[The difficulty of choosing a good control group in a retrospective study (author's transl)].

The choice of either hospital or population controls in retrospective studies is guided by rules implying underlying hypotheses. In a study on aesophageal cancer in relation to alcohol and tobacco consumption, two such control groups were available and were compared. It was found that some of the currently accepted assumptions relating to representativeness of cases or controls were verified but many others were not, particularly in hospital controls. These were found to give less correct estimates of drinking and smoking habits than those obtained from population controls. The limitations in the use of either type of controls is discussed, in relation to the purpose of the study concerned.