Modelling Time-Varying Hazards and Covariate Effects

Authors: 
Mike West
Duke University

Nov 30 1990

Models for survival analysis in the presence of possibly time-varying covariate effects are constructed as piecewise proportional hazards models in which the regression parameters changing according the random walk mechanisms, modeling the prior view that parameter vales in consecutive intervals of the time axis are more likely to be close than those well separated in time. analysis uses Bayesian updating to approximate posterior distributions for such time-varying parameters, providing estimates of their values over time - the parameters 'trajectories', and also approximations to posterior predictive survival distributions for further cases. Ill Illustration is provided here in analyses of nasopharynx cancer survival data.

Manuscript: 

PDF icon 1991-09.pdf