Statistical Inferences about the Mechanism of Action in Carcinogenicity Studies

An innovative approach to dose-response modeling provides statistical insight into the relative likelihood of different mechanisms of action in cancer dose-response studies. Two illustrative examples are given based on time-to-tumor data on mammary fibroadenoma and adenocarcinoma in female Sprague-Dawley rats using 34 different dose metrics. The likelihood for the study outcome was calculated for each dose metric and compared with the background likelihood using a likelihood-ratio test. In the first example, fibroadenomas were strongly related to the presence or absence of mammary secretory activity, galactoceles, pituitary tumors, and abnormal diestrous days in weeks 1 to 26. Adenocarcinomas were the most strongly related to the number and percentage of abnormal estrous days. In these examples, the usual dose metric based on the dietary concentration of the pesticide had some explanatory ability but not nearly as much as the dose metrics more directly related to hormonal mechanisms of action.