Hospital statistics that contribute to epidemiology

Can a hospital population be used as a set of reference to a general population? The purpose of the present paper is to underline several conditions for which hospital statistics might be used for epidemiological inference, e.g. to map-out the pattern and the frequency of diseases in a general population. In several countries the population base for hospital in-patients is known and well-documented. Under this assumption, the frequency of diseases that require hospitalization, such as meningitis or some birth defects might be estimated in relation to the general population. In other instances, the population base remains unknown, all hospitals from a given region not being able to provide uniform statistics. There might be false associations called Berkson's bias. However, a study of the frequency of some diseases estimated from a large population of in-patients gave surprisingly similar results to those obtained from well defined populations. It is concluded that, even when the population base is unknown, hospital statistics might already contribute largely to epidemiology under two conditions: (1) the hospital sample must be large enough (above 10 000 patients) and include a case mix from several specialties (e.g., all cases from the departments of internal medicine and surgery), (2) discharge abstracts should be available, reliable and comparable for all inpatients and include all diagnoses relevant to each case. The importance of accuracy and uniformity in registration is stressed as well as the need to develop more widely population-based hospital statistics.