The I-SPY 2 TRIAL enrolls women with locally advanced, molecular high-risk breast cancer. An integrated Residual Cancer Burden (iRCB), based on MRI volume change through treatment, is used to predict pathologic complete response (pCR) in the randomization/evaluation Bayesian engine. With the goal of effective de-escalation of treatment for patients exhibiting an early response, biomarkers are being assessed for their ability to predict pCR, alone or with MR data, during treatment. Here, we present the results of a pilot study to examine if invasive tumor cellularity in mid-treatment tissue core biopsies predicts pCR in a 40-patient cohort of I-SPY 2 patients. Other pathologic variables evaluated include Ki67, tumoral histologic features, and stromal tumor-infiltrating lymphocytes (sTILs).
I-SPY 2 TRIAL pathologists (N=4) were provided images of H&E-stained and Ki67 IHC- labelled (DAKO/Agilent, clone MIB-1) core biopsy sections from 40 patients at the inter-regimen time point, ~12-weeks into treatment. Of the 40 patients, 35 had 4 cores, 3 had 3 cores, and 2 had 2 cores assessed. In total, images from 153 cores were evaluated. For each core, pathologists were asked to score the % area occupied by tumor bed (treatment changes and/or residual cancer), % of viable invasive tumor (0-100%) within tumor bed (with Nottingham grading, % Ki67 labelled, and % sTILs, using standardized guidelines). As decided by the pathologist group, only cores with identified tumor bed were included in the initial analysis. Concordance between pathologists was assessed for all scored criteria, using % agreement for dichotomous variables, and Pearson correlation (r)/standard deviation (sd) for continuous variables. The maximum and average cellularity recorded over all cores/patient, averaged over all pathologists, were analyzed for association with pCR using t-tests (significance threshold: p<0.05). Fisher’s Exact test was used for dichotomous variables, and Pearson’s correlation for association of continuous variables with the residual cancer burden (RCB) index.
Pathologist were in general agreement about the presence or absence of tumor bed, with greater than 82% agreement between any two (83-96%), and an overall agreement of 77%. For scoring the % of the tumor bed involved by invasive cancer, correlations between pairs of pathologists ranged from 0.79-0.95 (mean(r)=0.87, sd=5%), and agreement on a binary presence/absence of invasive cancer was 78%. Both the mean (t-test: p=7.59E-05) and maximum (t-test: p=0.0012) %invasive tumor at 12 weeks, scored as an average over all pathologists, were significantly higher in patients who did not achieve pCR than in responders. We also treated %invasive cellularity as a dichotomous variable (present/absent). 90% (9/10) of patients scored by all pathologists as 0% invasive tumor cells (absent) achieved a pCR, vs only 20% (6/30) of patients scored as >0% invasive cellularity by one or more pathologists (present) (OR=32, Fisher p=0.0005); yielding a positive predictive value for pCR of 0.9. Ki67 and sTILS at 12 weeks were fairly concordant across pathologists ((r,sd)=(0.92, 8.45%) and (0.82,5.5%), respectively), but did not associate with response (p>0.05 for pCR, RCB01, or RCB index). Tumor histologic grade at 12 weeks, assessed in 29/30 patients with non-zero cellularity, trended toward association (Fisher p=0.078): 44% (4/9) with Grade 3 went on to have a pCR, vs. 15% (2/13) with Grade 2 and 0 with Grade 1. These data demonstrate the utility of invasive tumor cellularity as a predictor of pCR in a clinical setting.
In this pilot study we demonstrate that the absence of invasive cancer cells within identified tumor bed in mid-treatment core biopsy samples is highly predictive of pCR.