The measures of cancer prognosis most commonly reported by cancer registries are 5- or 10-year absolute or relative survival rates. Because cancer survival rates often vary by the age of cancer patients, and because the age structure of cancer patients often varies between populations, age adjustment is crucial for comparative analyses of cancer survival rates. However, traditional age adjustment often breaks down for long-term survival rates, particularly for sparse data, and it may provide inconsistent results for relative survival rates, even if age adjustment is made to the study population's own age distribution. In this manuscript, we propose an alternative approach to age adjustment of both absolute and relative survival rates to overcome both the practical and conceptual problems inherent in traditional age adjustment. We outline the computational realisation of this approach, and we give an empirical illustration of its application using data from the nationwide Finnish Cancer Registry.