Prediction of pathologic complete response to chemoimmunotherapy in triple-negative breast cancer using tumor-infiltrating lymphocytes: Exploiting cutoff values.

Abstract

e12629 Background: Triple-negative breast cancer (TNBC) prognosis is significantly influenced by tumor-infiltrating lymphocytes (TILs), but the lack of validated cutoff values has limited their clinical applicability. In the era of neoadjuvant pembrolizumab plus chemotherapy (P+CT) as a new standard of care, the role of TILs as predictors of response to chemoimmunotherapy remains uncertain. Methods: This study aimed to assess TILs as predictors of pathologic complete response (pCR) within the Neo-Real study, a multicenter, real-world data investigation on neoadjuvant P+CT in TNBC. TILs were evaluated using the standardized methodology of the International TILs Working Group. Logistic regression and receiver operating characteristic (ROC) curve analysis were performed to evaluate the predictive ability of TILs expression and multivariable models for pCR. Results: The analysis included 128 patients (pts) treated with neoadjuvant P+CT (68% stage II, 26% stage III). The proportion of pts in each TILs’ category and results of ROC curve analysis are presented in the Table. A cutoff value of 10% demonstrated the highest accuracy in predicting pCR, while a high specificity was observed at a cutoff value of 50%. Thus, the probability of residual disease if TILs ≥ 50% is considerably low. Incorporating categorized clinical and pathological variables, a multivariable logistic regression model, using TILs (≥ 10% vs < 10%), Ki67 (≥ 50% vs < 50%), and tumor stage (III vs II), exhibited the highest AUC (0.688) for predicting pCR. Conclusions: Our study underscores the predictive value of TILs for pCR in TNBC following neoadjuvant P+CT. The cutoff value of ≥ 50% identified patients with a very high probability of pCR. The results reinforce TILs’ use as a biomarker for treatment de-escalation, especially for pts with TILs ≥ 50%. Further enhancement of TILs' predictive potential may be achieved through multivariable models. [Table: see text]