|
Activity Number:
|
388
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 15, 2002 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Biometrics Section*
|
| Abstract - #301197 |
|
Title:
|
Weighted Logrank Tests for Correlated Failure Time Data
|
|
Author(s):
|
Jeffrey Buzas*+ and Joe Carpenter and Jeffrey Horbar
|
|
Affiliation(s):
|
University of Vermont and Vermont Oxford Network and Vermont Oxford Network and University of Vermont
|
|
Address:
|
16 Colchester Ave, Burlington, Vermont, 05401, USA
|
|
Keywords:
|
Cluster randomization ; Survival analysis
|
|
Abstract:
|
The logrank test and weighted logrank test are used to compare survival distributions between two or more groups. A common strategy is to augment the unweighted logrank test with tests based on several weights so that differences in survival distributions that occur early or late can be detected with adequate power. However, when there are correlations among survival times, the tests are not valid. Correlated survival data occur when individuals are grouped into clusters or several events are measured on the same individual. In this talk, we present an extension of weighted logrank tests for the Tarone-Ware and Fleming-Harrington weights that are appropriate for correlated survival data when the number of clusters is large and the number of individuals within clusters is relatively small. The validity of the test does not require specific assumptions on the correlation structure. Simulation results indicate that the test is valid in most practical situations. The methodology has been implemented in a computer program written in SAS.
|
