|
Activity Number:
|
15
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Sunday, July 29, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Biopharmaceutical Section
|
| Abstract - #309302 |
|
Title:
|
Extensions of Cure Models for Clustered Time-to-Event Data
|
|
Author(s):
|
Jeremy Taylor*+ and Yingwei Peng
|
|
Companies:
|
University of Michigan and Queen's University
|
|
Address:
|
1420 Washington Heights, Ann Arbor, MI, 48109,
|
|
Keywords:
|
cure models ; clustered data
|
|
Abstract:
|
Cure models are attractive approaches in survival analysis when there is a rationale for a non-susceptible group and there is good follow-up beyond the typical event times. The standard mixture cure model has a logistic model for long term incidence and a Cox model for latency amongst the susceptible group. We extend this model to the situation of clustered observations. In a conditional approach random effects are introduced in both the incidence model and the latency model to incorporate the cluster effects. In a marginal approach the variance-covariance of the estimates is obtained from a sandwich estimator. For both methods estimation is performed by extensions of the EM algorithm. Simulation studies are performed. The methods are applied to a multi-institutional study of local recurrence of tonsil cancer patients who received radiation therapy.
|
