Toward a Clearer Definition of Confounding

This paper illustrates that bias can result when adjustment is made for any factor which is caused in part by the exposure under study and is also correlated with the outcome under study.

Abstract

Epidemiologists are aware that the estimated effect of an exposure can be biased if the investigator fails to adjust for confounding factors when analyzing either a prospective or retrospective etiologic study. Standard texts warn, however, that intervening factors are an exception: one should not adjust for any factor which is intermediate on the causal pathway between the exposure and the disease. Other factors which are not on the causal pathway but are caused in part by the exposure are often adjusted for in epidemiologic studies. This paper illustrates that bias can result when adjustment is made for any factor which is caused in part by the exposure under study and is also correlated with the outcome under study. Intervening variables are only one example of this phenomenon. The misleading effects of this practice are illustrated with examples.