Miettinen O S, Caro J J
Montreal Neurological Institute, PQ, Canada.
Stat Med. 1994 Feb 15;13(3):201-9; discussion 211-5. doi: 10.1002/sim.4780130302.
Three decades ago, the thesis was adduced that setting diagnostic probabilities requires, by the inherent nature of diagnosis-pertinent medical knowledge, the use of Bayes' theorem. That paper was both vague and inconsistent in its delineation of the nature of the parameters involved in this formulation, and subsequent authors have only added to the confusion. Nevertheless, that thesis has been, and continues to be, enthusiastically embraced by clinical scholars. We here posit what those parameters must be taken to represent in principle; and this explication reveals that their quantification poses generally unsurmountable epistemologic challenges. The implication of this is not that informed setting of diagnostic probabilities is generally infeasible. Our conclusion is, instead, that the seminal thesis was founded on an untenable pair of premises about the nature of scientifically attainable knowledge pertinent to diagnosis.
三十年前，有人提出这样一个论点，即根据与诊断相关的医学知识的固有性质，确定诊断概率需要运用贝叶斯定理。该论文在阐述这一公式中所涉及参数的性质时既模糊又不一致，后续作者也只是徒增混乱。然而，这一论点一直以来并将继续受到临床学者的热烈追捧。我们在此阐明这些参数原则上必须代表的内容；而这一阐释表明，对它们进行量化通常会带来难以克服的认识论挑战。这并不是说明智地确定诊断概率通常是不可行的。相反，我们的结论是，这个开创性的论点是基于一对关于与诊断相关的科学可获取知识的性质的站不住脚的前提。
