Purpose: In the on-going I-SPY2 TRIAL in neoadjuvant setting, 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) prior to surgery (second regimen). Future patients in the I-SPY2 TRIAL will have the option to skip AC (de-escalation of neoadjuvant therapy) if they are highly likely of early pathologic complete response (pCR) at inter-regimen (12 weeks). To guide selection of the candidates for the option, a MRI prediction model based on a quantitative measure of tumor volume combined with mid-treatment core needle biopsy pathology is being studied. The combined predictor is referred to as predictive residual cancer burden (pre-RCB). This study aimed to investigate the possible benefit of adding radiologic review after mid-treatment biopsy to further refine the performance of pre-RCB in selecting the candidates for de-escalation of neoadjuvant therapy.

Methods: We retrospectively reviewed 87 I-SPY2 patients who had a mid-treatment core needle biopsy at inter-regimen, had serial MRIs, and completed both first and second regimens with surgical pathologic assessment of pCR (n =34) or non-pCR (n = 53). Of the 87 patients, 27 patients were selected for a blind radiologic review to include patients for whom subtype-specific MRI model and at least 1 of 11 I-SPY2 pathologists agreed on predicted early pCR, but the final pathology was non-pCR. The selection was balanced with cases that the consensus correctly predicted pCR. One radiologist retrospectively reviewed MRIs at pre-treatment and inter-regimen for the 27 patients in a blinded fashion, and labeled the presence of residual disease on MRI at inter-regimen as follows: class 0, no residual disease; class 1, possible residual disease; class 2, obvious residual disease.

To evaluate the accuracy of selecting the 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. Pre-RCB predicts a patient as early pCR if both MR prediction model and mid-treatment core needle biopsy pathology predict early pCR (see abstract PXXX for data). After adding radiologic review, patients labeled as class 2 were removed from the candidates with predicted early pCR. PPV and sensitivity before and after radiologic review were compared.

Results: The 27 patients included 21 pCR patients, and 6 non-pCR patients. The patients were labeled as follows by the radiologist’s review: class 0, 5; class 1, 9; class 2, 13 (Table 1). The pre-RCB predictors presented PPV of 91% (range 83–100%) and sensitivity of 91% (range 76–100%). When patients labeled as class 2 by radiologic review were removed from the candidates with predicted early pCR, PPV increased to 99% (range 93–100%) at the expense of lower sensitivity of 58% (52–62%).

Conclusion: Our study showed that adding radiologic review improved PPV by ruling out patients with obvious residual disease in MRI from the de-escalation candidates based on pre-RCB. This improvement is achieved with some cost to sensitivity. However, achieving high PPV is our top priority to ensure high accuracy of directing early pCR patients to therapy de-escalation. The results of this study will be used to guide selection of candidates for de-escalation of neoadjuvant therapy in future I-SPY2 TRIAL.