improvements in muscle strength and lean mass compared with RT alone [9–12]. However, previous meta-analyses exhibit several critical limitations. First, many syntheses focus on a single outcome (e.g. lean mass) or narrow populations, failing to capture the full spectrum of training adaptations. Second, conventional two‐level random‐effects models typically include only one effect size per study or treat multiple outcomes as independent, which can lead to information loss and an overestimation of statistical significance. Third, moderator analyses in past work have largely been confined to simple subgroup comparisons (e.g. by age or sex), lacking a comprehensive multivariable evaluation of training and supplementation parameters [12,13].
To address these gaps, the present study introduces three key innovations: (1) three-level meta-analysis: decomposing total variance into sampling error (Level 1), within-study variance (Level 2), and between-study variance (Level 3), allowing simultaneous inclusion of multiple correlated outcomes from the same trial. (2) Integrated moderator analyses: systematically evaluating the influence of participant characteristics (e.g. age, sex) and intervention factors (e.g. training duration, weekly frequency, creatine dosage/regimen, training experience) on RT+CR effects. (3) Multidimensional outcome integration: synthesizing both upper- and lower-body strength and multiple body composition indices (body mass, lean mass, fat mass, body-fat percentage) to comprehensively characterize creatine's augmentation of resistance training adaptations.
By leveraging these methodological advances, this study aims to provide more robust and generalizable evidence of RT+CR efficacy, elucidate its differential effects across various conditions and populations, and inform sports nutrition interventions and muscle health promotion.
2. Methods
This meta-analysis was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) statement [14], and the protocol was prospectively registered in the PROSPERO database (registration number CRD420251078957).
2.1. Information sources
We systematically searched PubMed, Web of Science, Cochrane Library, Embase, and EBSCO from database inception through April 21 2025. We included only full-text, peer-reviewed articles published in English, with no restrictions on publication date or sample size. Reference lists of all included studies were hand-searched for additional eligible trials.
2.2. Search strategy
We employed the following Boolean search string: (“resistance training” OR “strength training” OR “resistance exercise”) AND (“creatine supplementation” OR “creatine intake” OR “creatine loading” OR “creatine monohydrate”) AND (“body composition” OR “fat mass” OR “lean mass” OR “muscle mass” OR “skeletal muscle” OR “percent body fat” OR "body fat percentage") AND (“muscle strength” OR “isometric strength” OR “1-RM” OR “one repetition maximum” OR “muscle power” OR “maximal strength”).
For example, in Web of Science we used: TS = ((“resistance training” OR “strength training” OR “resistance exercise”) AND (“creatine supplementation” OR “creatine intake” OR “creatine loading” OR “creatine monohydrate”) AND (“body composition” OR “fat mass” OR “lean mass” OR “muscle mass” OR “skeletal muscle” OR “percent body fat” OR “body fat percentage”) AND (“muscle strength” OR “isometric strength” OR “1-RM” OR “one repetition maximum” OR “muscle power” OR “maximal strength”)).
2.3. Selection process
All retrieved records were imported into EndNote X9 and deduplicated by one reviewer. Two independent reviewers then screened titles and abstracts against the predefined eligibility criteria. Disagreements were resolved by consultation with a third reviewer. Full texts of potentially eligible studies were assessed independently by the same two reviewers to determine final inclusion.
2.4. Eligibility criteria
We applied PICOS criteria to define study eligibility.
Population (P): adults aged ≥18 years.
Intervention (I): resistance training combined with creatine supplementation.
Comparator (C): resistance training plus placebo (e.g. maltodextrin, dextrose, glucose).
Outcomes (O): at least one of the following—upper-body strength, lower-body strength, body mass, fat mass, body-fat percentage, or lean mass.
Study design (S): randomized controlled trials with a control group.
We excluded ecological studies, reviews, animal studies, observational studies, research on participants <18 years, and trials without randomization or a placebo/control condition. Trials in which creatine was co-ingested with other active supplements were also excluded, in order to isolate the effect of creatine supplementation as much as possible.
2.5. Data extraction
Two reviewers independently extracted data into a pre-piloted Excel form (version 16.93, Microsoft, Redmond, WA). Extracted information included authorship, study characteristics, participant demographics, intervention details, and outcome data (pre- and post-intervention means, standard deviations, and sample sizes). Discrepancies were resolved by consensus, without the need for third-party arbitration. When numerical data were missing or only presented graphically, we contacted authors; if unsuccessful, we digitized figures using WebPlotDigitizer 4.1 (https://automeris.io/WebPlotDigitizer). Studies with irretrievable data were excluded. Inter-rater agreement for extracted data was 96.52%.
2.6. Risk of bias assessment
Two independent reviewers assessed risk of bias using the Cochrane Risk of Bias tool for randomized controlled trials [15].
2.7. Quality of evidence
The certainty of evidence for each outcome was independently assessed by two reviewers using the GRADE framework, which evaluates risk of bias, consistency, directness, precision, and publication bias to classify evidence as “high,” “moderate,” “low,” or “very low” [16].
2.8. Statistical analysis
The analytic workflow is presented in Figure 1. and followed established guidelines and methods [17–22]. Hedges' g was calculated as the effect size metric [23], and effect sizes and variances were computed in Comprehensive Meta-Analysis (CMA) consistent with Aksayli et al. [24]. Subsequent meta-analyses were performed in R 4.1.2 (R Core Team, 2021) using the metafor package [25]. The effect‐size formula and detailed statistical procedures are described below.
Figure 1.

The data analysis workflow diagram.
Effect‐size calculation
The standardized mean difference (Hedges' g) was computed according to Borenstein et al. [23]:
where M e_post and M e_pre denote the experimental group's post-test and pre-test means, respectively; M c_post and M c_pre denote the control group's post-test and pre-test means, respectively; SDpooled_change is the pooled standard deviation of the pre–post change scores (post minus pre) across both groups. N represents the total sample size.
In repeated-measures designs, the most accurate formula for the sampling variance of Hedges' g requires the pre–post correlation coefficient [26]. However, because few of the included studies reported this coefficient, we followed Aksayli et al. [24] in using the following approximation:
2.8.1. Meta-analysis
In this study, some trials reported multiple outcome variables or presented results for one control group and multiple experimental groups, leading to dependency among effect sizes. To retain all available effect sizes and to address this dependency, multilevel or multivariate approaches are required [18–20]. Because most studies did not report the correlations needed for multivariate meta-analysis, we chose a multilevel meta-analysis. This method accounts for three levels of variance among effect sizes: sampling variance (Level 1), variance among effect sizes within the same study (Level 2), and variance between studies (Level 3).
First, we used a three‐level random-effects model to estimate the overall effect size (Hedges' g classified as small < 0.20; small‐to‐moderate 0.20 ≤ g < 0.50; moderate 0.50 ≤ g < 0.80; large ≥ 0.80) [27]. Following previous recommendations [17,18], we applied the “t” test and REML (restricted maximum likelihood) method in the rma.mv function. Given inconsistencies in participant characteristics and intervention features across studies, and our aim to generalize the findings to a broader population, we adopted a random‐effects model as recommended for model selection [23].
Second, we conducted an influence analysis to identify influential cases and assess their potential impact on the results [23]. An influential case was defined as an effect size whose exclusion substantially altered the fitted model. The meta‐analytic models were run both including and excluding these influential effect sizes.
Third, we used the Q statistic to test overall heterogeneity. Cochran's Q is calculated as the inverse‐variance–weighted sum of squared deviations of each effect size from the overall mean [28]. Simultaneously, we used I2 to estimate heterogeneity at Levels 1, 2, and 3, assessing variance distribution across levels [17]. Additionally, two log‐likelihood ratio tests evaluated whether the three‐level model explained data variability better than a two‐level model [17]. Specifically, we first set Level 2 variance to zero and fitted a two‐level model (Levels 1 and 3), comparing AIC and BIC between it and the full three‐level model. Then, we set Level 3 variance to zero and compared the full model to a two‐level model (Levels 1 and 2).
Fourth, for moderator analyses, we employed three‐level mixed‐effects models [23]. We used the omnibus Q test to determine whether subgroup differences exceeded what could be explained by sampling error alone [18]. When more than one moderator was statistically significant and sufficient data were available, significant moderators were considered in the same model to examine whether the associations persisted after mutual adjustment [18].
Moderator analyses included the following variables: participant characteristics: (a) age group (young adults [18–44 years] vs. middle-aged adults [45–59 years] vs. older adults [≥60 years]); gender (male vs. female); and training experience (athletes vs. resistance-trained vs. untrained). Intervention parameters: intervention duration (≥8 weeks vs. <8 weeks); weekly training frequency (>3 sessions/week vs. ≤3 sessions/week); creatine dosing strategy (fixed dose vs. weight-based); daily creatine dose (>5 g/day vs. ≤5 g/day); and creatine protocol (loading only vs. loading + maintenance vs. maintenance only).
