Contribution of tumour and immune cells to PD-L1 as a predictive biomarker in metastatic triple-negative breast cancer (mTNBC): analysis from keynote-119

Abstract

Background: We evaluated whether PD-L1 expression on tumour cells contributes to PD-L1 as a predictive biomarker for pembrolizumab in mTNBC. Methods: Patients with mTNBC received pembrolizumab or chemotherapy. PD-L1 expression (PD-L1 IHC 22C3 pharmDx) was quantified per tumour proportion score (TPS; percentage of PD-L1-expressing tumour cells [partial or complete membrane staining] relative to total number of tumour cells) and combined positive score (CPS; number of PD-L1-staining cells [tumour cells, lymphocytes, macrophages] divided by total number of viable tumour cells, multiplied by 100). Quantitative immune cell density (QID) was CPS minus TPS. TPS, CPS, and QID ability to predict ORR (ROC analysis) and overall survival (OS) hazard ratio (HR) (pembrolizumab vs chemotherapy) were evaluated. Results: With pembrolizumab, area under ROC curve (95% CI) was 0.69 (0.58–0.80) for CPS, 0.67 (0.56–0.77) for QID, and 0.55 (0.46–0.64) for TPS. At each cutoff, QID had lower sensitivity and Youden Index and larger OS HR than CPS. QID appeared to be orthogonal to TPS. Conclusions: Trends suggest tumour cell expression is an important component of PD-L1 as a predictive biomarker of pembrolizumab in mTNBC. When immune cells alone were used, a meaningful number of responders was missed, and OS benefit trended toward higher HR estimates.