A model for excess mortality of cancer patients when the life table providing background mortality does not depend on a demographic variable of interest

Net survival, the survival probability that would be observe where the cancer of interest would be the only possible cause of death, is a key indicator in population-based cancer studies. Accounting for mortality due to others causes, it allows cross countries comparisons or trends analysis and provides a useful indicator for public health decision making. We focus on the modelling method which is based on the decomposition of the all-cause mortality (which is observed) into the mortality due to cancer, the excess mortality, and the mortality due to other causes, the latter being assumed to be reflected by the background mortality derived from population life tables. However, if the available life tables do not provide mortality rates according to certain demographic variables (e.g. socio-professional category, socioeconomic), the estimates of the effects on the excess hazard of both this variable and, to a lesser extent, of the other covariables are biased. To deal with the impossibility to construct life tables adjusted for additional demographic variables, we propose to adjust for such an additional variable directly into the excess hazard model. We will present the context and the concept of net survival and excess mortality, and the impact of additional variable on background mortality. Then, we will describe our proposed model and some results of simulation study assessing its performances. An illustration will be presented using a real population-based dataset.