In meta-analyses of proportion studies, funnel plots were found to be an inaccurate method of assessing publication bias