2.5. | Data analysis
Effect sizes were indicated as MD and 95% confidence interval (CI) and
P≤0.05 were considered as statistically significant for each parameter
in this meta-analysis. Net changes in the explored parameters (change
scores) were computed by subtracting the value at baseline from the one
after intervention, in the treatment group, and in the control one.
Standard Deviation (SDs) of the
Weighted Mean Differences (WMD) were obtained as reported by Follmann,
Elliott, Suh, and Cutler (1992): SD difference = square
root [(SD pre-treatment)2+(SDpost-treatment)2 - (2× R × SDpre-treatment× SD post-treatment)],
assuming a correlation coefficient (R) = 0.8 as it is a conservative
estimate for an expected range of 0-1. If the trials did not reported
means and (SDs) of outcome measures, we converted the available
statistical data into means and (SDs) by suitable formula: SD = SEM ×
√n, being “n” the number of subjects in any group. If medians and
inter-quartile range were reported, mean and SD values were computed by
the method described by Hozo et al[39].
I² testing performed find the potential sources of between-study
heterogeneity. Fixed effect model
chosen for meta-analysis due to (\(I^{2}\) was below 50% with p-value
<0.1)[40], and selected the random-effects model if
(\(I^{2}\) was above 50%) [41]. All analyses were performed by
STATA software (version 14.0). Potential publication bias was explored
using Egger’s regression test (Egger’s test) (21).