Abstract P2-07-08: Standardized prediction of Oncotype DX^®risk classes by local RT-qPCR

Background: Recent results from the prospective validation of the Oncotype DX® recurrence score (RS) have underlined the clinical validity of the assay for the prediction of chemotherapy benefit in ER+/HER2- early stage breast cancer patients. Due to health economic restrictions, some patients have no easy access to the test. A pre-selection of tumor samples may help identify patients with a high likelihood to be spared chemotherapy. Histology and semi-quantitative IHC are hence used to select samples for Oncotype testing, but these suffer from intra- and inter-observer variability, especially for Ki-67 which is a main factor in most RS prediction algorithms. We have established and validated a tool for the prediction of RS risk classes (TAILORx cutoff RS ≤25) based on highly standardized, reproducible and locally performed RT-qPCR measurements of ERBB2, ESR1, PGR and MKI67 mRNA using the CE-marked IVD MammaTyper®. Methods: Total RNA was extracted from whole surface 10μm sections from FFPE breast cancer samples with a known RS result and a tumor cell content ≥20%. ERBB2, ESR1, PGR and MKI67 mRNA expression was measured by RT-qPCR on a CFX96 qPCR cycler using the MammaTyper® kit. A prediction model for an RS ≤25 result was established using multivariable logistic regression. Based on this model and the training data two cutoffs for confident prediction of low chemotherapy benefit patients in a clinical setting were established at 95% and 97.5% specificity. The model and the cutoffs were then fixed and validated in a second, separate set of breast cancer samples. ROC analysis was used to characterize predictive power of the continuous values resulting from the prediction model. Positive and negative predictive values for detection of an RS ≤25 result were also determined on the validation samples using the two pre-defined cutoffs. Results: The sample set for training of the prediction model encompassed 202 samples including 29 samples (14.4%) with an RS >25. In an initial multivariable model with all four markers, PGR and MKI67 were the strongest predictors while the influence of ESR in the model was lower, but still significant. ERBB2 was no significant predictor in this set of ERBB2 negative samples and was therefore excluded from the final model which was based on three markers only. This three marker model achieved an AUC of 0.920 (95% CI: 0.871-0.968) in the training samples. When applying the fixed model from the training dataset to a second separately collected set of 104 samples containing 20 samples (19.2%) with an RS >25, an AUC of 0.883 (95% CI: 0.810-0.955) was documented. When further applying the two predefined cutoffs established in the training set, 45 and 36 of the 104 validation samples (43.3% and 34.6%) had a predicted low chemotherapy benefit result (RS ≤25). Even with the less stringent cutoff, not a single one of the RS >25 cases from the validation cohort was falsely predicted as RS ≤25 sample. Conclusion: We have established a highly reliable method for prediction of Oncotype DX® low chemotherapy benefit results based on local and cost effective mRNA measurements. This method enables local pathologies to pre-assess routine samples using a highly precise molecular tool and thereby reserve the Oncotype DX® test for cases with ambiguous cancer biology. Citation Format: Lehr H-A, Aulmann S, Etzrodt A, Laible M, Hartmann K, Gurtler C, Wirtz RM, Sahin U, Varga Z. Standardized prediction of Oncotype DX® risk classes by local RT-qPCR [abstract]. In: Proceedings of the 2018 San Antonio Breast Cancer Symposium; 2018 Dec 4-8; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2019;79(4 Suppl):Abstract nr P2-07-08.