2.5. Subgroup Meta-analysis
We intended to generate random-effects pooled effect estimates, separately for RCTs and NRSIs, but only 3 RCTs and 2 NRSIs with numerical data on primary outcomes were identified. Therefore: (i) we generated frequentist (Mantel-Haenszel relative risk, Paul-Mandel for τ2, Hartung-Knapp adjustment) and Bayesian [vaguely informative prior for ln(RR) (mean=0, SD=4, half-normal for τ] random-effects pooled estimates and prediction intervals specifically to illustrate uncertainty (CI width) and heterogeneity of the RCT outcomes (width of prediction intervals); (ii) adjusted proportions retrieved from two NRSIs were used to calculate individual study risk ratios (PVI/Cox-Maze) by the Miettinen-Nurminen method for more intuitive presentation 17. We used package meta in R18 for the frequentist and package bayesmetafor Bayesian meta-analysis in R (R Core Team (2020). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/)19,20.