Different multiple regression models for estimating survival: use in a population-based series of colorectal cancers

Two additive models for estimating relative survival, recently available in statistical packages, were compared to the Cox model in order to define practical criteria for choosing one of these three models. The three models were applied to a series of 1062 colorectal cancer incident cases recorded at the "Registre des Tumeurs Digestives de Côte d'Or" to study the impact on survival of age, sex and socio-economic status. The respective advantages of each method and facilities of each program were identified. This study emphasized the importance of multiple regression models for estimating relative survival: from the epidemiologist's point of view, they reinforce the validity of international comparisons of survival statistics, from the clinician's point of view, they permit to identify the real prognostic significance of variables related to life expectancy in the general population.