These cut-points were selected before the moderator analyses to balance prior literature, practical interpretability, and the distribution of the included trials. Intervention duration was categorized as ≥8 weeks vs. <8 weeks because the included resistance-training programs ranged from 4 to 12 weeks and an 8-week threshold is commonly used to distinguish shorter from longer training interventions. Weekly training frequency was categorized as >3 vs. ≤3 sessions per week because three sessions per week represented a common training frequency in the included trials, whereas more than three sessions per week reflected higher-frequency training. Daily creatine intake was categorized as >5 g/day vs. ≤5 g/day because 3–5 g/day is commonly used as a maintenance dose of creatine monohydrate, whereas doses above 5 g/day reflect higher daily intake. These thresholds were used for exploratory subgrouping and were not intended to define definitive physiological or clinical cut-offs.
Because moderator analyses were conducted across multiple outcomes and several participant- and intervention-level variables, these analyses were considered exploratory and hypothesis-generating rather than confirmatory. No formal adjustment for multiple testing was applied; therefore, statistically significant moderator findings were interpreted cautiously, with emphasis placed on between-subgroup tests and consistency across outcomes rather than isolated within-subgroup statistical significance. To aid interpretation of subgroup precision, moderator findings were interpreted alongside subgroup representation. In the tables, K denotes the number of modeled effect sizes rather than the number of unique studies because the three-level model retained multiple dependent effects from some trials; in the Results, participant totals are additionally reported for the age subgroups, which were the only statistically significant moderator comparisons.
Finally, to assess publication bias, we used funnel plots [29], Egger's regression test [29], and the trim‐and‐fill method [30]. Funnel plots and trim‐and‐fill analyses were based on a two‐level model treating all effect sizes as independent—since these methods have not yet been extended to multilevel models [21,22]. Egger's regression test was conducted for both two‐ and three‐level models; in the three‐level analysis we used the recommended predictor 2√N (where N is total sample size) [21,22].
3. Results
3.1. Search results
Database and citation searches identified 3472 records, including 3462 records from five databases and 10 records from reference lists. After 1668 duplicate records were removed, 1804 records remained for title and abstract screening. Following this screening, 1685 records were excluded. Full texts of the remaining 119 reports were assessed for eligibility, of which 56 were excluded with reasons, leaving 63 studies for quantitative analysis (Figure 2).
Figure 2.

PRISMA 2020 flow diagram of study selection.
3.2. Characteristics of included studies
A total of 63 studies involving 1489 participants were included. Of these, 970 were male, 248 were female, and 271 did not report gender. Individual study sample sizes ranged from 8 to 45 participants, with ages spanning 18 to 75 years. Regarding training history, 28% of participants were resistance‐trained, 49% were untrained, and 23% were athletes. Intervention durations ranged from 4 to 12 weeks, with training frequencies of 3–5 sessions per week. All studies used creatine monohydrate. Supplementation protocols were classified as loading only (7%), loading+long maintenance (54%), and maintenance only (39%). Detailed participant and intervention characteristics are presented in Appendix 1, Tables 1 and 2. Because some studies contributed more than one dependent effect size, study counts, effect-size counts, and participant totals are reported separately where relevant in the Results.
3.3. Synthesis of results
3.3.1. Upper‐limb muscle strength
-
(1)
Overall effect of RT+CR on upper‐limb strength
A total of 31 studies (57 effect sizes; N = 1273) were included. The three-level random-effects model yielded a moderate overall effect of resistance training plus creatine (RT+CR) on upper‐limb muscle strength (g = 0.54, 95% CI [0.37, 0.71]; 95% PI [–0.13, 1.20]; p < 0.001) (see Figure 3). Results from a two-level random-effects model were similar (g = 0.56, 95% CI [0.43, 0.69]; 95% PI [0.04, 1.08]; p < 0.001). The heterogeneity test indicated significant between-effect variance (Q (56) = 80.18, p = 0.019). These findings suggest that RT+CR was associated with greater improvements in upper‐limb strength than RT alone, although the 95% prediction interval crossed zero and included negative values, indicating that the magnitude and direction of effects may vary across future study settings.
Figure 3.

Primary pooled effect sizes for the outcomes. 95%CI, 95% confidence interval; GRADE, grading of recommendations assessment, development, and evaluation (a system for evaluating the quality of evidence and strength of recommendations); Hedge's g, the effect size indicators used in the pooled; K, the total number of effects included in the pooled effect size; p-value, statistically significant p-values for pooled results; RT: resistance training; RT + CR: resistance training combined with creatine supplementation.
-
(2)
Influence analysis
An influence analysis following Viechtbauer and Cheung [31] found no single study exerted a disproportionate impact on the pooled effect estimate, indicating robust overall results (Appendix 3, Figure 2).
-
(3)
Heterogeneity analysis
The main effect analysis showed significant heterogeneity (Q (56) = 80.18, p = 0.019). Of the total variance, sampling variance (Level 1) accounted for 63.41%, within-study variance (Level 2) was <1%, and between-study variance (Level 3) accounted for 36.59%. Likelihood-ratio tests demonstrated significant Level 3 variance (χ2 = 9.03, p = 0.003, one-tailed), indicating the three-level model outperformed a two-level model with Level 3 variance constrained to zero. Level 2 variance was not significant (χ2 < 0.01, p > 0.49, one-tailed), suggesting no clear advantage of including Level 2 variance over a two-level model omitting it. Nonetheless, retaining the three-level structure is warranted because it more accurately reflects the nested data structure when multiple effect sizes per study are included [17] (see Figure 3).
-
(4)
Moderator analyses
Among all moderators tested, age was the only moderator showing a statistically significant between-subgroup difference (F (2,54) = 3.38, p = 0.04). Specifically, subgroup-specific estimates favored RT+CR over RT alone in younger adults (K = 43 effect sizes; N = 874; g = 0.62, p < 0.001) and older adults (K = 10 effect sizes; N = 291; g = 0.52, p = 0.004), whereas the effect in middle-aged adults (K = 4 effect sizes; N = 108; g = −0.10, p = 0.71) did not reach statistical significance. Pairwise contrasts indicated the middle-aged group's effect was significantly lower than that of young (β = –0.36, t (54) = –2.60, p = 0.01) and older adults (β = –0.31, t (54) = –1.99, p = 0.05) (see Table 1). However, this finding should be interpreted cautiously, as the distribution of studies across age categories was uneven, with relatively limited representation of middle-aged participants. The small number of modeled effects and participants in the middle-aged subgroup indicates limited precision and reduced statistical power for this comparison. No other subgroup analyses by gender, intervention duration (≥8 w vs. <8 w), weekly frequency (>3 vs. ≤3 sessions/wk), dosing strategy (fixed vs. weight-based), daily dose (>5 g/d vs. ≤5 g/d), supplementation protocol (loading only vs. loading+maintenance vs. maintenance only), or training experience (athlete vs. trained vs. untrained) reached statistical significance (all p > 0.05; see Table 1). Some within-subgroup estimates were statistically significant; however, in the absence of significant between-subgroup differences, these findings were interpreted descriptively rather than as evidence of effect modification. Given the number of moderator analyses conducted, the age-related finding should also be regarded as exploratory and hypothesis-generating.
Table 1.
Results of moderator analyses on the effects of resistance training combined with creatine supplementation on upper‐limb muscle strength.
