|
Activity Number:
|
193
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 4, 2008 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #301283 |
|
Title:
|
A Bayesian Cox-Regression Model with Informative Censoring
|
|
Author(s):
|
Niko Kaciroti*+
|
|
Companies:
|
The University of Michigan
|
|
Address:
|
300 N. Ingalls bldg., Ann Arbor, MI, 48109,
|
|
Keywords:
|
Time-to-event analysis ; Missing data ; Gibbs sampling ; Hypertension ; Randomized clinical trials
|
|
Abstract:
|
In randomized clinical trials the primary outcome of interest often is time to the occurrence of an event. Cox-Regression models are commonly used to analyze such type of data. The standard Cox model assumes that the censored data are non-informative. These assumptions are usually not testable from the observed data and may well introduce biased. This paper presents a method for analyzing time to event data using Cox-Regression when censoring is informative. We propose a Bayesian model to analyze this type of data by introducing informative prior distributions to identify the model. Sensitivity analysis is then used over a range of these prior distributions. The method is applied to analyze the data from the Trial of Preventive Hypertension Study.
|
