The Bayes’ Theorem Framework for Diagnostic Research

Abstract

Concerning the knowledge-base of diagnosis, the fundamental tenet of ‘clinical epidemiologists’ is the one that is traditional in medicine (and manifest in textbooks of medicine): that clinically-relevant knowledge generally is about particular illnesses and their respective manifestations, and that diagnosis-relevant knowledge therefore is about the relative frequencies of various case profiles in the presence (hidden) of the illness in question and that of its alternatives. This also was the predicate of an article on the theory of diagnosis in Science (in 1959), which deduced from it the idea that diagnostic probability-setting requires deployment of Bayes’ Theorem. As this outlook entails quite prohibitive problems of feasibility for diagnostic research, ‘clinical epidemiologists’ have materially reformulated that theoretical idea, thereby compromising the principles underpinning it. But more fundamentally, the idea’s predicate actually is untenable, and the Theorem is, even in principle, inapplicable for diagnosis and diagnostic research.