Projecting cancer incidence and mortality using Bayesian age-period-cohort models

Abstract

Background: We present a practical application of an age-period-cohort model in a Bayesian frame-work for making cancer-burden projections.

Methods: Second degree autoregressive smoothing was used on the age, period and cohort effects for estimating future incidence and mortality.

Results: We are able to demonstrate the feasibility, flexibility and strengths of this approach. Compared with previously used methods, it performed better for providing point estimates when past trends continued into the future. However, the extremely wide credible intervals need careful interpretation.

Discussion: Part of the uncertainty is attributable to the possible inadequacy of the model and not necessarily relevant in the prediction of what would happen if the present trends continue into the future.