Abstract Prediction of pathological complete response (pCR) for neoadjuvant treatment is an area of unmet clinical need, especially for triple negative breast cancer (TNBC) as pCR is correlated with better outcomes. Predicting which patients will have residual disease (RD) provides an opportunity to improve treatment planning. We developed a test to predict which patients are likely to achieve pCR or RD to the standard of care (taxane-based) neoadjuvant chemotherapy using gene expression profiling of 325 previously identified novel biomarkers. Three microarray datasets were used (GSE22226, GSE25055, and GSE25065) including a total of 594 stage II-III breast cancer patients of which 125 (21%) achieved pCR, and 469 (79%) RD. ER+ tumors were present in 57% of the patients and 52% were PGR+. Almost 90% of the patients were Her2-. Of 231 TNBC, 78 (33.8%) achieved pCR, while 153 (66.2%) RD. Of 303 ER+Her- patients 26 (8.6%) achieved pCR while 277 (91.4%) RD. The cohort was divided into balanced populations with 476 patients used for training (80%) and test (20%) rounds of model development, while 118 patients were reserved as a validation set. Combining a “winnowing” process to remove genes with least predictive power, and hundreds of thousands of step-wise runs, followed by ranking genes based on conditional probabilities, we developed a 17-gene cassette (BA100) which was locked-down in the validation set with ROC (AUC) = 0.818. With a cut-off of 83% sensitivity and 68% specificity (PPV 0.4; NPV 0.94), BA100 achieved a 16% true positive rate (true pCR) and 55% true negative rate (true RD) identifying 76% of the patients who achieved pCR, and 69% of the patients with RD. In TNBC, BA100 classified 29% as true positives (TP), 36% as false positive (FP), 30% true negative (TN), and 4.8% false negative (FN). Kaplan Meier (KM) curves showed a significant difference in 5-year disease-free survival (5Y DFS) between TP and TN (p=0.00453) or FP (p=2.09E-06). However, FP had even worse outcomes than TN patients. To improve the TP rate, additional genes expressed in TNBC plus the original 325 genes were subjected to a second round of gene selection to discriminate between TP and FP, resulting in a 16-gene cassette (BA100.1). With a cut-off of 95% sensitivity and 73% specificity (PPV 0.7; NPV 0.95), applying BA100.1 reduced the FP rates from 24% to 9%, while correctly identifying 88% of RD in the validation set. KM curves showed no significant difference in 5Y DFS between 124 TNBC (53.7%) classified as TN versus 29 TNBC (12.6%) classified as FP, while a significant difference in survival rate was found between TNBC classified as TN vs TP (Cox Proportional Harzard p=8.42e-05). Taken together, we developed a predictive test consisting of two gene cassettes that accurately identified 71% (88/104) of pCR, and 88% (417/469) of RD patients. Gene cassettes include several transcriptional repressors, PI3K signal transduction, components of telomerase, DNA repair genes, fatty acid metabolism and estrogen-independent proliferation. The test stratified TNBC with differential response to chemotherapy and survival rates so that novel approaches can be used without delay. Further validation will confirm the test utility. Citation Format: Fournier MV, Chen J, Obenauer J, Goodwin EC, Tannenbaum SH, Brufsky AM. A predictive test for neoadjuvant chemotherapy in breast cancer identifies a subset of triple negative patients with resistant disease and the poorest prognosis [abstract]. In: Proceedings of the 2017 San Antonio Breast Cancer Symposium; 2017 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2018;78(4 Suppl):Abstract nr P2-10-08.
Read full abstract