| Moderator | K | Hedge's g | 95% CI | F | df | P | Level 2 variance (I2) | Level 3 variance (I2) |
|---|---|---|---|---|---|---|---|---|
| Mean age | 3.38 | 2 | 0.04 | <1% | 30.09%** | |||
| Young adults (18–44 years) | 43 | 0.62 | [0.43, 0.81] | <0.001 | ||||
| Middle adults (45–59 years) | 4 | −0.1 | [−0.62, 0.42] | 0.71 | ||||
| Older adults (≥60 years) | 10 | 0.52 | [0.17, 0.88] | 0.004 | ||||
| Gender | 0.09 | 1 | 0.76 | <1% | 44.79%* | |||
| Male | 37 | 0.55 | [0.32, 0.79] | <0.001 | ||||
| Female | 4 | 0.45 | [−0.18, 1.09] | 0.16 | ||||
| Training status | 0.09 | 2 | 0.92 | <1% | 39.5** | |||
| Athletes | 13 | 0.5 | [0.16, 0.83] | 0.005 | ||||
| Trained | 20 | 0.51 | [0.19, 0.83] | 0.002 | ||||
| Untrained | 24 | 0.58 | [0.31, 0.84] | <0.001 | ||||
| Intervention duration | 0 | 1 | 0.95 | <1% | 38.2%** | |||
| Long intervention (≥8 w) | 38 | 0.54 | [0.33, 0.74] | 0.002 | ||||
| Short intervention (< 8w) | 19 | 0.53 | [0.21, 0.85] | <0.001 | ||||
| Training frequency | 0.01 | 1 | 0.93 | <1% | 38.1%** | |||
| High frequency (>3/pw) | 19 | 0.55 | [0.26, 0.83] | <0.001 | ||||
| Low frequency (≤3/pw) | 38 | 0.53 | [0.31, 0.74] | <0.001 | ||||
| Creatine loading | 0.57 | 2 | 0.55 | <1% | 37.95** | |||
| Loading + maintenance | 28 | 0.57 | [0.34, 0.81] | <0.001 | ||||
| Maintenance | 23 | 0.54 | [0.28, 0.81] | <0.001 | ||||
| Loading only | 6 | 0.25 | [−0.33, 0.82] | 0.39 | ||||
| Creatine intake forms | 0.5 | 1 | 0.48 | <1% | 37.62** | |||
| Fixed dose (grams) | 23 | 0.47 | [0.22, 0.72] | <0.001 | ||||
| Weight-based dosing | 34 | 0.59 | [0.36, 0.83] | <0.001 | ||||
| Creatine dose | 0 | 1 | 0.95 | <1% | 39.29%** | |||
| High dose (>5 g/d) | 18 | 0.57 | [0.28, 0.86] | <0.001 | ||||
| Low dose (≤ 5 g/d) | 33 | 0.56 | [0.32, 0.79] | <0.001 |
Note: Hedges' g, the effect size indicator used in the pooled analysis; CI: confidence interval; K: represents the number of effect sizes. **p < 0.01. *p < 0.05; F: the F statistic from ANOVA, used to test for group differences. df: degrees of freedom for the F test. K should not be interpreted as the number of unique studies, because several studies contributed more than one dependent effect size to the three-level model; subgroup precision should therefore be interpreted together with participant totals reported in the text.
-
(5)
Publication bias
Funnel plot inspection (two-level model) showed mild asymmetry (Figure 4), suggesting possible bias. Egger's test indicated significant bias (t (55) = 5.32, p < 0.001). A multilevel Egger test with the recommended predictor still detected bias (t (55) = 2.90, p = 0.005). Trim-and-fill analysis estimated ∼15 missing effect sizes on the left; after imputing these, the adjusted overall effect remained significant (g = 0.36, 95% CI [0.21, 0.51], p < 0.001), though it was attenuated compared to the unadjusted g = 0.54 (95% CI [0.37, 0.71], p < 0.001). Thus, although publication bias may be present, its impact on conclusions appears limited.
Figure 4.

Funnel plot of the publication bias analysis for the effects of creatine combined with resistance training on upper‐limb muscle strength.
3.3.2. Lower‐limb muscle strength
-
(1)
Overall effect of RT+CR on lower‐limb strength
A total of 34 studies (49 effect sizes; N = 1094) were included. The three‐level random‐effects model yielded a moderate overall effect of resistance training plus creatine (RT+CR) on lower‐limb muscle strength (g = 0.66, 95% CI [0.47, 0.85]; 95% PI [–0.37, 1.69]; p < 0.001) (see Figure 3). Results from a two‐level random‐effects model were identical (g = 0.66, 95% CI [0.47, 0.85]; 95% PI [–0.37, 1.69]; p < 0.001). The heterogeneity test indicated significant between‐effect variance (Q (48) = 117.51, p < 0.001). These findings demonstrate a clear advantage of RT+CR over RT alone for enhancing lower‐limb strength, although the 95% prediction interval suggests that future studies may observe effects ranging from non‐significant to negative.
-
(2)
Influence analysis
An influence analysis identified one outlier (Schweizer, 2021; g = 3.77; see Appendix 4, Figure 2). Excluding this effect size and re‐running the meta‐analysis produced a still‐significant overall effect (three‐level: g = 0.55, 95% CI [0.38, 0.71]; 95% PI [–0.04, 1.14]; p < 0.001; two‐level: g = 0.56, 95% CI [0.42, 0.69]; 95% PI [0.15, 0.97]; p < 0.001). To avoid undue influence, all subsequent analyses omitted this outlier.
-
(3)
Heterogeneity analysis
After excluding the outlier, heterogeneity among effect sizes was no longer significant (Q (47) = 54.72, p = 0.20). Sampling variance (Level 1) accounted for 69.18% of total variance, within‐study variance (Level 2) was <1%, and between‐study variance (Level 3) accounted for 30.82%. Likelihood‐ratio tests showed significant Level 3 variance (χ2 = 4.37, p = 0.037, one‐tailed), indicating the three‐level model outperformed a two‐level model with Level 3 variance constrained to zero; Level 2 variance remained non‐significant (χ2 < 0.01, p > 0.49, one‐tailed), showing no clear advantage of including Level 2 variance (see Figure 3).
-
(4)
Moderator analyses
Among the tested moderators, age was the only moderator showing a statistically significant between-subgroup difference for lower-limb strength (F (2,45) = 4.97, p = 0.01). Specifically, subgroup-specific estimates favored RT+CR over RT alone in younger adults (K = 35 effect sizes; N = 725; g = 0.65, p < 0.001) and older adults (K = 9 effect sizes; N = 261; g = 0.52, p = 0.001), whereas the effect in middle-aged adults (K = 4 effect sizes; N = 108; g = −0.10, p = 0.65) did not reach statistical significance. Pairwise contrasts indicated the middle-aged group's effect was significantly lower than that of young adults (β = –0.378, t (45) = –3.15, p = 0.003) and older adults (β = –0.31, t (45) = –2.31, p = 0.02) (see Table 2). This pattern should also be interpreted cautiously because the age categories were not evenly represented across studies, particularly for middle-aged participants. As with upper-limb strength, the middle-aged subgroup was represented by only four modeled effects and 108 participants, limiting the precision of the subgroup estimate and the statistical power of the between-subgroup comparison. No other subgroup analyses by gender, intervention duration, weekly frequency, dosing strategy, daily dose, supplementation protocol, or training experience reached significance (all p > 0.05; see Table 2). Given the number of moderator analyses conducted across outcomes, this age-related finding should be regarded as exploratory and hypothesis-generating rather than confirmatory evidence of age-dependent responsiveness to creatine.
Table 2.
Results of moderator analyses on the effects of resistance training combined with creatine supplementation on lower‐limb muscle strength (no outliers).
| Moderator | K | Hedge's g | 95% CI | F | df | P | Level 2 variance (I2) | Level 3 variance (I2) |
|---|---|---|---|---|---|---|---|---|
| Mean age | 5 | 2 | 0.01 | <1% | 16.04% | |||
| Young adults (18–44 years) | 35 | 0.65 | [0.48, 0.82] | <0.001 | ||||
| Middle adults (45–59 years) | 4 | −0.1 | [−0.55, 0.35] | 0.65 | ||||
| Older adults (≥60 years) | 9 | 0.52 | [0.22, 0.82] | 0.001 | ||||
| Gender | 0.9 | 1 | 0.35 | <1% | 24.34% | |||
| Male | 28 | 0.52 | [0.32, 0.72] | <0.001 | ||||
| Female | 6 | 0.76 | [0.28, 1.25] | 0.003 | ||||
| Training status | 0 | 2 | 0.99 | <1% | 34.46%* | |||
| Athletes | 12 | 0.56 | [0.22, 0.89] | 0.002 | ||||
| Trained | 15 | 0.56 | [0.25, 0.87] | <0.001 | ||||
| Untrained | 21 | 0.54 | [0.28, 0.80] | <0.001 | ||||
| Intervention duration | 1.5 | 1 | 0.23 | <1% | 32.35%* | |||
| Long intervention (≥8 w) | 33 | 0.48 | [0.278, 0.68] | <0.001 | ||||
| Short intervention (<8 w) | 15 | 0.7 | [0.40,0.99] | <0.001 | ||||
| Training frequency | 0.9 | 1 | 0.36 | <1% | 30.7%* | |||
| High frequency (>3/pw) | 13 | 0.43 | [0.1, 0.74] | 0.01 | ||||
| Low frequency (≤3/pw) | 35 | 0.6 | [0.40, 0.79] | <0.001 | ||||
| Creatine loading | 0.8 | 2 | 0.47 | <1% | 31.01%* | |||
| Loading+maintenance | 23 | 0.67 | [0.40, 0.87] | <0.001 | ||||
| Maintenance | 21 | 0.44 | [0.20, 0.68] | <0.001 | ||||
| Loading only | 4 | 0.64 | [0.08, 1.26] | 0.03 | ||||
| Creatine intake forms | 0.1 | 1 | 0.74 | <1% | 32.45%* | |||
| Fixed dose (grams) | 24 | 0.58 | [0.34, 0.81] | <0.001 | ||||
| Weight-based dosing | 24 | 0.52 | [0.28, 0.76] | <0.001 | ||||
| Creatine dose | 0.4 | 1 | 0.52 | <1% | 34.4%* | |||
| High dose (>5 g/d) | 17 | 0.47 | [0.19, 0.75] | 0.001 | ||||
| Low dose (≤5 g/d) | 27 | 0.59 | [0.36, 0.82] | <0.001 |
Note: Hedges' g, the effect size indicator used in the pooled analysis; CI: confidence interval; K: represents the number of effect sizes. *p < 0.05; F: the F statistic from ANOVA, used to test for group differences. df: degrees of freedom for the F test. K should not be interpreted as the number of unique studies, because several studies contributed more than one dependent effect size to the three-level model; subgroup precision should therefore be interpreted together with participant totals reported in the text.
