The Berkson Bias in Action

Thirty years ago Berkson recognized that differences in selection rates of different diseases for admission to the hospital will systematically change the frequency with which those diseases co-exist in hospitalized patients from the frequency rate in the general population. Mainland subsequently demonstrated that postmortem studies systematically show a lower co-morbidity rate for any two individually lethal diseases than would be expected from the individual prevalence of these diseases. In studying the concurrence of bacterial endocarditis and cirrhosis, we examined the relationship of these diseases at autopsy where, according to this concept, we would expect a negative association. We found the frequency of bacterial endocarditis to be three times greater in cirrhotic than in non-cirrhotic patients, a statistically significant difference that was even more convincing, since a negative correlation was anticipated. In accord with the Berkson-Mainland hypothesis, however, no such association was seen between bacterial endocarditis and either emphysema or myocardial infarction, two other chronic diseases of different lethality. Similarly, glioblastoma multiforme, a brain tumor with a high mortality rate, showed a negative correlation with cirrhosis, emphysema, and myocardial infarction. A corollary of this bias—that the mean age at death should be lower in patients dying with two lethal diseases than in patients dying of either disease alone—was supported by our study. This investigation provides evidence to validate the Berkson-Mainland hypothesis, and suggests that rather than being always an adverse bias, it may be used beneficially to document the validity of the increased co-existence of diseases at autopsy.