Additional Diagnostic Utility of a Variable After the Adjustment of Other Variables

• Activity Number: 504
• Type: Contributed
• Date/Time: Thursday, August 2, 2007 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Caixia Li*+ and Ying Lu
• Companies: University of California, San Francisco and University of California, San Francisco
• Keywords: ROC curves ; Linear Discriminant Function ; Linear Transformation ; Type I error ; Multivariate F-distribution

Abstract

Diagnostic information from multiple sources is common in medicine. The additional diagnostic value of a new variable after the adjustment of other variables can be evaluated by the incremental area under (AUC) the receiver operating characteristic (ROC) curves. However, such a statistical testing procedure fails to maintain proper type I error rates for practically used sample sizes. In this paper, we investigated a new test procedure based on a standardized mean difference between the location parameters of two ROC curves after a linear transformation of all diagnostic variables. We used our method to test hypothesis of superiority or non-inferiority. We proved, through simulation studies, that such a test can maintain better type I error rates as well as unbiased estimate of the standardized effect sizes. We used an example of hip fracture to illustrate our method.