Potential for misleading associations between exposure and disease when reviewing multiple epidemiologic studies

In reviewing evidence from multiple epidemiologic studies, a statistical artifact can leave an impression that a relationship exists between an exposure and a disease even if none exists. In this paper, two fundamental properties of probability were applied to hypothetical situations where results from independent studies were combined. It was evident that, even when a cause-effect relationship did not exist, several studies still appeared to confirm the relationship. This statistical artifact was related primarily to the total number of independent studies reviewed and to a lesser degree to the size of a given study. Thus, in examining evidence for a cause-effect relationship from multiple published articles, some studies are likely to show the association and some not, even if no association exists. If only the studies which show the association are considered, without regard to the total number of studies, the reviewer can be misled. This statistical effect is illustrated with an example from the rubber worker literature.