Bayesian comparison of cost-effectiveness of different clinical approaches to diagnose coronary artery disease.

The objective of this study was to compare the cost-effectiveness of four clinical policies (policies I to IV) in the diagnosis of the presence or absence of coronary artery disease. A model based on Bayes' theorem and published clinical data was constructed to make these comparisons. Effectiveness was defined as either the number of patients with coronary disease diagnosed or as the number of quality-adjusted life years extended by therapy after the diagnosis of coronary disease. The following conclusions arise strictly from analysis of the model and may not necessarily be applicable to all situations. As prevalence of coronary disease in the population increased, it caused a linear increase in cost per patient tested, but a hyperbolic decrease in cost per effect, that is, increased cost-effectiveness. Thus, cost-effectiveness of all policies (I to IV) was poor in populations with a prevalence of disease below 10%, for example, asymptomatic people with no risk factors. Analysis of the model also indicates that at prevalences less than 80%, exercise thallium scintigraphy alone as a first test (policy II) is a more cost-effective initial test than is exercise electrocardiography alone as a first test (policy I) or exercise electrocardiography first combined with thallium imaging as a second test (policy IV). Exercise electrocardiography before thallium imaging (policy IV) is more cost-effective than exercise electrocardiography alone (policy I) at prevalences less than 80%. 4) Noninvasive exercise testing before angiography (policies I, II and IV) is more cost-effective than using coronary angiography as the first and only test (policy III) at prevalences less than 80%. 5) Above a threshold value of prevalence of 80% (for example patients with typical angina), proceeding to angiography as the first test (policy III) was more cost-effective than initial noninvasive exercise tests (policies I, II and IV). One advantage of this quantitative model is that it estimates a threshold value of prevalence (80%) at which the rank order of policies changes. The model also allows substitution of different values for any variable as a way of accounting for the uncertainty inherent in the data. In conclusion, it is essential to consider the prevalence of disease when selecting the most cost-effective clinical approach to making a diagnosis.