A0406
Title: Bayesian variable selection for survival data: Theory, methods, software and applications
Authors: Francisco Javier Rubio - University College London (United Kingdom) [presenting]
Abstract: Survival analysis is one of the main branches of Statistics, with applications in medicine, biology, and engineering, to name but a very few. Thanks to recent data linkage and data collection developments, survival data can be enriched with additional individual characteristics, such as sociodemographic, clinical, and genetic information. Thus, it is of interest to select the variables that explain the survival times (i.e. the variables that are associated with the survival times). Two of the most popular models for analysing survival data will be presented: the proportional hazards model and the accelerated failure time model. After that, we will present an overview of different methods for conducting variable selection using these models, including penalised likelihood methods, and Bayesian methods based on spike-and-slab, local, and non-local priors. Then, we will focus the attention on the analysis of the theoretical/asymptotic properties of the methodology based on local and non-local priors, including a discussion of these properties when the model is misspecified (the most common case in practice). Finally, we will illustrate the use of available R packages for conducting variable selection based on these methods and models using simulated and real data. R code will be provided.
