Improving risk characterizations based on time to response

Abstract

One of the important aspects of quantitative cancer risk assessment is to model the frequency of a carcinogenic response as a function of exposure. Most of the current time-to-response modeling has been done using the "simple" model which does not include the effect of competing risks. As a result, the probabilities which are stated in terms of these models do not correspond to real world occurrences. Rather they correspond to a fictional world where no competing risks exist; that is, a world where only the risk being modeled is present. The cause-specific family of models is preferred since it incorporates the effect of competing risks. A method is presented for utilizing life table data and existing simple model maximum likelihood estimates to obtain cause-specific risk characterizations. Representations for various risk characterizations (e.g., time-to-response probability, mean response free period, mean free dose, and virtually safe dose) under both models are compared. The results show that the simple model always overstates the effect of the carcinogen. The overstatement can easily be in the 20%-200% range. Two alternatives are introduced for modeling time to death from tumor when the cause of death is uncertain. One alternative is to model the time to death irrespective of cause. The other alternative focuses on the increase in the hazard rate due to the presence of a tumor. Thus, deaths caused by the increase in hazard rates are considered as being related to the introduction of the carcinogen. Both alternatives give risk characterizations which may be more relevant to cancer risk assessment.