|
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
|
| Abstract - #306016 |
|
Title:
|
A Model-Free Approach to Combining Diagnostic Markers
|
|
Author(s):
|
Ruth Pfeiffer*+ and Efstathia Bura
|
|
Companies:
|
National Cancer Institute and The George Washington University
|
|
Address:
|
6120 Executive Blvd., Rockville, MD, 20852,
|
|
Keywords:
|
dimension reduction ; likelihood ratio ; random matrix ; save ; singular value decomposition ; NHANES III
|
|
Abstract:
|
A popular summary measure of the discriminatory ability of a continuous diagnostic marker for binary disease outcomes is the area under the ROC curve. We will present a model-free approach to combining several markers into a composite diagnostic test. Using sufficient dimension reduction techniques, we will replace the predictor vector with a lower-dimensional version obtained through linear transformations of markers that contain sufficient information for regression of the predictors on the outcome. We will combine the linear transformations using their asymptotic properties into a scalar diagnostic score via the likelihood ratio statistic and assess its performance by the ROC curve. We will derive the asymptotic distribution of the singular values of consistent estimates of a normally distributed random matrix.
|
