Data Synthesis and Analysis
Data were pooled and analyzed by using Comprehensive
Meta-Analysis Software, version 2 (Biostat, Englewood,
New Jersey). Because of low event rates for several
clinical outcomes, we used Peto odds ratios (ORs) (fixedeffects
model). Weighted mean differences were calculated
by using a random-effects model. Statistical heterogeneity
between trials was assessed by using the I2 test; a value of
50% or greater suggests moderate to substantial inconsistency
among studies (21). To explore heterogeneity, we
did subgroup analyses and tested for interactions (see the
sensitivity analysis section). The extent of publication bias
was evaluated through visual inspection of funnel plot
asymmetry and the linear regression– based test proposed