Confounding in epidemiologic studies: the adequacy of the control group as a measure of confounding.

A model for confounding in epidemiologic studies is developed based on the adequacy of the "control" group, instead of the widely used criterion of collapsibility, as a measure of confounding. It is shown that conditions for no confounding in both cohort and case-control studies derived under this model generally agree with the conditions for confounding derived inductively by Miettinen and Cook (1981, American Journal of Epidemiology 114, 593-603). The concept of confounding on which this model is based is compared with the collapsibility criterion in terms of its utility in the design and analysis of epidemiologic studies.