-
(5)
Publication bias analysis
The two‐level funnel plot showed slight asymmetry (see Figure 5), suggesting possible bias. Egger's test indicated significant bias (t (46) = 3.99, p < 0.001), and a multilevel Egger test still detected bias (t(46) = 2.52, p = 0.015). Trim‐and‐fill analysis estimated ~15 missing effect sizes on the left; after imputing these, the adjusted overall effect remained significant (g = 0.35, 95% CI [0.21, 0.50], p < 0.001), though it was lower than the unadjusted effect (g = 0.55, 95% CI [0.38, 0.71], p < 0.001). In summary, while publication bias may be present, its impact on conclusions appears limited.
Figure 5.

Funnel plot of the publication bias analysis for the effects of creatine combined with resistance training on lower‐limb muscle strength (no outliers).
3.3.3. Fat mass
-
(1)
Overall effect of RT + CR on fat mass
A total of 35 studies (41 effect sizes; N = 904) were included. The three‐level random‐effects model yielded a very small overall effect of resistance training plus creatine (RT+CR) on fat mass (g = −0.10, 95% CI [−0.23, 0.03]; 95% PI [−0.23, 0.03]; p = 0.14) (see Figure 3). The two‐level random‐effects model produced identical results (g = −0.10, 95% CI [−0.23, 0.03]; 95% PI [−0.23, 0.03]; p = 0.14). The heterogeneity test indicated no significant between‐effect variance (Q (40) = 7.30, p = 0.99). These findings indicate no statistically significant difference between RT+CR and RT alone for fat‐mass change.
-
(2)
Influence analysis
Influence analysis detected no individual study with a disproportionate impact on the pooled estimate, indicating that the overall effect size is robust (see Appendix 5, Figure 4).
-
(3)
Heterogeneity analysis
Main‐effect analysis showed no significant heterogeneity among effect sizes (Q (40) = 7.30, p = 0.99). In the total variance, sampling variance (Level 1) accounted for 100%, while within‐study (Level 2) and between‐study (Level 3) variances were both <1%. Likelihood‐ratio tests found neither Level 3 variance (χ2 < 0.01, p > 0.49) nor Level 2 variance (χ2 < 0.01, p > 0.49) to be significant, indicating no improvement of the three‐level model over two‐level models with either variance component constrained to zero (see Figure 3).
-
(4)
Moderator analyses
Subgroup analyses for all potential moderators (e.g. age, gender, intervention duration, training frequency, dosing strategy, daily dose, supplementation protocol, training experience) revealed no statistically significant differences (all p > 0.05; see Appendix 5, Table 1).
-
(5)
Publication bias analysis
Funnel‐plot inspection (two‐level model) showed a symmetric distribution (see Appendix 5, Figure 2), suggesting a low risk of publication bias. Egger's test found no significant bias (t (39) = 0.25, p = 0.80), and the multilevel Egger test likewise indicated no bias (t (39) = 0.28, p = 0.78). However, trim‐and‐fill analysis suggested approximately nine missing effect sizes on the left side of the funnel. After imputing these and recalculating, the adjusted overall effect became g = –0.16 (95% CI [–0.27, –0.04], p = 0.007), indicating a slightly larger and now statistically significant reduction in fat mass. Because trim‐and‐fill relies on symmetry assumptions and “filled” studies are hypothetical, this result should be interpreted with caution. Overall, multiple bias assessments support the robustness of these findings.
3.3.4. Lean mass
-
(1)
Overall effect of RT+CR on lean mass
A total of 44 studies (54 effect sizes; N = 1175) were included. The three-level random-effects model yielded a moderate overall effect of resistance training plus creatine (RT+CR) on lean mass (g = 0.30, 95% CI [0.18, 0.41]; 95% PI [0.18, 0.41]; p < 0.001) (see Figure 3). The two-level random-effects model produced the same results (g = 0.30, 95% CI [0.18, 0.41]; 95% PI [0.18, 0.41]; p < 0.001). The heterogeneity test indicated no significant between‐effect variance (Q (53) = 14.04, p = 0.99). These findings show that RT+CR significantly outperforms RT alone in increasing lean mass.
-
(2)
Influence analysis
Influence analysis detected no individual study with a disproportionate impact on the pooled estimate, indicating that the overall effect size estimate is robust (see Appendix 6, Figure 4).
-
(3)
Heterogeneity analysis
Main‐effect analysis showed no significant heterogeneity among effect sizes (Q (53) = 14.04, p = 0.99). Sampling variance (Level 1) accounted for 100% of total variance, while within‐study (Level 2) and between‐study (Level 3) variances were both <1%. Likelihood‐ratio tests found neither Level 3 variance (χ2 < 0.01, p > 0.49) nor Level 2 variance (χ2 < 0.01, p > 0.49) to be significant, indicating no improvement of the three-level model over two-level models with either component constrained to zero (see Figure 3).
-
(4)
Moderator analyses
None of the between-subgroup differences for the tested moderators reached statistical significance (all p > 0.05; see Appendix 6, Table 1). Although several within-subgroup estimates reached statistical significance, these findings were interpreted descriptively because the corresponding between-subgroup tests did not support differential effects across moderator categories. Given the number of moderator analyses conducted across outcomes, these subgroup-specific patterns should be regarded as exploratory rather than evidence of effect modification.
-
(5)
Publication bias analysis
Funnel plot inspection revealed slight asymmetry (see Appendix 6, Figure 2), suggesting possible bias. Egger's test found no significant bias (t (52) = 0.93, p = 0.36), and the multilevel Egger test likewise showed no bias (t (52) = 0.58, p = 0.56). Trim-and-fill analysis suggested six missing effect sizes on the right; after imputation, the adjusted overall effect remained significant (g = 0.33, 95% CI [0.23, 0.44], p < 0.001) and was slightly higher than the unadjusted estimate (g = 0.30, 95% CI [0.18, 0.41], p < 0.001). Thus, although a publication-bias risk cannot be ruled out, its impact on the results appears limited.
3.3.5. Body‐fat percentage
-
(1)
Overall effect of RT+CR on body‐fat percentage
A total of 41 studies (48 effect sizes; N = 1067) were included. The three‐level random‐effects model yielded a small but statistically significant overall effect of resistance training plus creatine (RT+CR) on body‐fat percentage (g = –0.14, 95% CI [–0.26, –0.02]; 95% PI [–0.26, –0.02]; p = 0.02) (see Figure 3). The two‐level random‐effects model produced identical results (g = –0.14, 95% CI [–0.26, –0.02]; 95% PI [–0.26, –0.02]; p = 0.02). The heterogeneity test indicated no significant between‐effect variance (Q(47) = 26.68, p = 0.99). These findings demonstrate that RT+CR significantly outperforms RT alone in reducing body‐fat percentage.
-
(2)
Influence analysis
Influence analysis detected no individual study with a disproportionate impact on the pooled estimate, indicating the overall effect size is robust (see Appendix 7, Figure 4).
-
(3)
Heterogeneity analysis
Main‐effect analysis showed no significant heterogeneity among effect sizes (Q (47) = 26.68, p = 0.99). Sampling variance (Level 1) accounted for 100% of total variance, while within‐study (Level 2) and between‐study (Level 3) variances were both <1%. Likelihood‐ratio tests found neither Level 3 variance (χ2 < 0.01, p > 0.49) nor Level 2 variance (χ2 < 0.01, p > 0.49) to be significant, indicating no improvement of the three‐level model over two‐level models with either component constrained to zero (see Figure 3).
-
(4)
Moderator analyses
None of the between-subgroup differences for the tested moderators reached statistical significance (all p > 0.05; see Appendix 7, Table 1). Although several within-subgroup estimates reached statistical significance, these findings were interpreted descriptively because the corresponding between-subgroup tests did not support differential effects across moderator categories. Given the number of moderator analyses conducted across outcomes, these subgroup-specific patterns should be regarded as exploratory rather than evidence of effect modification.
