[Generation and evaluation of etiologic hypotheses in epidemiology]

So far the problems of the generation and evaluation of etiologic hypotheses have been of too little concern to epidemiologists. Epidemiologic research usually deals with two fundamental etiologic questions: the first is 'why' an epidemiological phenomenon occurs; the second is 'how', and the question relates to the mediating mechanism. After having defined the nature of a valuable working hypothesis, we identify several ways by which hypotheses might be merged, and discuss for each of them the corresponding problems of evaluation. It is advocated that 'biological' (clinical) induction is a promising way of gaining insight into the etiology of disease. 'Statistical' induction, on the other hand, may be a useful though precarious way of generating hypotheses, mainly because of evaluation problems. As to deductive thinking, sometimes based on the process of analogy, its use is confined to rather well developed fields of knowledge. Furthermore, deductive thinking may be plagued by logical errors if the biological model of the studied disease is inadequate. Finally, Popperian deduction is also showed to be subjected to logical flaws if the causal model of relationships is poorly grounded into the biology of disease; specifically, the point is made that the refutation or confirmation of hypotheses is not straightforward and may be prevented in cases of complex biological situations. We conclude that epidemiologic analysis may be furthered by contacts with clinicians who would be of invaluable help in formulating and the testing of hypotheses, by study designs devised to reject specific models of disease, and by the rooting of putative causes of disease beyond simple risk markers.