Surviving fully Bayesian nonparametric regression models
Timothy Hanson, Alejandro Jara · Oxford University Press eBooks · 2013
We discussed, compared and illustrated flexible nonparametric models that can be used to introduce categorical and continuous covariates in the context of time–to–event data. The models correspond to generalizations of accelerated failure time models, based on dependent extensions of Dirichlet processes and Polya tree priors. Important advantages of the induced survival regression models include ease of interpretability and computational tractability. Furthermore, an important property of the proposed models is that the complete distribution of survival times is allowed o change with values of the predictors instead of just one or two characteristics, as implied for many commonly used survival models. The two extensions are compared by means of real–life data analyses.