Precision of cancer incidence predictions based on Poisson distributed observations

In this study, a method is proposed for calculating approximate confidence and prediction intervals both for the numbers of cases and for the age-adjusted incidence rates by assuming Poisson or extra-Poisson distribution of stratum-specific number of incident cases, the assumption commonly accepted and used in epidemiology. The method can be applied to predict incidence cases based on a model with any functional form of linear predictor or link function. A good prediction model, taking into account the availability of reliable data, should produce a prediction close to the future true observation with as high precision as possible. The models seem to represent well the main features of historical data and most of them preserve the age-specific incidence patterns of historical data in the period of prediction.