The purpose of this project is the development of biostatistical methods appropriate for epidemiological and experimental studies related to cancer prevention and control. Poisson regression methods are often used to analyze the relationship of a county's cancer mortality to its demographic, economic, and ecologic characteristics. Because of variability of cancer risk among members of the county's population, the Poisson assumption is theoretically invalid. This research will evaluate the actual degree of intra-county temporal heterogeneity for the major cancer sites. The proper time metric to be used when applying Cox regression techniques to analyze a cohort's cancer incidence over time is being studied by simulation methods. Data are being generated according to known cancer age-specific incidence and known mortality from all causes. The time metrics being studied are follow-up time from study initiation and age at risk. Different study durations are also being evaluated. In addition, different methods to properly estimate the effect of risk factors which are related to age at ascertainment are being evaluated. The assumption of """"""""proportional hazards"""""""" is commonly made when analyzing survival in patients treated for cancer. Only the scale parameter in parametric models and the proportionality parameter in semi-parametric models are assumed to vary among patients through their prognostic variables. Because extraneous hazard rate heterogeneity can produce substantial deviations from this assumption, this research will compare the fits of a Weibull model with a Gamma-mixture of Weibull models to estimate the degree of heterogeneity and its biasing effect on estimated parameters and on the hazard ratio.
