This study states that Regarding the statistical method used, when conducting a meta-analysis of proportions, transformations are usually performed to achieve sampling distributions closer to normal and better approximations of the sampling variance. The Freeman-Tukey double-arcsine transformation has become increasingly popular in meta-analyses because it works particularly well for normalizing and variance-stabilizing the sampling distribution of proportions compared with other techniques.3 Among the advantages of arcsine-based transformations is that the variances depend only on the sample sizes, which are typically fixed, known values. In contrast, the variances of the log and logit transformations also depend on the event counts, which are random variables.4

In the meta-analysis by Naazie et al,5 the overall event rates after transcarotid artery revascularization (TCAR) and corresponding Wald 95% confidence intervals were calculated using a logistic-normal random-effects model with logit transformation. As seen in the original studies included in both meta-analyses, many proportions of the outcomes with TCAR were close to zero.2,5 Thus, the estimate for the standard error is zero; therefore, the Wald 95% confidence intervals cannot be calculated, leading to the exclusion of studies.

Reference link-