|
Activity Number:
|
383
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 11, 2004 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
ENAR
|
| Abstract - #301463 |
|
Title:
|
Semiparametric Leasts-squares-based ROC Analysis
|
|
Author(s):
|
Zheng Zhang*+ and Margaret S. Pepe
|
|
Companies:
|
University of Washington and University of Washington
|
|
Address:
|
Dept. of Biostatistics, Seattle, WA, 98195-7232,
|
|
Keywords:
|
accuracy ; ROC curve ; regression ; biomarker ; diagnostic test
|
|
Abstract:
|
Receiver Operating Characteristic (ROC) curve is a standard statistical tool for evaluating the accuracy of a continuous diagnostic test (e.g., biomarker). It provides a complete description of the test performance and a meaningful way to compare the performances of different tests. We have developed regression models for the ROC curve based on least squares. It provides a simple yet elegant way to estimate the ROC curve using standard linear regression algorithms. Asymptotic theory shows the new estimators are asymptotically unbiased and normally distributed. Extensive simulation studies have shown the new estimators have the similar efficiency as the two previously proposed estimators and are 20-30% more efficient than the nonparametric estimators. The method is illustrated in a pancreatic cancer dataset. We have extended our method to ROC curves comparison and covariate effects modeling. For categorical covariates that are available for both diseased and nondiseased population, the method is readily applicable. Asymptotic theories have been developed for independent tests/covariates. Simulation studies and data analysis are also included.
|
