Cancer forecasting in New Zealand

Abstract

In 2002 Public Health Intelligence (PHI), a group in the Public Health Directorate of the Ministry of Health, produced Cancer in New Zealand: Trends and Projections (Ministry of Health 2002), which detailed the current rates and numbers of cases of cancer incidence and mortality in New Zealand, and gave projections for the same for the next 15 years. This publication included a comparison of rates in recent years between the Māori and non-Māori populations with those in the total population, and across socio-economic groups (defined by the NZDep96), and provided a breakdown of the various demographic drivers of change for the changing numbers of cases over the years. This thesis reviews the recent literature, giving particular emphasis to previous comparisons of alternate methods, records the methodologies used by the PHI group to prepare their report, and gives a detailed comparison of the various age-period-cohort models considered for use to provide the forecasts. These methods include generalised linear models, where age, period and cohort are fitted as factors, the number of cases is assumed to have a Poisson distribution and the log link function is used; models fitted, using two different software packages, in a Bayesian paradigm, with autoregressive priors for the three time measures; general additive models, where smoothing splines are used to model two of the time measures; local maximum likelihood models; and a model that is linear in time but not linear in the parameters, and in which the identity, not log, link function is used. Detail is given of the technique used to represent the relative contribution of changes in population size and age structure on the number of cases observed and projected, and of the possible effect on the mortality-to-incidence ratio of over- or under-estimation of the incidence rate.