An Examination of the Accuracy of a Composite Reference

• Activity Number: 194
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Ian Schiller*+ and Nandini Dendukuri and Michael Libman and Alula Hadgu
• Companies: Division of Clinical Epidemiology, McGill Univerity Health Centre and McGill University and Division of Infectious Diseases, McGill Univerity Health Centre and Centers of Disease Control and Prevention
• Keywords: Diagnostic tests ; Imperfect reference ; Composite reference ; Sensitivity ; Specificity ; Chlamydia Trachomatis

Abstract

In the absence of a gold-standard diagnostic test for a disease, new diagnostic tests may be evaluated by comparison to a composite reference standard (CRS). The CRS is defined using routinely used, but imperfect tests. Though this method is appealing due to its perceived simplicity compared to complex statistical models, the biases associated with it are not well understood. We derive expressions for the sensitivity and specificity of a CRS, and illustrate how it is affected by the number of component, imperfect tests. We also derive expressions for the sensitivity and specificity of a new test with respect to the CRS and show that at least one of these may be biased, with the bias increasing with the number of tests used to define the CRS. We consider cases where the new test is independent or dependent of the tests in the CRS conditional on the disease status. We illustrate these ideas using simulated data as well as by application to examples involving diagnostic tests for C. trachomatis.