Each year, a significant number of women in the United States are diagnosed with breast cancer. The ability to identify women at high risk for developing breast cancer as early as is possible would be invaluable for monitoring and management of this disease. As more information becomes available about the roles that different genetic, environmental, and personal health status factors play in determining breast cancer risk, it is important to develop methods for individualized breast cancer risk prediction that incorporate these factors. The goal of this application is to develop methods for individualized breast cancer risk prediction, using Bayesian networks with time dependencies, for women between the ages of 20 and 70 years. Bayesian networks provide methods for reasoning under conditions of uncertainty based on artificial intelligence and statistical principles. They allow for inclusion of expert opinion and empirical results from studies for a condition of interest. Specifically, we propose to (1) develop a Bayesian network based risk assessment tool for women aged 20 to 70 years, using knowledge provided by a panel of experts involved in cancer assessment and treatment research as well as relevant probabilities gleaned from the breast cancer literature; (2) perform a preliminary assessment of the accuracy of the risk assessment tool developed, using data obtained from 40 patients followed for at least 5 years through a high risk screening facility (the Yale Cancer Center Genetic Counseling Shared Resource); and (3) identify the steps required for a more formative validation of the risk assessment tool on a larger study population, and examine how the tool can be integrated with other quantitative methods designed for breast cancer risk prediction, such as Markov models.
