The practice of medicine involves the science of prediction. Prediction depends on clinical or laboratory variables or factors that are linked to outcome. The most common predictors in cancer medicine are the three variables: tumor size, regional lymph node status, and distant metastasis. The three variables are combined in a bin model to form the TNM Staging system, which is a major tool used to predict the outcome of cancer patients and guide therapy. However, the TNM system has only three variables. Therefore, its predictive accuracy is limited.
This research addresses issues of numerical computing and optimization in developing expanded cancer prognostic systems that can integrate multiple variables. Two sets of closely related tasks will be investigated, and they represent two major aspects of the intellectual merit of the study. One task is focused on how censored survival times and different types of variables are integrated into a clustering framework that works for a large volume of cancer patient data. Another task is to use the developed cluster analysis to establish prognostic systems, which will provide a more accurate prediction of outcome by taking multiple prognostic factors into account. The broader impact of this research includes many aspects. It will have a direct impact on future staging and classification of cancer patients. The work has general applications and can be adapted to studies of any non-cancer health problems. The investigator's study is expected to make significant contributions to advances in medicine. The research will also have an educational impact.