The long-term objective of our research is to develop statistical framework that will be used to translate the research findings to cancer prevention, diagnosis and management practices. The specific objective of this proposal is to develop a general framework for constructing risk prediction models (RPMs) for breast cancer that incorporates both environmental and genetic factors as well as family history, in the absence of genetic markers. The proposed framework can be used to develop RPMs for other cancers and possible other chronic diseases. This proposal has four specific aims: 1) to develop a general framework for constructing RPMs by incorporating known risk factors such as candidate genes, other biological markers, demographic variables, birth history, diet, history of medications, medical history and other lifestyle variables, which can be either time-independent or time-dependent; 2) to develop general framework for constructing RPMs by incorporating, (in addition to known risk factors described above), family history data as well as family data collected from family members; 3) to outline protocols on how to use such programs and to develop computer programs for all these new methodologies with a """"""""user friendly"""""""" interface, which will be disseminated to the scientific community via the Internet; 4) to develop RPMs for breast cancer based on the data sets collected in the Breast Cancer Detection Demonstration Program and Cancer Steroid Hormone study (among other available data sets in the Fred Hutchinson Cancer Research Center), to compare the RPMs that are constructed on different dat a sets and to validate them across study populations. Further, we will attempt to validate the refined RPM for breast cancer via ongoing studies in the Center. This development is rooted in recent advances made in statistics, epidemiology, genetics and genetic epidemiology. Since most of the risk factors are varying with age and outcomes are generally ages of onset, the proportional hazard model has been generalized to include the birth history and two-stage models, and is chosen as a basic model for constructing RPMs.
