Bayesian Survival analysis using a MARS Model

this paper we extend this Bayesian MARS model to cope with survival data in a similar way that the HARE model is an extension to the usual MARS methodology. Further, we generalise the methodology for multivariate survival data where the survival times are generated from the same group, person etc. These are known as frailty models and assign a random effect for each group from which the data arises. This random effect term is assumed to be due to the variability associated with unobserved covariates. These models are a generalisation of the normal proportional hazards model (Clayton and Cuzick, 1985) but we shall introduce them as an extension of the usual Bayesian model for survival data. In the next section we introduce the Bayesian MARS model we use for survival analysis. Section 3 gives examples of the Bayesian model, first on simulated and then on real survival data, whilst Section 4 outlines the extension to frailty datasets and gives a demonstration of the model on such a dataset. A brief discussion is contained in Section 5