2.4. Data analysis
One reviewer extracted descriptive characteristics and effect estimates
for each outcome of interest from the included studies and recorded them
on a data extraction form, which the remaining authors verified for
accuracy. Risk of bias for each included study was assessed in
selection, performance and attrition bias (5). We described the review
findings narratively for each predefined outcome, and when feasible,
pooled the effect estimates in a random effects meta-analysis according
to the DerSimonian and Laird method (5). RevMan (Review Manager (RevMan)
[Computer program]. Version 5.4. The Cochrane Collaboration, 2020)
was used to conduct these analyses.
To obtain an explicit judgement on the confidence in the review
findings, we classified the quality of evidence for each outcome as
high, moderate, low, or very low according to the GRADE working group
guidance (7). We also integrated judgements on the quality of the
evidence into a summary of findings table (8), which included the effect
estimates for the outcomes of interest.