Elizabeth S. Burnside MD, MPH, Presenter: Stockholder, Cellectar Biosciences, Inc Stockholder, NeuWave Medical Inc

Jie Liu, Abstract Co-Author: Nothing to Disclose

Charles David Page PhD, Abstract Co-Author: Nothing to Disclose

Catherine A McCarty PhD, Abstract Co-Author: Nothing to Disclose

Adedaoy A Onitilo MD,PhD, Abstract Co-Author: Nothing to Disclose

Peggy L. Peissig PhD, Abstract Co-Author: Nothing to Disclose

Terrie Kitchner, Abstract Co-Author: Nothing to Disclose

Amy Trentham-Dietz, Abstract Co-Author: Nothing to Disclose

Yirong Wu, Abstract Co-Author: Research Grant, Hologic, Inc

Ulrich Broeckel, Abstract Co-Author: Nothing to Disclose

Recent large-scale genome-wide association studies (GWAS) have identified new genetic variants that predict breast cancer. However the predictive ability of genetic variants compared to mammography (BI-RADS) features has not been evaluated. We conducted a retrospective case/control study to determine the predictive value of demographic risk factors (from the Gail model), germline genetic variants, and BI-RADS abnormality features in women undergoing image-guided breast biopsy.

We collected age-matched cases and controls from a population-based Personalized Medicine Research Project (PMRP), including women of Western European heritage with a plasma sample, a mammogram, and a breast biopsy within 12 months after the mammogram. We used Gail model risk factors from surveys and the EMR, mammographic findings according to BI-RADS extracted from free text reports, and 10 germline genetic variants (single nucleotide polymorphisms—SNPs). We built conditional logistic regression models to determine the predictive ability of single data types: 1) Gail, 2) SNPs and 3) BI-RADS as well as combined data types: 1) Gail + SNPs, 2) Gail + BI-RADS and 3) Gail + SNPs + BI-RADS. We evaluated each model in turn by calculating a risk score for each patient (using 10-fold cross validation); used this risk estimate to construct ROC curves; and compared the AUC of each model using the DeLong method.

With 373 cases and 395 controls, we found that models developed using a single data type, BI-RADS (AUC = 0.681) was superior to the Gail (AUC = .579; p < 0.001) and SNPs (AUC = .601; p < 0.001). Each data type augmented the baseline Gail model: Gail + SNPs (AUC = .622; p < 0.02), Gail + BI-RADS (AUC = .700; p < 0.001) and Gail + SNPs + BI-RADS (AUC = .718; p < 0.001).

Using a single data type, BI-RADS features were most predictive of breast cancer in this population. When combined, each data type augmented discriminative performance.

As genetic disease prediction gains momentum, we show that BI-RADS abnormality features alone outperform and together augment demographic and genetic risk factors in the prediction of breast cancer.

Burnside, E, Liu, J, Page, C, McCarty, C, Onitilo, A, Peissig, P, Kitchner, T, Trentham-Dietz, A, Wu, Y, Broeckel, U, The Predictive Value of BI-RADS Descriptors and Genetic Variants in Women Undergoing Breast Biopsy.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14045471.html