Bayesian hierarchical models for complex meta-analyses using MCMCglmm in R

Download poster as PDF

Download poster as PDF

Abstract text
Background: Available easy-to-use software packages for meta-analysis do not cope with multiple nested or correlated structures in data (e.g. subjects, hospitals, regions, countries) whereas more flexible and fully capable packages such as WinBUGS require detailed coding. Objectives: We provide a happy-medium solution to this problem by introducing a free R package whose use is relatively easy and whose capability is nearly as good, if not better, as other command-based software packages. Methods: Meta-analytic capability has been implemented in MCMCglmm, an R-pacakge for running generalized linear mixed-effects models using MCMC algorithms (i.e. Bayesian hierarchical or multilevel models). MCMCglmm can take arbitrary numbers of fixed effects (moderators) and random effects including a vector of measurement (sampling error) variances and, furthermore, it is able to incorporate any number of correlation matrices. Results: MCMCglmm has been used in a number of meta-analytic studies, especially in the fields of ecology and evolution, where data are highly complex and heterogeneous due to the inclusion of multiple populations and species. As an illustration of this fairly easy-to-use R package, we present our ‘comparative meta-analysis’ on the effects of dietary restriction on longevity across 36 species and over 100 studies. Conclusions: MCMCglmm can appropriately model complex meta-analytic data in a relatively easy manner and has potential to be used in meta-analysis in medical and social sciences.
Authors
Nakagawa S1, Hadfield JD2
1 Department of Zoology, University of Otago, New Zealand
2 Department of Zoology, University of Oxford, UK
Presenting author and contact person
Presenting author: 
Shinichi Nakagawa
Contact person Affiliation Country
Shinichi Nakagawa (Contact this person) University of Otago New Zealand