Arcsine‐based transformations for meta‐analysis of proportions: Pros, cons, and alternatives

Abstract

Meta‐analyses have been increasingly used to synthesize proportions (eg, disease prevalence) from multiple studies in recent years. Arcsine‐based transformations, especially the Freeman–Tukey double‐arcsine transformation, are popular tools for stabilizing the variance of each study's proportion in two‐step meta‐analysis methods. Although they offer some benefits over the conventional logit transformation, they also suffer from several important limitations (eg, lack of interpretability) and may lead to misleading conclusions. Generalized linear mixed models and Bayesian models are intuitive one‐step alternative approaches, and can be readily implemented via many software programs. This article explains various pros and cons of the arcsine‐based transformations, and discusses the alternatives that may be generally superior to the currently popular practice.

Bayesian model selection favors parametric over categorical fMRI subsequent memory models in young and older adults

“…Parametric regressors with more than two values are commonly coded as linear scales (Heinzel et al, 2005; Northoff et al, 2009) and, in the present study, a linear scale of recognition confidence was used in our default parametric model. We additionally employed an arcsine-transformed scale, which was inspired by the use of arcsine transformations for proportion data in statistics (Hernández et al, 2018; Lin and Xu, 2020). The choice of arcsine over other inverse sigmoid transformations like the logit or Fisher z-transformations was due to the non-asymptotic nature of the arcsine function.…”

The role of asymptomatic and pre-symptomatic infection in SARS-CoV-2 transmission—a living systematic review

“…A pooled analysis was carried out to generate summary estimates for the secondary attack rate in each subgroup analyzed (asymptomatic, pre-symptomatic and symptomatic index cases), in the framework of a random effect model. The Freeman – Tukey double-arcsine variance-stabilizing transformation was used to combine data, due to its advantage over log and logit transformations which did not allow to compute the proportion in the presence of zero event counts [ 15 ]. Secondary attack rates are presented as a proportion along with 95% CIs in forest plots.…”

Antibiotic prescribing in patients with COVID-19: rapid review and meta-analysis

“…Fourth, we stratified by study age group to evaluate differences in antibiotic prescribing between pediatric and adult patients. We pooled proportion data across studies via a random-effects meta-analysis using a generalized linear mixed model (GLMM) with logit link approach ( 10 , 11 ). Results were illustrated using forest plots.…”