Beta-Binomial Models for the Meta-Analysis of Binary Outcomes



Oliver Kuß, German Diabetes-Center

Annika Hoyer, German Diabetes-Center

Day: Sunday, September 6

Length of tutorial: half day

Technical level: medium


In recent years, several researchers argued that standard inverse-variance methods should no longer be used for the meta-analysis of binary outcome data. Actually, such methods can be avoided because the information needed (e.g. binary outcome, study, treatment group, true disease status) is in general available. Therefore, there is no need to calculate summary effect measures per single study which are summarized in a second step in a weigthed estimate. Instead, well-known methods for logistic regression models with correlated outcomes can be used which  avoid the assumption of known weights while they are actually estimated in inverse-variance models. In previous work, we successully applied beta-binomial models for the meta-analysis of binary outcomes in a variety of study designs. This tutorial aims to introduce such models with respect to statistical theory as well as hands-on computation in SAS and R.

Local Organizer:

Institute of Biometry and Clinical Epidemiology
Charité - Universitätsmedizin Berlin & Berlin Institute of Health