Analyses and measures of treatment effect
This meta-analysis was planed to extract information based on
differences of baseline and follow up times for every trial treatment
(mean change data); nonetheless, most articles reported follow-up data
in figures and we extracted follow-up times mean differences and SD. We
applied standardized mean differences for variables pooled on the
different scales. Heterogeneity was evaluated between studies by the
Cochrane’s Q and quantified by the I2 statistic, that
demonstrated percentage of the total variation through articles that is
ascribable to heterogeneity rather than to chance. P-value of
<0.05 and level of I2>40
percent was considered as significant heterogeneity.
For estimating the overall effect, we computed the weighted mean
differences (WMDs) with 95 percent confidence intervals (CIs) using
random effects model. To measure if the outcomes could have been
affected by a single study on the overall distinctly, a sensitivity
analysis was performed (28). Subgroup
analysis was also conducted, based on follow-ups (<24, 24, 48,
72 and 96 hours after exercise), dose of creatine (20 g/day or lower and
more than 20 g/day), duration of studies (1 week and lower and more than
1 week) and train status (trained and untrained). For evaluating the
publication bias we applied visual inspection of funnel plots, Egger’s
regression asymmetry and means of Begg’s rank correlation test. Funnel
plots depicted the effect sizes (differences in means) against their
corresponding standard errors. Also, statistical analyses were conducted
applying STATA 11.2 software (StataCorp, College Station, Texas, USA).