Regression Models for Positive and Negative Predictive Values

• Activity Number: 342
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Michael Sachs*+
• Companies: University of Washington
• Keywords: biomarkers ; regression ; predictive values ; diagnosis

Abstract

In many clinical studies, it is of interest assess the risk of incurring some binary outcome, such as death or disease, based on current medical test results. This paper focuses on the positive predictive value as a tool to assess the prospective accuracy of a continuous test. In some cases, a covariate may affect the accuracy of a continuous test, or different thresholds for high-risk versus low-risk are needed for different groups defined by some covariate. We present two regression models for the positive predictive value of a continuous test for a binary outcome. The first model directly links the positive predictive value to a linear predictor. The second model is an indirect approach. We present some theoretical results, and then illustrate the use of these models in two different studies. The first is a longitudinal study of depression, in which it is of interest to assess the change in accuracy of a screening test when administered repeatedly over time. The second illustration uses our model to develop education- and race-specific thresholds for the mini mental state examination in predicting clinical diagnosis of Alzheimer's disease.