Analysis
All analyses were performed using STATA 14.012. We have reported
results of both fixed and random effects meta-analysis.
Where some summary data were missing for the reported trial results, we
have imputed missing information (e.g. standard deviations or
intra-cluster correlation coefficients (ICCs)) using other information
in the trial paper or values from similar trials13. Sensitivity to
imputed values was assessed by imputing a range of different values.
Where the reported outcome data were from either an individually
randomised trial or a cluster trial where the results had already been
adjusted for clustering by the unit of randomisation, the standard
errors were utilised without adjustment. Where adjustment for clustering
was required the standard error was multiplied by the square root of the
design effect (DE); this requires the average cluster size (M) and the
intra-class correlation coefficient
(ICC)13.
\(DE=\) 1+(M-1)ICC
Where possible we report the I2 statistic as a measure
of heterogeneity 13.
The I² statistic estimates the percentage of variation across studies
that is due to study heterogeneity rather than chance.