Using the SEER database, which covers more than 10% of the US population, lung cancer incidence trends over the period 1973-1998 will be investigated using extensions of age-cohort-period models. Specifically, it is proposed to develop models in which non-specific age effects are replaced by parametric hazard functions that acknowledge the multistage nature of carcinogenesis. At least formally, this procedure finesses the well known problem of arbitrary linear trends that plagues the traditional age-cohort-period models. Furthermore, analyses of carefully collected incidence data in large registries such as SEER using biologically-based modelscan lead to insights in to the mechanisms underlying carcinogenesis in addition to shedding light on temporal trends. With respect to temporal trends, attempts will be made to relate trends in lung cancer incidence to trends in smoking habits in the US. Tobacco smoke contains both mutagens and agents that impact cell proliferation kinetics. Thus, tobacco smoke probably acts as both an initiator and a promoter in lung carcinogenesis. One goal of these analyses is to investigate whether non-specific and birth cohort and calendar year trends can be directly linked to specific aspects of carcinogenesis, such as initiation, promotion and progression. This possibility will be investigated by incorporating tobacco consumption trends into the parameters of the multistage model. Because of the large size of the SEER database it should be possible to conduct these analyses separately by race, sex and histologic type. Finally, the estimated parameters of the optimal multistage model will be used to project lung cancer incidence trends into the future under various assumptions regarding changes in composition of cigarettes and smoking habits.
