All results were reported with the random-effects model.
The Q statistic method was used to assess statistical heterogeneity
and the I-squared (I2) statistic was used to
evaluate the inconsistency between trials.14 15 All absolute
risk difference estimates were pooled by using the
DerSimonian and Laird methodology.16 We choose the
risk difference over the risk ratio in order to better
describe the direct clinical effects of our findings. The
quality of each trial was evaluated by the Jadad criteria.
Preferred reporting items for systematic reviews and
meta-analyses guidelines17 for reporting meta-analysis
were followed. All analyses were adjusted for the study
design to account for potential ascertainment bias. The
software used was Comprehensive Meta-Analysis V.2.0
(Biostat, Englewood, New Jersey, USA). Egger regression
and Begg and Mazumdar18 methods were used to evaluate
publication bias. Statistical power calculations were
performed based on the comparison of two independent
proportions by χ2 testing using the software Power
and Precision V.4.0 (Englewood, New Jersey, USA). The
two-group test of proportions was used to test the null
hypothesis that the proportion of cases meeting the