Breast cancer is a very heterogeneous disease. Some patients are cured by local therapy and survive for many years, while others experience an early recurrence of their disease followed shortly by death. In order to make the most appropriate treatment decisions, the patient and her oncologist must predict the clinical outcome of her particular breast cancer. Traditionally, several easily measured factors (e.g. axillary nodal status, tumor size, estrogen and progesterone receptor status, etc.) have been used to make these predictions. However, a series of Consensus Conferences and NCI Clinical Alerts have cast doubt about the predictive ability of these traditional factors. More recently, a number of new factors have been identified as potential predictors of clinical outcome in patients with breast cancer. The initial evaluations of these new factors have often been limited by small numbers of patients and/or relatively short clinical followup. The new factors are frequently evaluated in a univariate setting, or together with only a small number of other prognostic factors. What is needed is a large population of breast cancer patients with long-term clinical followup and accessible tumor specimens on which a complete panel of potential prognostic factors can be determined. Multivariate modeling techniques could then be used to integrate these factors and produce prognostic indices, or a set of classification schemes, that can be used to predict disease recurrence and mortality for an individual patient with breast cancer. This project will continue the development of a large bank of breast tumors, and will measure a panel of potential prognostic factors on tumors in this bank. The factors will include: traditional factors, factors from our earlier pilot studies, measures of proliferation, oncoproteins and growth factors, tumor suppressor genes, invasion-related factors, and promising markers from other projects in this Program Project. We will fully explore the practical application of three multivariate methods of analyzing potential prognostic factors: Cox modeling, recursive partitioning, and neural networks. It is anticipated that the best features of these different analytical approaches can be combined to develop a powerful, practical strategy for integrating potential prognostic factors. The overall objective is to construct and validate a set of clinically relevant prognostic classification schemes for patients with primary breast cancer, and thus to optimize individual treatment decisions.
Showing the most recent 10 out of 260 publications
