Fitting cancer mortality data with cumulative damage models

Cumulative damage models conceive the epidemiologically observed aspects of carcinogenesis as some kind of total balance over a complex biological process and suggest that this total balance might behave as a wear-and-tear process. The essential concepts of this mechanistic model are exposure to a carcinogenically damaging environment and resistance of a host system against those damages. Intensity of exposure and magnitude of host resistance are the parameters to be assessed. The paper describes (1) the statistical methods for fitting this model to birth cohort data; (2) for which cancer sites the model provides acceptable fits and for which it does not; and (3) how model extensions provide improvements in the goodness of fit. It is shown that from a theoretical viewpoint the consideration of extra-Poisson variation is needed for descriptive epidemiological applications. The practical examples indicate that the present version of the model provides acceptable fits for only a few cancer sites and that refinements are needed in the majority of sites. However, plausible model extensions suggest considerable improvements of goodness of fit.