Abstract PurposeIn the on-going I-SPY2 TRIAL, participants receive 12 cycles of weekly paclitaxel with or without addition of experimental agents for 12 weeks (first regimen), followed by 4 cycles of anthracycline-cyclophosphamide (AC, second regimen) prior to surgery. A de-escalation strategy currently being introduced in I-SPY2 will give patients the option to skip AC if it is highly likely that they have achieved early pathologic complete response (pCR) at the inter-regimen time point (12 weeks). To guide selection of candidates for this option, a prediction model combining subtype-specific MRI predictive probabilities with mid-treatment percutaneous core biopsy pathology has been developed. The combined model predicts a patient as early pCR if the MRI model based on a quantitative measure of functional tumor volume predicts a high likelihood of pCR, and pathology determines that there are no invasive cancer cells detected in the mid-treatment core biopsy. This study was performed to develop radiologic criteria for post-selection review to further reduce the possibility for incorrect de-escalation recommendation. MethodsA review of 87 I-SPY2 patients with serial MRI and inter-regimen core biopsy identified 25 patients where both the subtype specific MRI model and at least 2 of 11 I-SPY2 pathologists predicted early pCR. One radiologist retrospectively reviewed MRIs at pre-treatment and inter-regimen for the 25 patients in a blinded fashion, and labeled the presence of residual disease on MRI at inter-regimen as follows: rank 0, no residual disease; rank 1, possible residual disease; rank 2, obvious residual disease.To evaluate the accuracy of selecting candidates for de-escalation of neoadjuvant therapy (i.e. predicting patients with early pCR at inter-regimen), surgical pathology of pCR after the completion of NAC (both first and second regimens) was defined as a surrogate “truth” in this study. Positive predictive value (PPV) and sensitivity for the combined MRI-pathology model were computed for all pathologist pairs. After adding qualitative radiologic review, patients labeled as rank 2 were excluded from the candidates with predicted early pCR by the combined MRI-pathology model. PPV and sensitivity before and after radiologic review were compared. ResultsThe 25 patients included 21 patients that ultimately achieved pCR, and 4 patients that did not. Radiologic review of the inter-regimen MRI for these patients classified their MRIs as 20% (5/25) rank 0 (no residual disease), 36% (9/25) rank 1 (possible residual disease), and 44% (11/25) rank 2 (obvious residual disease). Most (3/4) non-pCR patients were classified rank 2, however one (1/4) was assessed rank 0. Eight of the 21 pCR patients (38%) were classified as having obvious residual disease on inter-regimen MRI. Without radiologic review, the combined MRI-pathology model predicted pCR with a mean PPV of 92% (range 83-100%) and a mean sensitivity of 91% (range 76-100%). When patients labeled as rank 2 by radiologic review were excluded from candidates with predicted early pCR, PPV increased to 99% (range 93-100%) at the expense of lower sensitivity of 59% (range 52-62%). Conclusion In this study, elimination of rank 2 lesions resulted in substantial improvement in PPV (92% to 99%) with a consequent reduction in sensitivity from 91% to 59%. Notably, none of the rank 1 lesions, where diagnostic interpretation was less certain, were found to be non-pCR at surgery. High PPV is essential to ensure high accuracy and safety in directing early pCR patients to therapy de-escalation. These findings will be further validated in continuing studies and radiologic re-review will be included in the de-escalation strategy for omission of AC in the I-SPY2 TRIAL. Citation Format: Natsuko Onishi, Wen Li, Sara J Venters, Denise M Wolf, David C Newitt, Elissa R Price, Jessica Gibbs, Barbara LeStage, the I-SPY 2 Pathology Working Group, the I-SPY 2 Imaging Working Group, the I-SPY 2 Consortium, William F Symmans, Nola M Hylton. Radiologic review to refine selection of candidates for de-escalation of neoadjuvant therapy [abstract]. In: Proceedings of the 2020 San Antonio Breast Cancer Virtual Symposium; 2020 Dec 8-11; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2021;81(4 Suppl):Abstract nr PS3-02.