A0713
Title: Robust Bayesian inference for accelerated failure time models with skewed and heavy-tailed survival data
Authors: Mehrdad Naderi - Northumbria University at Newcastle (United Kingdom) [presenting]
Abstract: Censored data analysis has been commonly used in clinical studies, where researchers often encounter limitations in measuring instruments and/or experimental design. Despite the growing adoption of survival statistical methods for analyzing censored data, the accurate specification of these models requires great care, especially in dealing with atypical observations displaying non-normal features. In such cases, models based on normality may fail to capture the underlying pattern adequately. Accelerated failure time modeling is proposed for survival-censored data in which the random errors conform to the normal mean-variance mixture (NMVM) distribution. The NMVM represents a diverse class of distributions offering enhanced flexibility in analyzing skewed and heavy-tailed data. The focus lies on Bayesian inference for model parameters, employing an efficient Markov chain Monte Carlo (MCMC) algorithm. A series of simulation studies are conducted to assess the effectiveness and robustness of the proposed methods in various scenarios of skew data with heavy tails. The methodology is then examined by assessing the relationship between covariates and survival time on real-world data. The results show that models with a skew and/or fat-tail distribution not only exhibit a superior fit to the data but also provide robust inference compared to the normally-based model.
