|
Activity Number:
|
374
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 4, 2009 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #305167 |
|
Title:
|
Bayesian Hazard Rate Estimation and Sufficient Dimension Reduction
|
|
Author(s):
|
Shraddha S. Mehta*+ and Surya T. Tokdar and Jayanta K. Ghosh and Bruce A. Craig
|
|
Companies:
|
Purdue University and Carnegie Mellon University and Purdue University and Purdue University
|
|
Address:
|
250 North University Street, West Lafayette, IN, 47906,
|
|
Keywords:
|
Bayesian ; Survival analysis ; Hazard rate ; Sufficient dimension reduction ; nonparametric
|
|
Abstract:
|
Logistic Gaussian process priors have been used in nonparametric Bayesian density estimation and sufficient dimension reduction in a regression setting. In this talk, I describe the extension of this approach to survival modeling with right-censored data. Our method simultaneously estimates the covariate subspace and the hazard rate given this subspace. This method overcomes the dimensionality problem and the inability to estimate non-linear covariate effects when using the traditional proportional hazards model. The hazard rate is estimated using both a partial likelihood approach and the full-likelihood. Preliminary simulation studies comparing this method, random survival forests, and Cox's proportional hazards model using partial likelihood shows that when the proportional hazards assumption is not valid, both our method and random survival forests perform better than Cox's method.
|
