Projection of mesothelioma mortality in Britain using Bayesian methods

Mesothelioma is a cancer that mainly affects the pleura (the membrane that covers the lungs and lines the internal chest wall) and the peritoneum (the membrane that forms the lining of the abdominal cavity). Of all deaths, 85% have been among men and the majority of these were caused by occupational exposure to asbestos fibres (Rake et al, 2009). The disease has a long latency period; symptoms usually emerge between 15 and 60 years after exposure to asbestos, after which mesothelioma is rapidly fatal. The majority of deaths occur among those >60 years of age, with few deaths occurring among those 1% of all cancer deaths. Predictions of mesothelioma mortality have been made in several countries. Clements et al (2007) modelled mesothelioma mortality in Australia using both an age–birth cohort model and a model based on that introduced by Hodgson et al (2005). Banaei et al (2000) predicted mortality in France using a method based on a risk function that links mortality with past exposure to asbestos. Segura et al (2003) used an age–period–cohort model to predict mortality in the Netherlands. Hodgson et al (2005) developed a model based on the estimated collective population exposure to asbestos and a specific form for the relationship between mesothelioma risk and time since first exposure at the population level. The model was fitted to mesothelioma mortality for Great Britain up to 2001 and projections of mortality were made by applying fitted mesothelioma rates to future population projections. Confidence intervals for the parameters and prediction intervals for future estimated annual deaths could not be made because of limitations in the optimisation approach adopted. This paper presents the results of a Bayesian statistical analysis to refit a modified version of this model using updated mortality data up to 2006 and a Markov Chain Monte Carlo (MCMC) method of computation. The Bayesian approach allowed credible and prediction intervals to be calculated; thus, informed statistical inferences about model parameters and predictions of future mortality, in particular the scale and timing of the peak in deaths, could be made.