|
Activity Number:
|
419
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Biopharmaceutical Section
|
| Abstract - #309059 |
|
Title:
|
Comparison of Predictive Values of Two Diagnostic Tests Using Conditional Model
|
|
Author(s):
|
Yoonjin Cho*+ and Andrzej S. Kosinski
|
|
Companies:
|
North Carolina State University and Duke University
|
|
Address:
|
1408 crest Rd 103, Raleigh, NC, 27606,
|
|
Keywords:
|
Positive predictive value ; Negative predictive value ; Diagnostic test ; Conditional model ; Logistic regression
|
|
Abstract:
|
Positive and negative predictive values are important measures when one compares performance of two diagnostic tests. Thus, the development of statistical methodology to test equality of these measures between two diagnostic tests is essential. In this presentation we consider a paired design in which each subject has both tests performed. We first review the existing methods for comparison of predictive values in such a design. The first is based on marginal regression and the second is weighted least squares. We propose a conditional model approach. Our model uses logistic regressions and Wald statistic is used to test equality of positive or negative predictive values between the two diagnostic tests. We conduct simulations under a variety of scenarios to evaluate type I error and power of the proposed testing approach and compare to the other two existing methods.
|
