A model for incorporating historical controls into a meta-analysis.

Overview

A method is presented for estimating the treatment effect in a meta-analysis when some of the studies are comparative studies, and the rest are noncomparative, historical control studies. A random-effects model is used in which the baseline effect in each study is random, but the treatment effect is constant. With this model the appropriate contribution of the historical studies can be determined. Extensions of the model are developed to accommodate preliminary tests for bias, and for the possibility that the treatment effect is heterogeneous.