A critical appraisal of the Bayesian approach to diagnose coronary artery disease.

The impact of disease prevalence on the predictive accuracy of a diagnostic procedure can be calculated using the Bayes theorum. In general, there is an inverse relation between the prevalence of disease and the predictive accuracy of the diagnostic test. However, at high prevalence of the disease, the diagnostic accuracy of other means of diagnosis is so great (that is, typical exercise-rest relation of angina in a middle-aged man with several coronary risk factors) that there may be little clinical value in performing the test. Limitations of Bayesian approach. The limitations and pitfalls of using the Bayesian approach to analyze clinical decision-making have been thoroughly discussed in a delightful essay by Alvan R. Feinstein entitled, "The Haze of Bayes, The Aerial Palaces of Decision Analysis, and the Computerized Ouija Board" (1). Feinstein summarized the many disparities between the mathematical simplicity of the Bayes formula and the realities of clinical reasoning and clinical data. Among these are the fact that the formula computes a numerical probability that is too vague and uncertain for the demands of modem diagnostic precision. If a patient has angina and it is the judgment of the clinician that the diagnosis of coronary artery disease must be established, coronary arteriography will be selected. This test will reliably exclude the absence of coronary disease (so that its probability is nearly 0) or confirm its existence (probability = 100%). One cannot insert the need to establish the diagnosis in the Bayesian formula. In addition, Bayesian clinical logic is aimed at diagnosis and it contains no provision for the situation where clinical reasoning terminates with a trial of nitroglycerin, rather than additional diagnostic tests.