Authors' reply: Variation of sensitivity, specificity, likelihood ratios and predictive values with disease prevalence by H. Brenner and O. Gefeller, Statistics in Medicine, 16, 981–991 (1997)

✍ Scribed by Hermann Brenner; Olaf Gefeller

• Publisher: John Wiley and Sons
• Year: 1998
• Tongue: English
• Weight: 50 KB
• Volume: 17
• Category: Article
• ISSN: 0277-6715
• DOI: 10.1002/(sici)1097-0258(19980430)17:8<947::aid-sim823>3.0.co;2-q

✦ Synopsis

sensitivity from a random sample of ill subjects, although the disease prevalence itself cannot be assessed. In other words, sensitivity is independent of the disease prevalence. This fact naturally holds only within the same probability space. If one changes the diagnostic criterion or the population, then the efficacy measures of the diagnostic test may not be constant. The efficacy of diagnostic tests is dependent on the population characteristics. If the sample used for estimation does not contain the whole spectrum of ill and healthy patients, respectively, estimation of test efficacy is subject to spectrum bias. This was described by Ransohoff and Feinstein nearly 20 years ago and was repeatedly reported in a number of papers thereafter. If the population characteristics change, sensitivity and specificity as well as disease prevalence may also change. Hence, the observed variation of sensitivity and specificity in the model of Brenner and Gefeller is trivial, but this observation does not mean that sensitivity and specificity vary in dependence on disease prevalence. The main point is the variation of the population characteristics not that of the disease prevalence. For example, it is possible that the sensitivity of a diagnostic test is different between two populations with equal disease prevalence but with more seriously ill persons in one population. Hence, it is misleading to say that sensitivity and specificity vary with disease prevalence.

It is to the credit of the authors that they underline the impact of measurement error on the validity of diagnostic tests. However, the calculations based on their simple model are by no means a theoretical framework for the explanation or quantification of the dependence of sensitivity and specificity on general population characteristics.