413 – Contributed Oral Poster Presentations: Section for Statistical Programmers and Analysts
Assessing the Performance of the Gail's Breast Cancer Risk Prediction Model
Vicky W. Li
Beth Israel Deaconess Medical Center
Mara A. Schonberg
Beth Israel Deaconess Medical Center
Long H. Ngo
Harvard Medical School
Background: The Gail model is the most widely available breast cancer risk prediction model. However, it has not been recently validated among postmenopausal women, and has never been validated among women aged 75 and older. Methods: We assessed performance of the Gail model in breast cancer prediction among postmenopausal women in a random selection of 20% (n=18,946) of Nurses' Health Study (NHS) participants from 2004 to 2009. The NHS sample was on average older (youngest women aged 57 at start of follow up) than the Breast Cancer Detection and Demonstration Project (BCDDP), the sample used in development of the Gail model. Results: The Gail model was found to have c-statistics of 0.61 in NHS women ages 57 to 64, 0.55 in women 65 to 74, and 0.63 in women 75 and older. Calibration was assessed through expected over observed (E/O) ratio of breast cancer cases by age (1.6 in women 57-64; 2.2 in 65-74; and 2.5 in 75 and older). Conclusion: We found the Gail model had poor discrimination and over-predicted breast cancer among postmenopausal women. We also found different strengths of association (relative risks) between Gail risk factors and breast cancer. We plan to publish a complete evaluation of the performance of the Gail model using additional cohort data in a clinical manuscript.