A meta-analysis methodology for utilizing study-level covariate information from clinical trials

Covariate information at the patient level is generally not available for a meta-analysis of clinical trials. Published papers or unpublished reports of clinical trials selected for a meta-analysis generally do not include such information at the patient level. However, such papers and reports usually include design and covariate information at the study level. For example, they generally report important patient population characteristics for the study in the form of descriptive statistics including design information, such as, dose levels, duration of treatment information, gender and race proportions, etc. This paper addresses a meta-analysis methodology for binary responses based on mixed-effects modeling which makes use of such study-level covariate information. This paper also reviews fixed and random effects meta-analysis methods which become special cases of the mixed-effects methodology.