Breast Cancer Risk Prediction with Heterogeneous Risk: Profiles According to Breast Cancer Subtypes

• Activity Number: 46
• Type: Invited
• Date/Time: Sunday, July 31, 2011 : 4:00 PM to 5:50 PM
• Sponsor: ENAR
• Author(s): Bernard Rosner*+ and Rulla Tamimi and Graham Colditz and Susan Hankinson and Robert Glynn
• Companies: Channing Laboratory and Channing Laboratory and Washington University at St. Louis and Channing Laboratory and Brigham and Women's Hospital
• Keywords: Poisson regression ; cancer risk modelling

Abstract

It is well known that breast cancer risk profiles vary by characteristics of the tumor. For example, pregnancy is generally protective for ER+ breast cancer (the most common type), but is either null or deleterious for ER- breast cancer. Technological advances make it possible to characterize tumors by a number of characteristics simultaneously. However, if tumors are categorized by the joint classification of several tumor characteristics and specific categories of tumors are compared by risk factor status, sample size can be small and power correspondingly low. In this talk, we propose an enhancement of Poisson Regression analysis to assess marginal association of risk factors with specific tumor characteristics, while controlling for the levels of other tumor characteristics. The methods will be illustrated using prospective data from the Nurses' Health Study based on 3398 breast cancer cases occurring from 1980 to 2006.