Limitations of Sensitivity, Specificity, Likelihood Ratio, and Bayes' Theorem in Assessing Diagnostic Probabilities: A Clinical Example

• Published in: Epidemiology (Cambridge, Mass.) Mass.), 1997-01, Vol.8 (1), p.12-17
• Main Authors: Karel G. M. Moons, van Es, Gerrit-Anne, Deckers, Jaap W., J. Dik F. Habbema, Grobbee, Diederick E.
• Format: Article
• Language: English
• Subjects: Adult; Aged; Bayes Theorem; Biological and medical sciences; Cardiology. Vascular system; Confidence Intervals; Coronary artery disease; Coronary Disease - diagnosis; Coronary heart disease; Cross-Sectional Studies; Disease models; Epidemiology; Exercise Test - statistics & numerical data; Female; Heart; Humans; Likelihood Functions; Male; Medical diagnostic tests; Medical sciences; Middle Aged; Patient medical history; Physical examinations; Selection bias; Sensitivity and Specificity; Stress tests; Workloads
• ISSN: 1044-3983; 1531-5487
• DOI: 10.1097/00001648-199701000-00002

We evaluated the extent to which the sensitivity, specificity, and likelihood ratio of the exercise test to diagnose coronary artery disease vary across subgroups of a certain patient population. Among 295 patients suspected of coronary artery disease, as independently determined by coronary angiography, we assessed variation in sensitivity and specificity according to patient history, physical examination, exercise test results, and disease severity in 207 patients with and 88 patients without coronary artery disease, respectively. The sensitivity varied substantially according to sex (women 30% and men 64%), systolic blood pressure at baseline (53% to 65%), expected workload (50% to 64%), systolic blood pressure at peak exercise (50% to 67%), relative workload (33% to 66%), and number of diseased vessels (39% to 77%). The specificity varied across subgroups of sex (men 89% and women 97%) and relative workload (85% to 98%). The likelihood ratio varied (3.8 to 17.0) across the same patient subgroups, as did the sensitivity. As each population tends to be heterogeneous with respect to patient characteristics, no single level of these parameters can be given that is adequate for all subgroups. Use of these parameters as a basis for calculating diagnostic probabilities in individual patients using Bayes' theorem has serious limitations.