On the problem of controlling confounding in case-referent studies.

Individuals in various exposure categories constitute an epidemiological study population and its health experience over time can be taken as the base of the study, which is either open with a turnover of individuals (deaths, migrations etc) or closed, when encompassing only originally defined individuals. In cohort studies all individuals are known in terms of membership and exposure, whereas case-referent (case-control) studies utilize a sample from the base, the referents, for information about the exposed and non-exposed domains of the base, and the referents are a sort of relative denominators for the cases. An association in the base between the exposure and another determinant of the disease means confounding and can be fully accounted for in the cohort approach, but not in the case-referent study as relying on a sample of the base. The reason is that the occurrence of the confounder among the referents is influenced by random variation in the sampling procedure and the referents may therefore not always reflect the true situation with regard to confounding in the base. Computer simulations of the acquisition process of referents illustrate that neither a confounded nor a non-confounded base might always be properly reflected in case-referent data. However, the variation of the confounding rate ratio (the crude rate ratio/SMR) might not be expected to vary by more than about a factor of two. Particularly in nested case-referent studies, a relatively large number of referents helps stabilize the random variation, but only the cohort approach provides full control of confounding, a fact, that becomes the more important, the smaller the effect under study.(ABSTRACT TRUNCATED AT 250 WORDS)