Data synthesis and statistical analysis
The
statistical analysis was performed with the statistical software open
source R program (Version 3.41). Categorical outcomes were estimated by
odds ratio (OR) with 95% CI. If “1” falls into the CI, the outcome
would be considered “not statistically significant”. The mean
difference (MD = mean of primary IOL – mean of primary aphakia) with a
95% confidence interval (CI) was adopted for continuous
outcomes. If “0” falls into the
CI, the outcome would be considered “not statistically significant”.
Heterogeneity across studies was
tested with Q test and I2 statistic. If there was no
heterogeneity across these studies (I2<50% ), the
Mantel–Haenszel fixed-effect model would be used.
I2 above 50%
and the P-value below 0.1 would constitute a significant heterogeneity
among these studies, and possible reasons would be explored by reviewing
the included studies. Three approaches were used to detect the source of
heterogeneity in this meta-analysis: sensitive analysis, subgroup
analysis and meta regression. Through sensitivity analysis, we could
determine whether the heterogeneity would decrease following exclusion
of each study one by one. Subgroup analysis and meta regression would be
performed according to the clinical characteristics of the included
studies. Through these two approaches, we could identify the factor that
induced heterogeneity in the meta-analysis. If sensitivity analysis and
subgroup analysis did not decrease the heterogeneity, the random-effects
model (DerSimonian-Laird) would be adopted to calculate pooled effect
size.28,29.