Improved breast cancer risk prediction is of critical importance for African American (AA) women in view of their younger age at diagnosis, higher incidence of the most aggressive subtypes (e.g., estrogen receptor negative (ER-) breast cancer), and 40% higher breast cancer mortality compared with white women. Several models, including the well-known Breast Cancer Risk Assessment Tool (Gail model), have been developed, largely in white populations, to assess a woman?s absolute risk of breast cancer; they have been used to identify high-risk women for supplemental screening, preventive treatment, and enrollment in chemoprevention trials. Only two prediction models have been developed specifically for AA women and both have low discriminatory accuracy. Recent research by our group and others indicates distinct etiologic pathways for breast cancer subtypes defined by ER status in that associations with parity, breastfeeding, postmenopausal hormone use, and body size differ by ER status. For example, high parity is associated with reduced risk of ER+ cancer and with increased risk of ER- cancer. The relatively poor discriminatory accuracy of breast cancer risk prediction models in AA women may reflect the failure to properly consider tumor subtypes. This is a lesser concern in other ethnic groups in which the vast majority of cases are ER+, whereas up to a third of AA cases are ER-. In a novel approach, we will first estimate ER specific relative risk estimates from analyses of pooled data from AA women in three population-based case-control studies, including 1382 ER- and 2275 ER+ breast cancer cases, as well as 3341 controls. We will then use those estimates, together with SEER age-incidence rates for ER+ and ER- breast cancer in AA women to estimate baseline age-specific hazard rates for ER+ and ER- cancer. Finally, we will combine relative risks and baseline hazards, taking into account competing risks, to estimate the probability of developing the first of either ER+ or ER- breast cancer over a pre-specified time interval given a woman?s age and risk factors. Performance of the ER specific models and the overall tool for predicting any breast cancer will be tested in prospective cohort data from the Black Women?s Health Study (BWHS), based on occurrence of 703 ER- and 1502 ER+ cases. Existing risk prediction models for AA women will also be applied to the prospective BWHS data in order to compare their performance with performance of our new tool. Although genome-wide genotyping is not yet a part of each patient?s medical records, to set the stage for such genotyping in the future, in a second aim we will add SNPs associated with breast cancer subtypes identified in GWAS and fine-mapping of AA women to the models and evaluate changes in performance. Improved breast cancer risk prediction models in AA women will lead to earlier detection and treatment of high risk women, and thereby to reduced breast cancer mortality.

Public Health Relevance

This novel approach to breast cancer risk prediction for AA women will produce a tool that will be immediately useful for personalized risk prediction in young women who have never had a mammogram, so that those at high risk can be referred for breast cancer screening before reaching current guideline-recommended ages. It will also meet a critical need for older women, given that current models, even models that include breast density measures, perform poorly in AA women. Finally, the risk models developed will be useful for determining eligibility for enrollment in cancer prevention trials and for decision-making on whether to take known chemopreventives such as Tamoxifen.

National Institute of Health (NIH)
National Cancer Institute (NCI)
Research Project (R01)
Project #
Application #
Study Section
Cancer, Heart, and Sleep Epidemiology B Study Section (CHSB)
Program Officer
Nelson, Stefanie A
Project Start
Project End
Budget Start
Budget End
Support Year
Fiscal Year
Total Cost
Indirect Cost
Boston University
Public Health & Prev Medicine
Organized Research Units
United States
Zip Code