Random-effects linear regression meta-analysis models with application to the nitrogen dioxide health effects studies.

Abstract

As the field of epidemiology grows and multiple studies of the same topic are more frequently available, increased focus is placed on quantitative methods for synthesis of results to yield an overall conclusion. A major difficulty encountered in practice has been the lack of convenient methodology for addressing groups of studies which are similar, but not exactly alike, in features which may affect study results. The age group from which subjects were selected, the general health of subjects when selected, and the specific health endpoint examined are examples of such features. Some previous investigators have addressed the problem using iterative techniques, although most have opted for simpler models which assume that differences in the studies do not appreciably affect the outcome under investigation. That is, he studies are taken to be homogeneous in that the underlying effect being investigated is the same in each study. This paper presents a random-effects linear regression technique which allows differences in the individual study features. The proposed methodology does not require iterative or other complicated procedures, making it more readily accessible to the applied researcher. We demonstrate this technique on a set of studies of the health effects of indoor NO2 exposure in children. It is seen that odds ratios from these studies vary considerably according to subject age, the study location, and the health endpoint considered. A simple synthesis which does not account for these differences may be misleading.