Analyzing Polytomous Data from Combined Use of Two Tests in Cervical Cancer Screening: Predictive Values, Specificity and Sensitivity

• Activity Number: 437
• Type: Topic Contributed
• Date/Time: Thursday, August 7, 2003 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Teng S. Weng*+
• Companies: Food and Drug Administration
• Keywords: Pap smear test ; HPV DNA test ; combination test ; polytomous data ; predictive values ; diagnostic accuracy

Abstract

Current interest in cervical cancer screening has focused on the combined use of conventional Pap smear test and the newly introduced human papilloma (HPV) DNA test. In spite of both tests giving polytomous outcomes, the current practice is to dichotomize the outcomes of each test as positive (+) or negative (-) whereby a specimen is deemed as + if either test is +, and - otherwise. Those women having a + result are followed up with a colposcopy/histology procedure (as gold standard) to detect cervical intraepithelial neoplasias (CIN). In this paper, a loglinear model (LLM) was used to fit the polytomous data from a published study whereby predictive values for future histologic outcomes (CIN+ or CIN-), given the two tests' joint outcomes, were obtained via a logit parameterization. These values were then utilized to construct the specificity and sensitivity for the combined test "relative to the observed CIN status." This approach is capable of fully exploiting the clinical information revealed by the combined test. Not only was the dilution of clinical information caused by dichotomization avoided, but also was the necessary conditional independence assumption totally abolished.