We would like to thank Dr Papageorgiou for his interest in our article. We are grateful for the opportunity to offer a few clarifications. While conducting and reporting this meta-analysis, we strictly followed the MOOSE statement, which was released in 2000 and is probably the only guideline for meta-analysis of observational studies in epidemiology.
Compared with meta-analyses of randomized controlled trials, meta-analyses concerning prevalence (MCPs) are relatively new and rare in the literature. Variations in methodology are common. First, many authors of MCPs based their choice of summary methodology on the significance of heterogeneity, and we referred to them. In our analysis, a switch from a fixed-effect model to a random-effect model would generate no change in the conclusions.
Second, owing to the commonly found considerable heterogeneity across studies and a chief aim to merely provide pooled prevalence, many MCPs reported subgroup differences only in a descriptive manner rather than ambitiously giving the statistical significance of the differences. According to Higgins and Green, nonoverlap (and even overlap to a small degree) of the confidence intervals of the summary estimates is an indication of statistical significance. To avoid exaggeration, we used this relatively conservative method to estimate the rough degree of differences while describing them.
As for publication bias, previous MCPs either made no mention or provided only an overall value of relevant statistics for all included studies (same as what we did). Actually, the chief concern for publication bias was that studies with negative results are less likely to be published. However, results of epidemiologic surveys concerning prevalence could not be considered either positive or negative. Furthermore, a previous study has shown that study size was not consistently associated with the probability of publication. Thus, to MCPs, the importance of publication bias and the traditional methods to estimate it might not be absolutely applicable. Conversely, information bias, selection bias, and performance bias have been highlighted for MCPs, all of which we endeavoured to avoid and evaluate in our analysis.
For the last 3 or 4 years, MCPs have been rapidly gaining the interest of researchers. The elimination of inconformity and the improvement of quality of future MCPs might be one task that the Cochrane Collaboration could undertake before its 25th anniversary!