A0636
Title: Hierarchical penalized distributional regression models for survival data
Authors: Fatima-Zahra Jaouimaa - University of Limerick (Ireland)
Il Do Ha - Pukyong National University (Korea, South)
Kevin Burke - University of Limerick (Ireland) [presenting]
Abstract: A distributional regression approach is taken to analyse survival data, whereby explanatory variables can enter the hazard regression model through its scale and shape parameters; this enables flexible modelling beyond proportional hazards. A penalised hierarchical likelihood estimation approach is adopted to facilitate automatic variable selection and account for hierarchical data structures (e.g. clustered clinical trials). This very general procedure applies an adaptive lasso to both the scale and shape hazard parameters while incorporating correlated bivariate frailty in both parameters. The estimation and inferential performance of the proposal are investigated using simulation studies; the method on a real data example is demonstrated.
