J. Sun1, W. Sun1, D. Chen2, J. Li2, R. Roetzheim1, C. Laronga1, M. C. Lee1 1Moffitt Cancer Center And Research Institute,Breast Oncology,Tampa, FL, USA 2Moffitt Cancer Center And Research Institute,Biostatistics And Bioinformatics,Tampa, FL, USA
Introduction:
The Gail Model estimates 5-year risk of developing breast cancer (BC) in unaffected women using health information. Our objective was to develop a clinical assay to quantify risk of BC occurrence in women receiving benign breast biopsy using a unique gene expression signature to estimate risk of subsequent BC.
Methods:
A 56-gene malignancy risk (MR) gene signature assay was validated for formalin-fixed paraffin-embedded (FFPE) tissue. Women who developed BC <2 years from a benign breast biopsy were identified and matched by age and biopsy date to unaffected controls with 2+ years of follow-up; biopsies were obtained at a single institution from 2007-2011. The MR was applied to benign breast and available BC specimens. Receiver operating characteristic curves were generated to determine diagnostic accuracy of the MR score and the Gail Model. BC risk was calculated by MR score only, Gail score only, and combined MR and Gail scores, then analyzed using leave-one-out-cross validation.
Results:
663 women with benign breast biopsies were identified. 100 women had concurrent BC biopsies and were excluded, leaving 563 women with benign results; 48 women developed BC (cases) within 2 years of biopsy. Follow-up data was collected at 2- and 5-year follow-up; 2 control patients developed BC between years 2-5 and were reclassified. 30 BC cases were matched to 60 unaffected controls, but analyzable tissue-based assays were only obtained for 17 benign “pre-cancer” case biopsies and 35 unaffected benign control biopsies due to poor cellularity of FFPE. 28 malignant tissue case assays were also obtained for comparison (Figure 1). Kruskal-Wallis rank sum test confirmed a difference between the three groups (Figure 1). MR assay signatures were higher in the case group than control group (p=0.09).
When comparing the MR and Gail scores, the MR score did not reach statistical significance when calculated for control and pre-case groups in two separate cohorts (n=52, p=0.246; n=43, p=0.75), while the Gail score suggested difference between the two groups (n=43, p=0.107, AUC=0.683). The MR score demonstrated poor predictive value when applied alone (AUC=0.538-0.613). However, the two tests demonstrated the best predictive value when combined (AUC=0.710).
Conclusion:
Due to small sample size, we were not able to demonstrate predictive capability of the MR signature alone; however, we demonstrated the feasibility of using FFPE gene expression assays to develop a predictive test for BC. Our current technique was limited by poor cellularity of benign breast samples. We intend to further investigate the combination of MR signature and Gail Model score as a more accurate risk assessment for women undergoing benign breast biopsy.
