Prognostic value of visually and computationally-assessed tumor-infiltrating lymphocytes in early-stage triple-negative breast cancer (TBCRC-030).

Abstract

BACKGROUND Tumor-infiltrating lymphocytes (TILs), assessed by visual examination (VE), are prognostic and predictive in early-stage triple-negative breast cancer (TNBC). Computational assessment (CA) may provide a complementary approach. We evaluated the prognostic value of TILs by VE and CA. METHODS TBCRC 030 was a randomized phase II trial enrolling patients with BRCA1/2-proficient stage I-III TNBC to receive preoperative cisplatin or paclitaxel. The primary endpoint was pathologic response at surgery. TILs were visually scored on digitized pre-treatment biopsies per International TILs Working Group recommendations. CA used the 4D QPOR platform to generate TILs, immune heterogeneity index (IHI), and a combined immune/cell cycle biomarker (CmbI). Predictive performance for residual cancer burden (RCB) 0/1 was assessed using ROC curves and odds ratios (ORs) with 95% CIs; all statistical tests were two-sided. RESULTS Of 139 response-evaluable patients, 121 had matched VE and CA data (59 cisplatin, 62 paclitaxel). Median VE TILs were higher in responders (40.0% vs. 10.0%, p = .002) and predicted response (OR 1.86, 95% CI 1.24-2.87, AUC 0.69, 95% CI 0.57-0.80). CA CmbI differed by response group and predicted RCB 0/1 (OR 3.20, 1.05-11.07; AUC 0.62, 0.51-0.73). CA TILs and IHI were not predictive. VE TILs and CA CmbI predicted response to paclitaxel (OR 2.91, 1.56-6.14; OR 9.17, 2.01-66.39, respectively), but not to cisplatin. CONCLUSION VE TILs and CA CmbI were each associated with response to NAC in TNBC in the overall cohort and the paclitaxel arm. CA CmbI did not outperform visual assessment. Further validation is needed before clinical implementation of computational approaches.