Elsevier

Mayo Clinic Proceedings

Volume 74, Issue 11, November 1999, Pages 1179-1182
Mayo Clinic Proceedings

Editorial
The Wizard of Odds: Bayes' Theorem and Diagnostic Testing

https://doi.org/10.4065/74.11.1179Get rights and content

Section snippets

How Should We Quantify the Diagnostic Accuracy of a Particular Test Procedure?

Predictive Accuracy.—When symptomatic patients with abnormal electrocardiographic stress test responses first came to be referred for coronary angiography, the proportion with disease—the predictive accuracy of an abnor­mal test response—was noted to be very high. But when the identical testing procedure was later extended to asymp­tomatic subjects, the proportion with disease was surprisingly low. This puzzling paradox was eventually resolved through Bayes' theorem, which states that the

How Should We Interpret a Particular Test Response in an Individual Patient?

Result-Specific Measures.—As noted earlier, sensitivity and specificity (and their associated likelihood ratios) are usually expressed in terms of categorical thresholds (all test responses beyond some putative cutoff being considered equivalently abnormal). These thresholds have no effect on our assessment of the test's discriminant accuracy (in terms of ROC area, for example), but they are wholly unsuited to the matter of clinical interpretation. Individual patients have a particular

Health Care implications

Arcane as they may be, quantitative measures of diagnostic testing have practical relevance to all those who provide health care resource utilization, technology assessment, and patient management, as well as to those who must ultimately judge the quality of such efforts—journal editors, granting agencies, third-party payers, and practitioners. The following examples illustrate some potential applications.

Resource Utilization.—There is some range of prior probability of disease (P) within which

References (15)

There are more references available in the full text version of this article.

Cited by (16)

  • Diagnostic Yield of Routine Noninvasive Cardiovascular Testing in Low-Risk Acute Chest Pain Patients

    2015, American Journal of Cardiology
    Citation Excerpt :

    Foy et al8 showed in 421,774 patients that risk of infarction is not different between those who do versus those who do not undergo noninvasive testing, again questioning the value of routine testing for patients in the ED with chest pain. The reasons for the low yield of routine testing are rooted in Bayes' theorem that indicates the importance of the pretest likelihood of a diagnosis on the outcome of a test in a particular population.9 Our study population consisted of patients selected to be at very low pretest likelihood of cardiovascular disease.

View all citing articles on Scopus
View full text