|
Activity Number:
|
217
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 3, 2009 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistics in Epidemiology
|
| Abstract - #304925 |
|
Title:
|
Improvement in Performance by Combining Biomarkers in Diagnostic Medicine
|
|
Author(s):
|
Aasthaa Bansal*+ and Margaret Pepe
|
|
Companies:
|
University of Washington and University of Washington
|
|
Address:
|
Department of Biostatistics, Seattle, WA, 98195,
|
|
Keywords:
|
diagnostic tests ; classification ; biomarkers ; ROC
|
|
Abstract:
|
BACKGROUND: When an existing marker does not have sufficient diagnostic accuracy on its own, new markers are sought with the goal of yielding a combination with better performance. Understanding the properties that a new marker should have in order to improve performance would help biomarker development. METHODS: Assuming the joint distribution of baseline and new markers is bivariate normal in cases and controls, we quantify the improvement in the ROC curve as a function of the correlations and the relative performance of the new marker. RESULTS: The ROC is typically improved substantially only when the new marker performs well on its own and is weakly correlated with the existing marker. Surprisingly, in some realistic settings a very highly correlated marker can yield substantial improvement. We also show that combining markers incorrectly can lead to decrements in performance.
|
