Abstract P6-06-17: Gene signature model predicts metastatic onset better than standard clinical markers - Nested case-control design uniquely enables enrichment for biologically relevant features

Abstract

Abstract Purpose Considering that breast cancer is a heterogeneous disease, we aimed to improve prediction of patients at high-risk for metastatic disease utilizing a nested case-control design uniquely enabling enrichment for relevant phenotypes. Methods In Sweden cancer registration has a legal basis and the Swedish Cancer Registry has a breast cancer coverage of more than 96% in validation studies. We identified women diagnosed with primary breast cancer from January 1, 1997, to December 31, 2005, in the Stockholm health care region. Patients developing distant metastatic disease (cases) were selected and controls (free from distant disease) were randomly matched by adjuvant therapy, age and calendar period at diagnosis. The nested case-control study included 768 study subjects (621 patients including two patients with bilateral breast tumors) with detailed manually collected clinical information and complete follow-up (according to the national guidelines by the Swedish Breast Cancer Group (SweBCG)). Primary tumor grade, Human Epidermal Growth Factor Receptor 2 (HER2) and Ki-67 expression was reassessed and scored by a breast cancer pathologist. All patients were profiled with the Affymetrix array (Human Cancer G110) containing in total 52,378 probes, both including control probes of different types, as well as probes corresponding to human transcripts. The study subjects were randomly and equally divided into the discovery set or validation set. Metastatic onset predictive capacity was compared including either clinical variables only (1) or combining clinical and genetic information (2), estimating ROC curves and AUC. Results Convincingly, genes and pathways found to be differentially expressed in our nested case-control study included a wide-spectra of well known as well as candidate regulators of the metastatic cascade. In our study, 313 annotated genes (under the strict Bonferroni cut-off) were differentially expressed in patients developing distant metastatic disease in contrast to patients free from disseminated disease. Our metastasis gene signature model accommodates 50 genes predicting risk for metastatic disease after adjustment of standard clinical markers. Regulation of immune response activity was a common feature among the genes as included. It is clear from the area under the ROC curve, AUC, that the predictive capacity of the metastasis gene signature model (2), also validated in the validation set of study subjects, was superior to the model with clinical variables only (1), as evidenced by AUC (0.86; 95% CI, 0.83 to 0.89 for the metastasis gene signature and 0.73; CI, 0.71 to 0.76 for clinical variables only). The gene signature model predicted both high-risk of early and later onset of metastatic disease. Conclusion In our nested case-control study, we identified genes and pathways differentially expressed in patients developing distant metastatic disease compared to patients without disseminated disease. Our validated metastasis gene signature model enabled better risk prediction compared to standard clinical markers, capturing both high-risk of early and later metastatic disease onset, of potential vital importance in the clinical setting. Citation Information: Cancer Res 2013;73(24 Suppl): Abstract nr P6-06-17.