Statistical modeling of a novel clinical trial design using neoadjuvant therapy (NAT) to personalize therapy in patients (pts) with triple-negative breast cancer (TNBC).
595 Background: 40-50% of pts with TNBC develop pathologic complete response (pCR) with adriamycin/cyclophosphamide (AC)àtaxane (T) NAT; thus, most pts treated in randomized trials (RCTs) adding experimental drugs (ED) to standard NAT do not benefit from trial participation. A personalized trial design that enriches for non-pCR to standard NAT would diminish toxicity from ED in pts who do not need them and enrich ED in high-risk pts that are most likely to benefit. Methods: ARTEMIS (NCT02276443) is a non-randomized trial to study personalization of NAT in TNBC. Tumor biopsies were performed pre-NAT and volumetric change by ultrasound (VCU) after 4 cycles of AC (or upon clinical progression) assessed response. Pts with sensitive TNBC (VCU >=70% after AC) had T as the second phase of NAT. Pts with <70% VCU were offered phase II trials. pCR was assessed at surgical resection. 273 pts had available pCR status and 222 had complete data to generate a model predictive of response using multivariate logistic regression with common clinical factors. Data was randomly divided into training (n=111) and validation (n=111) sets. Results: 85 pts (38%) had pCR and VCU after AC x 4 was the strongest predictor of pCR. Other factors significant on multivariate analysis and included in the model were T stage (T1-4), stromal TIL, Ki67 and PD-L1. When applied to the validation data set, the accuracy of this model for predicting pCR was 76.6%, sensitivity 78.6% and specificity 75.4%. The PPV was 66.0% and the NPV was 85.2% with a ROC curve AUC of 82.4%. Using these data, ED exposure (table) was estimated for the ARTEMIS study design vs a 1:1 or a 2:1 RCT design (with an estimated pCR in control arm=40%), with a demonstrated benefit for personalization. Conclusions: This modeling indicates that personalization of NAT trials has the potential to enrich ED exposure for non-responsive disease as well as diminish ED exposure in pts likely to achieve pCR with standard NAT. Improved prediction of pCR would further enhance personalized trial design. Clinical trial information: NCT02276443 . [Table: see text]