Background: Various factors in the conduct of clinical trials may bias the intervention effect. On the basis of theoretical and empirical considerations, Cochrane identifies several factors to be assessed for risk of bias in randomized trials. Empirical evidence comes from meta-epidemiology, where the intervention effects from the studies in a collection of meta-analyses are compared between trials with and without a particular characteristic. Meta-epidemiological studies can be performed with a two-step approach (Sterne 2002) or a one-step approach (Siersma 2007). Investigation of heterogeneity, sensitivity analyses and graphics are key points. Outcomes can be continuous or discrete.
Objectives: To develop an R package for standardizing the procedure of a meta-epidemiologic study.
Methods: Several functions are developed in the R package allowing for: 1) fitting two- and one-step models; 2) performing meta-regression and subgroup analyses; 3) drawing Baujat plots, forest plots and funnel plots. The package’s functions are illustrated on data from a meta-epidemiologic study investigating the impact of deviations from intention-to-treat on the intervention effect (provisionally accepted for publication).
Results: The functions of the package deploy the R utilities in manipulating data, the lme4 package for mixed-effects modelling and the base graphics system. The functions are easy to use and allow for well-structured meta-epidemiologic analysis.
Conclusions: The R package metaepi is a convenient tool for facilitating the statistical analysis in a meta-epidemiologic study and can provide guidance for the steps required in performing such challenging task.