Diagnostic Imaging Procedures: Defining and Analyzing Test Results To Account for Unknown Disease Loci

• Activity Number: 407
• Type: Topic Contributed
• Date/Time: Wednesday, August 9, 2006 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Gene Pennello*+ and Brandon D. Gallas
• Companies: U.S. Food and Drug Administration and U.S. Food and Drug Administration
• Keywords: partial gold standard ; conditional independence ; verification bias ; Bayesian methods ; free-response receiver operating characteristic curve ; latent class

Abstract

Diagnostic imaging is a vital tool in helping to identify and biopsy tissue suspicious of being diseased. Examples are use of imaging in colposcopy and mammography to search the cervix and breast for cancerous loci. At the subject level, common definitions for the diagnostic test result do not account properly for unknown disease loci. This causes logical problems, such as defining a diseased subject as a false positive because all of the biopsies are negative. The problems are apparent in studies with and without complete disease ascertainment (i.e., follow-up and nonfollow-up studies). We consider alternative definitions for subject-level diagnostic test results in an attempt to address unknown disease loci. Under these definitions, we explore statistical methods of analysis that can be used to estimate diagnostic performance (e.g., sensitivity, specificity, FROC).