Cancer Models and Cancer Genetics

There have been many attempts to develop stochastic multistage models for cancer. The models relate hypothesized biological processes occurring at the cellular level to the occurrence of tumors at the experimental or epidemiologic level of individuals in a population. The existing models fit a variety of data, but none is fully satisfactory and they are in some ways inconsistent with each other. Recently, substantial new data have become available on the nature of cancer-associated mutations observed directly at the cellular level. These data suggest that the number of stages may be greater and more variable among individual tumors of the same organ than has been thought. There may be many pathways to cancer, and the mutations responsible may not constitute a fixed set or sequence. This pattern resembles the genetics of quantitative rather than qualitative traits, and may also be consistent with the variable histology and behavior of tumors of a given organ. Simulations using such models suggest that cancer in the general population may have such heterogeneous etiology, a possibility that has important implications for screening, risk projection, and prevention. Risk-generating processes of a rather generic kind may generate similar hazard functions for diverse chronic diseases in the age ranges often used in epidemiologic studies. This phenomenon raises questions about the purpose and interpretation of statistical epidemiologic models.