Relationship between programmed death ligand 1 (PD-L1) expression and clinical outcome in patients (pts) with melanoma (MEL) treated with pembrolizumab (pembro; MK-3475)

Background: In an initial analysis of the relationship between PD L1 expression and efficacy of the anti-PD-1 monoclonal antibody pembro (training set; n = 125), the preliminary cutpoint for PD-L1 positivity was an Allred proportion score (APS) ≥2 (ie, staining in ≥1% of tumor cells). Using this cutpoint, the PD-L1+ pts (71%) had a significantly higher ORR (P = 0.0007) and longer PFS (P = 0.0051) than PD-L1− pts. We report the results of an independent validation set to confirm the cutpoint and provide pooled results of the training and validation sets. Methods: The validation set included 150 PD-L1-evaluable pts with ipilimumab na¨ıve and treated MEL treated with pembro 10 mg/kg Q3W or 2 mg/kg Q3W in KEYNOTE-001. Response was assessed every 12 wk by RECIST 1.1 by independent central review. PD-L1 expression was assessed in pretreatment tumor biopsies by IHC using the 22C3 antibody. APS included staining in tumor cells. Response rate, PFS, and OS were compared between PD-L1+ and PD-L1− pts using Miettinen and Nurminen’s method (ORR) and Cox regression (PFS, OS). Results: 82% of validation set pts were PD-L1+ (APS≥2). There was a significantly higher response rate and significantly longer PFS and OS for PD-L1+ pts in the validation set (Table). The positive (PPV) and negative (NPV) predictive values for ORR in the validation set were 36% and 96%. 77% of pooled set pts were PD-L1+ (Table). PPV and NPV for the pooled set were 42% and 91%. Conclusions: Tumor PD-L1 expression was associated with higher ORR and prolonged PFS and OS in advanced MEL pts treated with pembro. Although NPV is >90%, the high prevalence of PD-L1 positivity, durable responses in PD-L1− pts, and unknown prognostic value and dynamic nature of tumor PD-L1 expression in MEL suggest PD-L1 expression may not be useful for selecting MEL pts for pembro treatment.