Forecasting causes of death by using compositional data analysis: the case of cancer deaths

Cause‐specific mortality forecasting is often based on predicting cause‐specific death rates independently. Only a few methods have been suggested that incorporate dependence between causes. An attractive alternative is to model and forecast cause‐specific death distributions, rather than mortality rates, as dependence between the causes can be incorporated directly. We follow this idea and propose two new models which extend the current research on mortality forecasting using death distributions. We find that adding age, time and cause‐specific weights and decomposing both joint and individual variation between different causes of death increased the forecast accuracy of cancer deaths by using data for French and Dutch populations.