|
Activity Number:
|
469
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #308293 |
|
Title:
|
Assessing Fit in Cure-Mixture Models
|
|
Author(s):
|
E. Paul Wileyto*+ and Daniel F. Heitjan
|
|
Companies:
|
University of Pennsylvania and University of Pennsylvania
|
|
Address:
|
Tobacco Use Research Center, Psychiatry, Philadelphia, PA, 19104-3309,
|
|
Keywords:
|
residuals ; cure models ; acellerated failure time ; proportional hazards ; survival ; behavioral medicine
|
|
Abstract:
|
Cure-models fit event times for a fraction of subjects truly at risk with a parametric survival model, and membership in the cured fraction using logistic regression. We have modified the Schoenfeld residual (from Cox regression) to assess adequacy of the parametric cure model for effects of covariates. The original residual is the difference between observed and expected covariate values for subjects failing at each event time; expectation is calculated by weighting covariate values in the risk set by proportional hazards (PH). Our alternative weights use the full hazard in place of PH, and account for the probability of membership in the cured class. We tested performance by simulation. The residuals assess fit well in both PH and accelerated failure-time models, and detect misspecification. We illustrate using relapse times from a smoking cessation study. Support: NCI/NIDA P5084718.
|
