ABSTRACT
This chapter focuses on time-to-event data used as outcome measures, particularly in Phase III randomized trials. The Bayesian approach to survival analysis is well developed and very flexible, with model fitting based on Markov Chain Monte Carlo methods available in both specialist Bayesian and general statistical packages. In the context of survival analysis, and more generally in non-linear models, multiple imputation is not straightforward but such methods are naturally catered for within the simulation-based approach we adopted. The chapter discusses the Bayesian model for clustered time to event data using a trial of laser coagulation as a treatment to delay diabetic retinopathy. The limited space allocated to each contribution to a clinical journal means that adequate reporting of the full model and sensitivity analyses in the main trial report is difficult. Frequentist approaches can be divided into moment-based approaches and likelihood-based approaches.
