Application of Markov chain Monte Carlo methods to projecting cancer incidence and mortality

Projections based on incidence and mortality data collected by cancer registries are important for estimating current rates in the short term, and public health planning in the longer term. Classical approaches are dependent on questionable parametric assumptions. We implement a Bayesian age-period-cohort model, allowing the inclusion of prior belief concerning the smoothness of the parameters. The model is described by a directed acyclic graph. Computations are carried out by using Markov chain Monte Carlo methods (implemented in BUGS) in which the degree of smoothing is learnt from the data. Results and convergence diagnostics are discussed for an exemplary data set. We then compare the Bayesian projections with other methods in a range of situations to demonstrate its flexibility and robustness.