-
(5)
Publication bias analysis
The two‐level funnel plot was symmetric (see Appendix 7, Figure 2), suggesting a low risk of publication bias. Egger's test found no significant bias (t (46) = –1.28, p = 0.21), and the multilevel Egger test likewise showed no bias (t (46) = –0.84, p = 0.40). Trim‐and‐fill analysis indicated no missing effect sizes on either side of the distribution. Taken together, these assessments suggest that publication bias is unlikely to have materially influenced the results.
3.3.6. Body weight
-
(1)
Overall effect of RT+CR on body weight
A total of 50 studies (55 effect sizes; N = 1180) were included. The three-level random-effects model yielded a very small overall effect of resistance training plus creatine (RT+CR) on body weight (g = 0.12, 95% CI [0.06, 0.23]; 95% PI [0.06, 0.23]; p = 0.04) (see Figure 3). The two-level random-effects model produced the same result (g = 0.12, 95% CI [0.06, 0.23]; 95% PI [0.06, 0.23]; p = 0.04). The heterogeneity test indicated no significant between-effect variance (Q (54) = 23.16, p = 0.99), suggesting little detectable heterogeneity in this very small body-weight effect.
-
(2)
Influence analysis
An influence analysis identified one outlier (Schweizer, 2021; g = –1.24; see Appendix 8, Figure 4). Excluding this effect size and re-running the meta-analysis produced a still-significant overall effect (three-level: g = 0.14, 95% CI [0.03, 0.26]; 95% PI [0.03, 0.26]; p = 0.01; two-level: g = 0.14, 95% CI [0.03, 0.26]; 95% PI [0.03, 0.26]; p = 0.01). To avoid undue influence, all subsequent analyses omitted this outlier.
-
(3)
Heterogeneity analysis
After excluding the outlier, heterogeneity among effect sizes was not significant (Q (53) = 12.37, p = 0.99). Sampling variance (Level 1) accounted for 100% of total variance, while within-study (Level 2) and between-study (Level 3) variances were both <1%. Likelihood-ratio tests found neither Level 3 variance (χ2 < 0.01, p > 0.49) nor Level 2 variance (χ2 < 0.01, p > 0.49) to be significant, indicating no improvement of the three-level model over two-level models with either component constrained to zero (see Figure 3).
-
(4)
Moderator analyses
None of the between-subgroup differences for the tested moderators reached statistical significance (all p > 0.05; see Appendix 8, Table 1). Although several within-subgroup estimates reached statistical significance, these findings were interpreted descriptively because the corresponding between-subgroup tests did not support differential effects across moderator categories. Given the number of moderator analyses conducted across outcomes, these subgroup-specific patterns should be regarded as exploratory rather than evidence of effect modification.
-
(5)
Publication bias analysis
The two-level funnel plot was symmetric (see Appendix 8, Figure 2), and neither Egger's regression test nor the multilevel Egger test detected significant bias (all p > 0.05). Trim-and-fill analysis indicated no missing effect sizes on either side of the distribution. Together, these assessments suggest that publication bias is unlikely to have materially influenced the body weight findings.
3.4. Meta-regression analysis
To examine potential linear relationships between age or creatine dose and intervention effect sizes, we conducted exploratory meta-regression analyses. Age was inversely associated with the effect size for upper-limb strength (β = –0.008, p = 0.03, R2 = 35.49%) and lower-limb strength (β = –0.0079, p = 0.03, R2 = 51.7%), suggesting larger strength effects in studies with younger participants. These findings should be interpreted cautiously because the analyses were conducted at the study level, multiple moderator tests were performed, and the age distribution across included studies was uneven (see Figure 6). Because these analyses were based on aggregate study-level characteristics rather than individual participant data, they cannot determine whether age modifies creatine responsiveness at the individual level and may be affected by ecological bias. Age was not significantly associated with any body composition outcome, and creatine dose was not significantly associated with either strength or body composition outcomes.
Figure 6.

Meta‐regression analysis of the effects of age and creatine dosage on effect size. Note: the shaded area around the black dashed line represents the 95% confidence interval (CI), and the shaded area around the light blue dashed line represents the 95% prediction interval (PI).
3.5. Risk of bias
Detailed risk-of-bias assessments for each study are presented in Appendix 2, Table 1. Of the 63 included trials, 41 were judged to have low risk of bias, 16 moderate risk, and 6 high risk.
3.6. GRADE assessment
Appendix 9, Table 1 summarizes the GRADE evaluations. Across all examined outcomes and domains, the overall certainty of evidence ranged from low to moderate.
4. Discussion
4.1. Muscle strength
The present meta-analysis demonstrates that resistance training combined with creatine supplementation (RT+CR) produces significantly greater gains in muscle strength than resistance training alone. Both upper- and lower-body strength showed moderate improvements (Hedges' g ≈ 0.5–0.6) favoring RT+CR over placebo [32,33]. These findings reinforce a large body of evidence that creatine is one of the most effective ergogenic aids for enhancing high-intensity exercise performance and training adaptations [34]. The mechanism underlying these strength gains is well-founded: creatine loading elevates intramuscular phosphocreatine stores, which allows for improved adenosine triphosphate (ATP) resynthesis during short, intense exercise. This translates to the ability to perform extra repetitions or lift heavier loads in training, ultimately leading to superior neuromuscular adaptations [35,36]. Creatine may also facilitate faster between-set recovery and increase training volume, compounding its impact on strength development. Additionally, there is evidence that creatine's cell-swelling effect and upregulation of anabolic signaling (e.g. elevated IGF-1 expression and satellite cell activity) contribute to greater gains in muscle fiber size and contractile force output over time [10,37]. Collectively, these mechanisms may help explain the greater strength gains observed with RT+CR compared with resistance training alone.
Despite the overall positive effect, our results indicate some heterogeneity and potential context-specific differences. In the exploratory moderator analyses, age was the only participant-level variable showing a statistically significant between-subgroup difference for both upper- and lower-body strength. Subgroup-specific estimates favored RT+CR in young and older adults, whereas the estimates in middle-aged adults were smaller and did not reach statistical significance. This apparent age-related pattern should not be interpreted as confirmatory evidence of age-dependent responsiveness to creatine. Multiple moderators were examined across several outcomes, no formal adjustment for multiplicity was applied, and the age subgroups were unevenly represented, particularly the middle-aged category. Therefore, this finding is best viewed as hypothesis-generating and should be tested in future trials or individual participant data meta-analyses. Importantly, the meta-regression analyses were conducted using aggregate study-level mean ages, not participant-level data; therefore, the observed age association may reflect between-study differences in study design, training programs, outcome measurement, baseline fitness, or nutrition rather than a true individual-level age effect. Another consideration is that this apparent age-related pattern may partly reflect differences in how strength was assessed across studies, as most trials reported absolute rather than relative strength outcomes. Because relative strength measures were not reported consistently enough for separate quantitative analysis, it remains unclear whether the observed age-related differences would persist after normalization to body mass or lean mass. The limited number of studies and participants in the mid-life category may also have reduced the precision of this comparison. This interpretation should also be considered alongside evidence summarized by Chilibeck et al. [12], suggesting that age-related differences in intramuscular phosphocreatine/creatine stores may be muscle-specific rather than uniform across all muscle groups. In their discussion, older adults did not consistently show lower phosphocreatine stores overall, but lower levels were observed in the vastus lateralis when studies of other lower-leg muscles were excluded, which may partly help explain why some older adults still respond favorably to creatine supplementation during resistance training.
For other moderators, the subgroup analyses did not provide clear evidence that the strength response to RT+CR differed by sex, training frequency, intervention duration, or creatine dosing strategy. Although the male subgroup estimate reached statistical significance whereas the female subgroup estimate did not, the between-subgroup comparison by sex was not significant. This pattern should therefore be interpreted descriptively rather than as evidence of a sex-specific effect. The direction of effect in women still favored creatine, and prior research confirms females can benefit similarly from creatine given adequate training stimulus [38,39]. Thus, the lack of significance in females here should be interpreted cautiously, not only because fewer studies included women, but also because the available literature predominantly reported absolute rather than relative strength outcomes. Because relative strength measures were not reported consistently enough for separate quantitative synthesis, it remains unclear whether these apparent differences between males and females would persist after normalization to body mass or lean mass.
