|
Activity Number:
|
372
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 6, 2003 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #300808 |
|
Title:
|
On Comparison of Two Classification Methods with Survival Endpoints
|
|
Author(s):
|
Ying Lu*+ and Hua Jin and Jie Mi
|
|
Companies:
|
University of California, San Francisco and University of California, San Francisco and Florida International University
|
|
Address:
|
3333 California St., Suite 375, San Francisco, CA, 94118,
|
|
Keywords:
|
classifications ; variance of mean survival time ; degree of separation ; noninferior test
|
|
Abstract:
|
Methods of prognostic prediction and classification, such as staging systems for diseases, are commonly used to assist decision making in clinical management. Multiple classification methods are often available for the same disease population. Comparison of these methods to identify the optimum classification for patient survival or to determine noninferior alternatives to the optimum classification is necessary. We used variance on the restricted mean lifetimes as an index for the degree of separation (DOS) to measure the efficiency in prognostic separation by a classification method. This index was used to compare the efficiency of two classification methods with survival time as the endpoint. We provided mathematical relationships between DOS and partial ordering of classification methods. We proposed estimating and testing procedures of two DOSs based on paired data. Our simulation results suggested that the proposed testing procedure have a desired type I error rate under reasonable study sample sizes. Two clinical examples demonstrated its uses in medical research.
|
