|
Activity Number:
|
110
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 9, 2004 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Nonparametric Statistics
|
| Abstract - #300691 |
|
Title:
|
Practical Consideration on Optimal Combination of Multiple Tests
|
|
Author(s):
|
Hua Jin*+ and Ying Lu
|
|
Companies:
|
University of California, San Francisco and University of California, San Francisco
|
|
Address:
|
185 Berry St., San Francisco, CA, 94107,
|
|
Keywords:
|
multiple diagnostic predictors ; receiver operating characteristic (ROC) curve ; likelihood ratio score ; Mann-Whitney statistic ; permutation test
|
|
Abstract:
|
Multiple alternative diagnostic tests for one disease are commonly available to clinicians. It's important to use all the good diagnostic predictors simultaneously to establish a new predictor with higher statistical utility. The likelihood ratio score leads to the uniformly most powerful test achieving the largest area under the receiver operating characteristic (ROC) curve on the basis of the famous Neyman-Pearson. However, it seems complex for practical uses. One may ask when it is possible to replace use of the complicated likelihood ratio score with a relatively simple forms such as the most commonly used linear combinations. We propose a formal statistical framework to deal with this problem. We used the Mann-Whitney statistic to estimate the area under the ROC curve and a permutation reference distribution to test the null hypothesis that the simple linear or polynomial combinations of markers have similar area under ROC curves as those of the maximum likelihood ratio scores. Monte Carlo simulations were conducted to evaluate the performance of the proposed test. The algorithm applied to data from the Study of Osteoporotic Fractures (SOF) for illustration.
|