Another point of consideration is the variability in creatine supplementation protocols. In our analysis, whether creatine was administered in a fixed dose (e.g. 5 g/day) or relative to body weight, and whether a loading phase was used, did not significantly alter strength outcomes. This aligns with recent evidence that common dosing strategies (e.g. ~0.1 g/kg/day or a 5–7 g daily dose after loading) all achieve muscle creatine saturation and yield comparable training benefits [40,41]. We did observe that studies employing a maintenance phase (with or without an initial loading phase) tended to report significant strength improvements, whereas those with only a short loading period did not show a clear benefit. It stands to reason that sustained creatine intake throughout the training program is necessary to maintain elevated intramuscular creatine levels and realize strength gains. From a practical standpoint, athletes and practitioners can be confident that even lower daily doses (3–5 g/day after saturation) are sufficient to elicit strength gains, which may simplify supplementation protocols and reduce cost.
The funnel plot and Egger's regression analysis suggested a risk of publication bias in the strength outcomes (especially for upper body strength), as smaller studies with null or negative results may be underrepresented. After statistically “filling” missing studies, the corrected mean effect size for upper body strength (approximately g = 0.36) was lower than the observed value (g = 0.54) but remained significant. This implies that the true magnitude of creatine's strength benefit might be modestly smaller than our raw estimate, yet still practically meaningful. In any case, the strength improvements appear genuine and not solely an artifact of biased reporting.
Overall, our findings concur with the consensus that creatine monohydrate is the single most effective nutritional supplement for increasing muscle strength and high-intensity exercise capacity during resistance training [12,42]. Athletes involved in strength/power sports, as well as recreational weightlifters seeking greater gains, can expect a moderate enhancement in strength development from adding creatine to their regimen.
4.2. Body composition
In addition to strength gains, RT+CR had significant effects on body composition—primarily by increasing lean mass—without inducing any substantial loss of fat mass. Our results showed a significant association between creatine supplementation and greater gains in lean mass when combined with resistance training. The overall effect size for lean mass was small-to-moderate (g ≈ 0.30) but highly significant (p < 0.001), indicating that individuals who supplemented with creatine achieved greater increases in muscle size/lean tissue than those who performed resistance training alone, as evidenced by recent meta-analyses [43,44]. For example, a 2024 systematic review in young adults found that creatine users gained about 1.14 kg more lean mass than placebo across training programs [43]. Likewise, a comprehensive dose–response meta-analysis reported that creatine (at ~5–7 g/day or 0.3 g/kg) leads to ~0.8–1.0 kg additional fat-free mass increase relative to controls [13]. These findings are consistent with the present results and support the view that creatine can augment lean mass gains during resistance training.
Mechanistically, the lean mass gains with creatine are likely supported by multiple, complementary factors. As discussed, creatine allows for more intense training, which stimulates muscle hypertrophy to a greater extent [32]. Beyond training performance, creatine has osmotic properties—it drives water into muscle cells—which can acutely increase muscle volume. This cell volumization is not merely “water weight;” cellular swelling is hypothesized to act as an anabolic signal, reducing proteolysis and increasing protein synthesis within muscle fibers [44]. Over weeks of training, creatine users often exhibit greater muscle fiber cross-sectional area and contractile protein content than non-users, as shown in muscle biopsies from controlled trials [37]. Creatine may also upregulate intramuscular growth factors; For instance, elevated expression of myogenic regulatory factors (like MRF4 and MyoD) and IGF-1 have been observed with creatine supplementation, potentially enhancing muscle fiber differentiation and growth [44]. It is important to acknowledge that a portion of the initial weight gain from creatine (commonly 0.5–1.5 kg in the first week) is due to increased water retention inside muscles. However, intracellular and extracellular water outcomes were not reported consistently enough across the included studies to permit separate quantitative synthesis. Future studies should report these variables more systematically to clarify the extent to which creatine-related increases in body mass and lean mass reflect intracellular water retention versus longer-term tissue accretion. Importantly, this should not be viewed negatively—the additional water is intracellular (not bloating or extracellular edema) [45] and may accompany the muscle hypertrophy process. Indeed, the meta-analytic evidence suggests that the extra weight gained with creatine is largely reflected in lean mass (inclusive of intracellular water and new muscle tissue), with no significant gain in fat mass [13,43]. Our analysis found that total body mass increased slightly more in the creatine groups than in controls (g = 0.12, p = 0.01). Notably, this weight gain was accompanied by a non-significant change in fat mass (Hedges' g = –0.10, 95% CI –0.23 to 0.03, p = 0.14), indicating that creatine did not significantly reduce or increase fat weight. In other words, participants gained extra weight from creatine, but it was more consistent with increases in lean mass than with fat accumulation. Our results and prior studies therefore do not support the concern that creatine might cause fat gain—if anything, fat mass tended to be slightly lower in the creatine group, though not to a statistically significant degree.
Although creatine did not significantly reduce absolute fat mass in this meta-analysis, it did lead to a small but significant reduction in body-fat percentage. We observed a collective decrease in body-fat percentage (g = −0.14) favoring the creatine condition (p = 0.02). This seemingly paradoxical result—no fat loss, yet lower body-fat percentage—is explained by the increase in the lean mass denominator. Essentially, as creatine users gained more lean weight while fat mass remained unchanged, their proportion of body fat declined slightly relative to the placebo group. This phenomenon has been documented in other reviews: for instance, a recent meta-analysis of younger adults (<50 y) found that creatine combined with resistance training produced a very small but significant drop in body-fat percentage (~1.2% on average) with no corresponding loss of fat mass (–0.18 kg, p = 0.76) [46]. Similarly, among older adults, creatine has been shown to reduce body-fat percentage by ~0.5% versus controls despite no difference in fat weight [47]. Taken together, the evidence suggests that creatine's primary body-composition effect is to increase lean mass rather than decrease fat. Any improvements in body fat percentage are largely a result of diluting the fat mass with additional muscle. This nuance is important for interpreting changes on the scale: a small weight increase on creatine should not be mistaken as fat gain—rather, in the context of resistance training, it may reflect favorable changes in fat-free mass. From a health and athletic perspective, a lower body fat percentage at equal or higher body weight can be beneficial, as it reflects a greater muscle-to-fat ratio. For athletes in weight-class sports, however, the slight weight gain might need consideration; strategies like timing the supplementation or focusing on muscular benefits during off-season could be used to manage weigh-ins.
The moderator analyses did not identify statistically significant between-subgroup differences for body composition outcomes. Although some within-subgroup estimates reached statistical significance, these patterns should be interpreted descriptively because the corresponding between-subgroup tests did not support differential effects across moderator categories. Given the number of moderator analyses conducted across outcomes, these subgroup-specific findings should be regarded as exploratory rather than evidence of effect modification. The available evidence therefore supports a primary body-composition effect of RT+CR on lean mass, but it does not establish that this effect differs reliably by sex, age, training status, intervention duration, training frequency, or supplementation protocol. Indeed, controlled trials in females have reported lean mass increases from creatine similar to those in males [48]. However, the smaller number of studies including women limits the precision of sex-specific estimates. Regarding dosing regimens, as noted earlier, no clear advantage was seen for higher daily doses or for utilizing a loading phase in terms of final body composition. This comports with recent dose–response analyses suggesting that even low-dose regimens (e.g. 3–5 g/day without loading) can gradually saturate muscle creatine stores and induce comparable hypertrophic effects over the long term [13].
In scenarios where body recomposition is the goal (simultaneously building muscle and losing fat), creatine can be a useful tool to maximize muscle preservation/growth, while standard nutritional approaches (e.g. caloric deficit, high protein intake) must drive the fat loss component. Preliminary evidence suggests creatine might help attenuate muscle catabolism during calorie-restricted weight loss, thereby improving the quality of weight loss (more fat relative to muscle lost), although more research is needed in this area. From a health standpoint, the combination of resistance training and creatine yields improvements in body composition characterized by greater lean mass and stable fat mass—changes that are associated with better metabolic health, strength, and functional capacity in various populations [43]. For older adults, in particular, the ability of creatine to enhance muscle mass without affecting fat may contribute to combating sarcopenia while not exacerbating adiposity [49]. In young athletes, creatine can increase muscle size and potentially improve power-to-weight ratio (if fat remains the same or slightly lower proportionally). In sum, creatine's impact on body composition should be viewed primarily as a lean mass booster rather than a fat reducer, yielding a slightly leaner physique mainly through muscular gains. Notably, the vast majority of included studies used creatine monohydrate; therefore, the present findings primarily reflect the effects of this form of creatine, and there was insufficient variability to formally assess creatine type as a moderator.
4.3. Practical applications
In practical application, resistance training combined with creatine supplementation can be recommended as an effective strategy to enhance muscle strength and lean body mass. Athletes and recreational weight trainers aiming to maximize strength gains or muscle hypertrophy over a training cycle are likely to benefit from creatine. A typical dosing regimen (e.g. ~5–7 g of creatine monohydrate per day, or ~0.1 g/kg body mass) is sufficient to produce ergogenic effects. However, practical recommendations should be applied in conjunction with an appropriately structured resistance-training program and adequate dietary support, because protein intake, total energy intake, and concurrent training exposures were not uniformly controlled across the included trials.
