On the generalization of a time-to-response cancer risk assessment model

Abstract

This dissertation generalizes the hazard rate in the Hartley-Sielken model from a product form in dose and time to a nonproduct form. Also a latency period (the time from onset of cancer until it is clinically detectable) is considered. The likelihood function of the nonproduct model is concave when the latency period is not present. Two measures of goodness-of-fit are presented for situations in which the time to the tumor or similar response is not observable. When the time to the specified response is observable the goodness-of-fit statistics in Akritas (1985) can be applied. Cancer risk characterizations such as Virtually Safe Dose, Late Risk Dose, and Mean Free Period can be obtained using the nonproduct model. A computer program was developed to facilitate the use of the generalized model in a personal computer environment. A small simulation experiment was performed. In that experiment a better fit did not necessarily imply better cancer risk characterizations. Also, the true and estimated frequencies were closer together than the observed and estimated frequencies. The nonproduct model fit the experimental data on the presence of liver carcinoma in sacrificed mice in the ED[01] study better than the product model did.