Multistage prediction of cancer in serially dosed animals with application to the ED01 study.

The ED01 study is a large-scale experiment in which mice were exposed to the known carcinogen 2-acetylaminofluorene (2-AAF). The dosing was continuous until time of death for most of the mice. However, for some mice, dosing was terminated at specified intervals prior to sacrifice time. Exploratory model fitting of the continuous dosing data of sacrificed mice using liver neoplasms as the endpoint was conducted by Brown and Hoel (1982). They report that the multistage model of Armitage and Doll (1954) fits the data moderately well, however a much improved fit with fewer parameters is attainable by a modified multistage model. These two models, with parameter estimates obtained from the data on continuous dosing, are referred to in this paper as Model 1 and Model 2, respectively. Models 1 and 2 are extended to provide for termination of dosing prior to time of sacrifice. The resultant equations are used to predict cancer responses in the serial dosing part of the ED01 study, and then are compared to the actual observed tumor frequencies. The objective is to evaluate the change in cancer risk over time after dosing is terminated, and to establish which stages in the multistage process are affected by dose. The predictive capabilities of the two models are compared and contrasted. Model 1, which presumes four stages of which two are affected by dose, predicts only moderately well and is inconclusive as to which two of the stages are the ones affected. Model 2, which differs from the multistage model by using a "J-shaped" curve instead of a polynomial for the dose metameter, provides an improvement over Model 1.(ABSTRACT TRUNCATED AT 250 WORDS)