4.4. Limitations and future directions
While this meta-analysis provides evidence of creatine's benefits, certain limitations should be acknowledged when interpreting the findings. One limitation is the moderate sample size and scope of included studies. We synthesized results from 63 trials, many with relatively small sample sizes, which leaves uncertainty about subgroup effects such as sex differences, age-related patterns, or optimal dosage. The majority of participants in the analyzed studies were young to middle-aged men, and fewer data are available for women. This limits the generalizability of conclusions to female trainees, who may respond differently; indeed, our subgroup analysis hinted at smaller, non-significant strength gains in women taking creatine. Future research should therefore specifically target female populations to confirm efficacy and investigate whether hormonal factors or lower baseline creatine stores play a role in the attenuated response. Similarly, more research is needed in older populations. Although numerous studies in older adults indicate benefits of creatine on muscle aging, differences in recovery capacity and nutrition in the elderly merit further investigation (e.g. whether creatine needs to be combined with higher protein intake or longer training durations for maximal benefit in sarcopenic individuals). A second important limitation is the lack of uniform reporting and control of nutrition across the primary studies. Habitual protein intake, total caloric intake, macronutrient distribution, hydration status, supplement timing, and overall energy balance were either unreported or reported inconsistently in many trials. These factors could plausibly influence changes in lean mass, body mass, and strength, and their incomplete reporting limited our ability to code or analyze them as moderators. Therefore, the present findings should be interpreted as the effects of creatine added to resistance training under the nutritional conditions used in the included trials, rather than as the isolated effect of creatine under standardized dietary intake. A third limitation concerns training heterogeneity. Although all included studies involved resistance training, protocols varied in exercise selection, intensity, volume, progression, supervision, and weekly frequency. In addition, some participants may have engaged in sport training or concurrent aerobic exercise outside the experimental resistance-training program, but these exposures were not consistently quantified. Such variation may have contributed to between-study differences in adaptation and may partly obscure whether creatine effects differ across specific training contexts. Future trials and meta-analyses would benefit from focusing on more homogeneous combinations of nutrition and training, such as creatine combined with standardized protein intake, creatine during energy deficit or surplus, creatine with resistance training alone versus concurrent training, or creatine effects in specific sport populations. Finally, the moderator and meta-regression findings should be interpreted cautiously because they were based on aggregate study-level data. Aggregate-level meta-regression is vulnerable to ecological bias and residual confounding; therefore, it cannot establish how individual-level variables such as age, sex, baseline fitness, dietary intake, or training volume modify a participant's response to creatine. Individual participant data meta-analyses or well-powered randomized trials with standardized nutrition and detailed training monitoring are needed to test these questions more directly.
5. Conclusion
This meta-analysis indicates that resistance training combined with creatine supplementation produces greater improvements in muscle strength and lean body mass than resistance training alone, without appreciably affecting fat mass. The modest increase in total body weight was likely driven primarily by gains in lean body mass, resulting in a slight reduction in body fat percentage. Among participant characteristics, age appeared to be a potential moderator, with larger benefits observed in younger individuals; however, this finding should be interpreted cautiously given the uneven representation of age subgroups across studies. Intervention-protocol variables did not significantly influence the observed effects. Because nutritional intake and non-resistance-training exposures were inconsistently reported across trials, future research should use more standardized dietary control and training monitoring to clarify the conditions under which creatine provides the greatest benefit.
Supplementary Material
Appendix.docx
Acknowledgements
We would like to thank all experts who provided guidance and advice during the preparation of this manuscript, and we express our sincere respect to all authors whose work was cited in this article.
Funding Statement
This work was supported by the Science and Technology Strengthening Project of Capital University of Physical Education and Sports (Grant No. 155225002/011).
Disclosure statement
No potential conflict of interest was reported by the author(s).
Data availability statement
All data and material reported in this review and meta-analysis were from peer-reviewed publications. The datasets supporting the conclusions of this article are included within the article and its additional files.
Ethics approval and consent to participate
Not applicable.
Consent for publication
Not applicable.
Clinical trial number
Not applicable.
Supplementary material
Supplemental data for this article can be accessed at https://doi.org/10.1080/15502783.2026.2718316.
Author contributions
Hai Wang: data curation, formal analysis, methodology, supervision, writing–original draft, writing–review and editing. Hao Wang: data curation, formal analysis, writing–review and editing. Xianyang Xin: data curation, formal analysis, methodology, writing–review and editing. Zheng Yi: formal analysis, supervision, validation, writing–review and editing.
References
- [1]. Volek JS, & D. Performance and muscle fiber adaptations to creatine supplementation and heavy resistance training. 1999. [DOI] [PubMed]
- [2]. Rawson PES, Persky PAM. Mechanisms of muscular adaptations to creatine supplementation. International Sportmed Journal. 2007;8(2):43–53. [Google Scholar]
- [3]. Greenhaff PL. The nutritional biochemistry of creatine. J Nutr Biochem. 1997;8(11):610–618. doi: 10.1016/S0955-2863(97)00116-2 [DOI] [Google Scholar]
- [4]. Kaviani M, Abassi A, Chilibeck PD. Creatine monohydrate supplementation during eight weeks of progressive resistance training increases strength in as little as two weeks without reducing markers of muscle damage. J Sports Med Phys Fitness. 2019;59(4):608–612. doi: 10.23736/S0022-4707.18.08406-2 [DOI] [PubMed] [Google Scholar]
- [5]. Casey A, Greenhaff PL. Does dietary creatine supplementation play a role in skeletal muscle metabolism and performance? Am J Clin Nutr. 2000;72(2):607S–617S. doi: 10.1093/ajcn/72.2.607S [DOI] [PubMed] [Google Scholar]
- [6]. Bonilla DA, Moreno Y. Molecular and metabolic insights of creatine supplementation on resistance training. Revista Colombiana De Química. 2016;44(1):11–18. doi: 10.15446/rev.colomb.quim.v44n1.53978 [DOI] [Google Scholar]
- [7]. Farshidfar F, Pinder MA, & Myrie SB, et al. Creatine supplementation and skeletal muscle metabolism for building muscle mass- review of the potential mechanisms of action. 2017. https://www.eurekaselect.com/article/83899 [DOI] [PubMed]
- [8]. Kreider RB, Kalman DS, Antonio J, et al. International Society of Sports Nutrition position stand: safety and efficacy of creatine supplementation in exercise, sport, and Medicine. J Int Soc Sports Nutr. 2017;14(1):18. doi: 10.1186/s12970-017-0173-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- [9]. Mirdha P, Nalgirkar V, Patil A, et al. Efficacy and safety of creatine supplementation on strength and muscle mass in resistance trained individuals: a prospective study. Int J Curr Res Rev. 2021;13(12):150–154. doi: 10.31782/IJCRR.2021.131230 [DOI] [Google Scholar]
- [10]. Cribb PJ, Williams AD, Hayes A. A creatine-protein-carbohydrate supplement enhances responses to resistance training. Med Sci Sports Exercise. 2007;39(11):1960–1968. doi: 10.1249/mss.0b013e31814fb52a [DOI] [PubMed] [Google Scholar]
- [11]. Candow DG, Vogt E, Johannsmeyer S, et al. Strategic creatine supplementation and resistance training in healthy older adults. Appl Physiol Nutr Metab. 2015;40(7):689–694. doi: 10.1139/apnm-2014-0498 [DOI] [PubMed] [Google Scholar]
