ABSTRACT
Meta-analytic methods have been developed for determining the effectiveness of a single treatment versus a control or standard by combining such comparisons over a number of studies. When the endpoint is dichotomous (e.g., effective or not effective, side effects present or absent, life or death), effectiveness is typically measured in terms of differences in risks (proportions) or odds ratios. In large studies, more than one treatment may be involved, with each treatment being compared to the common control. This is particularly true of pharmaceutical studies, in which the effects of several drugs or drug doses are compared in order to identify the most promising choices. Because investigators may have different goals, or are prevented by financial or other constraints from testing all possible treatments, different studies may involve different treatments. When later a meta-analytic review is attempted of all studies that involve the treatments of interest to the researcher, the facts that some studies may be missing one or more of the treatments and that (because of the common control) the effect sizes within studies are correlated need to be accounted for in the statistical analysis.
