Reduction of anthracycline use with a combined imaging and pathology prediction model in the neoadjuvant I-SPY2 trial

The I-SPY2 Trial is a randomized phase II platform trial testing novel neoadjuvant therapies. Serial MRI and pathology are used to evaluate response. New agents (+/- taxane) are tested in a 12 week (wk) treatment block, followed by Adriamycin/Cytoxan (AC). We sought to determine whether MRI functional tumor volume (FTV)-based predictive models, in conjunction with breast core biopsy, could identify early responders who had reached pCR and thus proceed to surgery without receiving AC, sparing them additional toxicity. The “predicted Residual Cancer Burden” (preRCB) models were developed using retrospective I-SPY data and prospectively tested in I-SPY2. Eligible patients (pts) were required to be on preRCB-designated treatment arms and consented to participate in the optional preRCB process. PreRCB included breast MRI at baseline, wk 3 and wk 12 of the initial I-SPY2 treatment block and a clinical core biopsy of the tumor bed at 12-weeks. Pts who met combined preRCB criteria (predicted probability of pCR above threshold in a subtype specific FTV model and no invasive tumor on breast biopsy) were eligible to proceed to surgery without AC after return of results (ROR) and consultation with their clinician. From 1/20-6/22, 172 pts underwent preRCB and completed surgery. 51/172 (30%) met combined MRI/biopsy preRCB criteria and 40/51 proceeded to surgery without AC (Table). 96% (49/51) pts who met preRCB criteria had RCB0/1 vs. 48% (57/120) of pts who did not meet criteria (OR 27 [6.5-235]; Fisher’s exact test p=1.04E-10). Overall, preRCB led to a 24% reduction in AC use without significant difference in RCB0/1 rate between those with and without AC (p=1). Implementation of preRCB in I-SPY2 identified patients with pCR prior to AC. High uptake of early surgery recommendations led to reduced use of AC while maintaining optimal outcome. I-SPY 2.2 is testing next-generation models to enable further optimization.Table: 355PPreRCB met (n=51)PreRCB not met (n=120)No AC40/51 (78%)Received AC11/51 (22%)No AC22/120 (18%)Received AC98/120 (82% )RCB 0/1RCB 2/3RCB 0/1RCB 2/3RCB 0/1RCB 2/3RCB 0/1RCB 2/338 (95%)2 (5%)11 (100%)0 (0%)12 (55%)10 (45%)45 (46%)53 (54%) Open table in a new tab