- [12]. Chilibeck P, Kaviani M, Candow D, et al. Effect of creatine supplementation during resistance training on lean tissue mass and muscular strength in older adults: a meta-analysis. Open Access J Sports Med. 2017;8:213–226. doi: 10.2147/OAJSM.S123529 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [13]. Pashayee-Khamene F, Heidari Z, Asbaghi O, et al. Creatine supplementation protocols with or without training interventions on body composition: a GRADE-assessed systematic review and dose-response meta-analysis. J Int Soc Sports Nutr. 2024;21(1):2380058. doi: 10.1080/15502783.2024.2380058 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [14]. Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021;n71. doi: 10.1136/bmj.n71 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [15]. Higgins JPT, Altman DG, Gotzsche PC, et al. Cochrane Bias Methods Group; Cochrane Statistical Methods Group . The Cochrane collaboration's tool for assessing risk of bias in randomised trials. BMJ. 2011;343(oct18 2):d5928–d5928. doi: 10.1136/bmj.d5928 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [16]. Schünemann HJ, Higgins JP, Vist GE, et al. Completing ‘summary of findings’ tables and grading the certainty of the evidence. In: Higgins JPT, Thomas J, Chandler J, Cumpston M, Li T, Page MJ, Welch VA, editors. Cochrane Handbook for Systematic Reviews of Interventions. Wiley; 2019. p. 375–402. doi: 10.1002/9781119536604.ch14 [DOI] [Google Scholar]
- [17]. Harrer M, Cuijpers P, Furukawa T, et al. Doing meta-analysis with R: a hands-on guide. Chapman and Hall/CRC; 2021. doi: 10.1201/9781003107347 [DOI] [Google Scholar]
- [18]. Assink M, Wibbelink CJM, Assink M. Fitting three-level meta-analytic models in R: a step-by-step tutorial. The Quantitative Methods for Psychology. 2016;12(3):154–174. doi: 10.20982/tqmp.12.3.p154 [DOI] [Google Scholar]
- [19]. Cheung MW-L. Modeling dependent effect sizes with three-level meta-analyses: a structural equation modeling approach. Psychol Methods. 2014;19(2):211–229. doi: 10.1037/a0032968 [DOI] [PubMed] [Google Scholar]
- [20]. Cheung MW-L. A guide to conducting a meta-analysis with non-independent effect sizes. Neuropsychol Rev. 2019;29(4):387–396. doi: 10.1007/s11065-019-09415-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [21]. Rodgers MA, & Pustejovsky JE. Evaluating meta-analytic methods to detect selective reporting in the presence of dependent effect sizes. 2021. [DOI] [PubMed]
- [22]. Schindler J, Richter T. Text generation benefits learning: a meta-analytic review. Educational Psychology Review. 2023;35(2):44. doi: 10.1007/s10648-023-09758-w [DOI] [Google Scholar]
- [23]. Borenstein M, Hedges LV, Higgins JPT, et al. A basic introduction to fixed-effect and random-effects models for meta-analysis. Research Synthesis Methods. 2010;1(2):97–111. doi: 10.1002/jrsm.12 [DOI] [PubMed] [Google Scholar]
- [24]. Aksayli ND, Sala G, Gobet F. The cognitive and academic benefits of Cogmed: a meta-analysis. Educational Research Review. 2019;27:229–243. doi: 10.1016/j.edurev.2019.04.003 [DOI] [Google Scholar]
- [25]. Viechtbauer W. Conducting meta-analyses in R with the metafor package. J Stat Softw. 2010;36(3):1–48. doi: 10.18637/jss.v036.i03 [DOI] [Google Scholar]
- [26]. Hunter JE, Schmidt FL. Methods of meta-analysis: Correcting error and bias in research findings. SAGE; 2004.. [Google Scholar]
- [27]. Cohen J. Statistical power analysis for the behavioral sciences (2). Routledge; 2013. doi: 10.4324/9780203771587 [DOI] [Google Scholar]
- [28]. Cochran WG. The combination of estimates from different experiments. Biometrics. 1954;10(1):101. doi: 10.2307/3001666 [DOI] [Google Scholar]
- [29]. Egger M, Smith GD, Schneider M, et al. Bias in meta-analysis detected by a simple, graphical test. BMJ. 1997;315(7109):629–634. doi: 10.1136/bmj.315.7109.629 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [30]. Duval S, Tweedie R. A nonparametric “trim and fill” method of accounting for publication bias in meta-analysis. Journal of the American Statistical Association. (world). 2000;95(449):89–98. doi: 10.1080/01621459.2000.10473905 [DOI] [Google Scholar]
- [31]. Viechtbauer W, Cheung MW-L. Outlier and influence diagnostics for meta-analysis. Research Synthesis Methods. 2010;1(2):112–125. doi: 10.1002/jrsm.11 [DOI] [PubMed] [Google Scholar]
- [32]. Wang Z, Qiu B, Li R, et al. Effects of creatine supplementation and resistance training on muscle strength gains in adults <50 years of age: a systematic review and meta-analysis. Nutrients. 2024;16(21):3665. doi: 10.3390/nu16213665 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [33]. Nunes JP, Ribeiro AS, Schoenfeld BJ, et al. Creatine supplementation elicits greater muscle hypertrophy in upper than lower limbs and trunk in resistance-trained men. Nutr Health. 2017;23(4):223–229. doi: 10.1177/0260106017737013 [DOI] [PubMed] [Google Scholar]
- [34]. Cooper R, Naclerio F, Allgrove J, et al. Creatine supplementation with specific view to exercise/sports performance: an update. J Int Soc Sports Nutr. 2012;9(1):33. doi: 10.1186/1550-2783-9-33 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [35]. Chrusch MJ, Chilibeck PD, Chad KE, et al. Creatine supplementation combined with resistance training in older men. Med Sci Sports Exercise. 2001;33(12):2111–2117. doi: 10.1097/00005768-200112000-00021 [DOI] [PubMed] [Google Scholar]
- [36]. Moon A, Heywood L, Rutherford SM, et al. Creatine supplementation in the elderly: is resistance training really needed? Journal of Nutrition and Health Sciences. 2015;2(2):1–9. doi: 10.15744/2393-9060.2.203 [DOI] [Google Scholar]
- [37]. Swinton P, Forbes S, Aragon AA, et al. The effects of creatine supplementation combined with resistance training on regional measures of muscle hypertrophy: a systematic review with meta-analysis. Nutrients. 2023;15:2116. doi: 10.3390/nu15092116 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [38]. Santana CC, de A, Santos EEP, et al. Efficacy of creatine supplementation combined with resistance training on muscle strength and muscle mass in older females: a systematic review and meta-analysis. Nutrients. 2021;13:3757. doi: 10.3390/nu13113757 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [39]. Syrotuik D, Ferguson TB. Effects of creatine monohydrate supplementation on body composition and strength indices in experienced resistance trained women. Journal of Strength & Conditioning Research. 2006;20:939. doi: 10.1519/R-18485.1 [DOI] [PubMed] [Google Scholar]
- [40]. Chilibeck P, Zello G, Farthing J, et al. Comparison of creatine supplementation before versus after supervised resistance training in healthy older adults. Research in Sports Medicine. 2014;22:61–74. doi: 10.1080/15438627.2013.852088 [DOI] [PubMed] [Google Scholar]
- [41]. Romance R, Benítez-Porres J, Kreider R, et al. Creatine enhances the effects of cluster-set resistance training on lower-limb body composition and strength in resistance-trained men: a pilot study. Nutrients. 2021;13:2303. doi: 10.3390/nu13072303 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [42]. Forbes S, Candow D, Chilibeck P. Creatine supplementation and aging musculoskeletal health. Endocrine. 2014;45:354–361. doi: 10.1007/s12020-013-0070-4 [DOI] [PubMed] [Google Scholar]
- [43]. Pandit A, Desai I, Kaakoush N, et al. The effect of creatine supplementation on resistance training–based changes to body composition: a systematic review and meta-analysis. Journal of Strength & Conditioning Research. 2024;38:1813–1821. doi: 10.1519/JSC.0000000000004862 [DOI] [PubMed] [Google Scholar]
- [44]. Arazi H, Eghbali E, Suzuki K. Supplementing with which form of creatine (hydrochloride or monohydrate) alongside resistance training can have more impacts on anabolic/catabolic hormones, strength and body composition? Physiol Res. 2024;73(5):739–753. doi: 10.33549/physiolres.935323 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [45]. Candow DG, Forbes SC, Roberts MD, et al. Creatine o'clock: does timing of ingestion really influence muscle mass and performance? Frontiers in Sports and Active Living. 2022;4:893714. doi: 10.3389/fspor.2022.893714 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [46]. Candow DG, Prokopidis K, Forbes SC, et al. Resistance exercise and creatine supplementation on fat mass in adults < 50 years of age: a systematic review and meta-analysis. Nutrients. 2023;15(20):4343. doi: 10.3390/nu15204343 [DOI] [PMC free article] [PubMed] [Google Scholar]
- [47]. Candow D, Forbes S. Creatine and strength training in older adults: an update. Translational Exercise Biomedicine. 2024;1:212–222. doi: 10.1515/teb-2024-0019 [DOI] [Google Scholar]
- [48]. Delpino FM, Figueiredo LM, Forbes SCet al. Influence of age, sex, and type of exercise on the efficacy of creatine supplementation on lean body mass: a systematic review and meta-analysis of randomized clinical trials. Nutrition. 2022;103–104. doi: 10.1016/j.nut.2022.111791 [DOI] [PubMed] [Google Scholar]
- [49]. Moriarty T, Candow D. Effects of creatine monohydrate supplementation on muscle, bone and brain- hope or hype for older adults? Curr Osteoporos Rep. 2024;23(1):1. doi: 10.1007/s11914-024-00895-x [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Appendix.docx
Data Availability Statement
All data and material reported in this review and meta-analysis were from peer-reviewed publications. The datasets supporting the conclusions of this article are included within the article and its additional files.
