Estimating the Agreement and Diagnostic Accuracy of Two Diagnostic Tests When One Test Is Conducted on Only a Subsample of Specimens

• Activity Number: 126
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Hormuzd A. Katki*+ and Yan Li and Philip E. Castle
• Companies: National Cancer Institute and The University of Texas at Arlington and National Cancer Institute
• Keywords: two-phase design ; verification bias ; gold standard ; diagnostic testing

Abstract

Specimen repositories are used to compare a new diagnostic test against pre-existing tests for agreement and diagnostic accuracy. Usually all pre-existing diagnostic tests will already have been conducted on all specimens. To minimize study costs and specimen consumption, we propose to retest only a judicious subsample of specimens by the new diagnostic test. We introduce methods to estimate agreement statistics and diagnostic accuracy when one test is conducted on only a subsample. The methods use inverse-probability weighting by treating the subsample as a stratified two-phase sample. Stratification targets the most informative specimens for retesting. Unlike verification bias, subsampling is under investigator control. We present examples where subsampling achieved adequate statistical efficiency and greatly reduced the number of specimens requiring re-testing.