Assigning a cause to an individual event is a natural impulse. This impulse sometimes finds its way into epidemiologic research, if researchers assign a specific cause to some cases of a disease under study. In epidemiologic studies of drug effects, assigning a cause to specific cases is often used as a reason to exclude these cases from cohort or case-control studies, because the disease in these cases is thought to be the result of a 'known cause'. These exclusions are invalid. It is always valid, however, to exclude all subjects with 'known' causes from a study, cases from the numerator and the people or person-time at risk from the denominator (or from the control series in a case-control study). If the exclusions from the denominator would be negligibly small or the relative number of excluded cases with 'known' causes is small, then the exclusion of just the cases with 'known' causes will introduce little bias. Copyright (C) 2002 John Wiley Sons, Ltd
