Abstract
[PURPOSE] We examined aspects of the multistage frailty models for carcinogenesis. [METHODS] We see that the generalized Armitage-Doll model, which can examine the age-, dose-, and age-at-exposure- dependent effects of external exposure to a carcinogen, do not, because of various unobserved factors, account for population heterogeneity. Thus we developed multistage frailty models that assume that population heterogeneity may come from either different sensitivities to exposure or different background cancer risks, and that frailty distributions are gamma and inverse-Gaussian distributions. Using these models, we demonstrated how population heterogeneity affects estimates of model parameters, particularly excess risks of cancer. [EXAMPLE] We reanalyze a solid cancer incidence data set of atomic bomb survivors exposed before they were 40 years old, and we assess the magnitude of population heterogeneity, the excess absolute and relative risks of solid cancers from exposure to a 100-millisievert colon dose, and the goodness of fit of the models. [CONCLUSION] We see that the multistage frailty models are useful for understanding the time course of carcinogenesis with effects of exposure and unobserved heterogeneity factors in epidemiologic data. [J Radiat Res 44:388 (2003)]
