Abstract Abstract #1062 Background: The performance of current prognostic breast cancer tests is limited, because they only address simple additive molecular risks, and the results are often difficult to interpret in the context of overlapping clinicopathologic risks. Gene expression-based tests can be further confounded by contamination with non-tumor cells. Our goal was to develop a protein-based profile that addresses these limitations. Methods: ER, PR, Her2, EGFR, BCL2, p27/Kip1, and p53 (IHC) and MYC (FISH) were scored in tumor cells of FFPE tissue from patients with operable, pN0-2, hormone receptor-positive breast cancer at two clinical sites. A consensus prognostic model was trained using robust cross-validation in 290 hormone therapy (HT)-treated patients using statistical pattern recognition methods to account for complex marker interactions. Tumor grade, pT, and pN were directly incorporated into the model to the extent they were not replaced by the molecular markers. Continuous risk scores ranging from 0 to 10+ were generated for the 290 HT-treated patients, 90 untreated patients, and 119 patients treated with chemotherapy and HT. Results: A predetermined threshold of 3.8 separated HT-treated patients into high and low risk groups with hazard ratios of 8.8 (p<1E-4) for metastasis and 13.0 (p<1E-4) for disease-specific survival (DSS) without the need for an equivocal intermediate risk group. NPV/PPV were 96%/35% for metastasis at 10 yr and 97%/27% for DSS at 8 yr. In multivariate analyses, the profile was independent and fully replaced the significance of all individual prognostic factors and clinical treatment guideline combinations. Within treatment guideline elevated risk groups (Adjuvant! relapse risk >15%, NPI >3.4, and St. Gallen intermediate/high), the profile reclassified as low risk 84%, 81%, and 85% of pN0 patients, and 43%, 43%, and 32% of pN1 patients, respectively. The reclassified patients had <10% 10-yr metastasis and <5% 8-yr DSS event rates. In HT-treated patients with profile risk scores >7.0, addition of chemotherapy produced a >20% DSS benefit at 8 years (p=0.04). Conclusion: This 8-marker profile with super-additive algorithm achieved significantly higher classification accuracy than treatment guidelines and could aid selection of the most appropriate level of adjuvant therapy in patients with operable breast cancer.
 
 Citation Information: Cancer Res 2009;69(2 Suppl):Abstract nr 1062.