282 Pan-tumor analysis of the association between PD-L1 combined positive score and response to pembrolizumab monotherapy

BackgroundPD-L1 is expressed on both tumor and immune cells; however, the mechanism by which PD-L1 modulates the adaptive immune response on tumor versus immune cells may differ. Additionally, the prevalence of PD-L1 expression and the partitioning between tumor and immune compartments varies by tumor type. While PD-L1 expression on tumor or immune cells can be scored separately, the PD-L1 combined positive score (CPS) captures both tumor and immune cell expression in one aggregate score. We performed a retrospective, exploratory analysis of the effectiveness of CPS as an enrichment biomarker across several studies of pembrolizumab monotherapy in patients with multiple tumor types.MethodsPD-L1 expression was assessed using PD-L1 IHC 22C3 pharmDx. Expression was measured using CPS (defined as the number of PD-L1–staining cells [tumor cells, lymphocytes, macrophages] divided by the total number of tumor cells, multiplied by 100) in tumor samples from single-arm (KEYNOTE-052 [UC], KEYNOTE-059 cohort 1 [G/GEJ], KEYNOTE-086 [TNBC], KEYNOTE-158 [cervical; SCLC], KEYNOTE-180 [EC], KEYNOTE-224 [HCC], KEYNOTE-427 [RCC]) and randomized (KEYNOTE-040 [HNSCC], KEYNOTE-045 [UC], KEYNOTE-061 [G/GEJ], KEYNOTE-119 [TNBC], KEYNOTE-240 [HCC]) pembrolizumab studies. Data were pooled across tumor types for pembrolizumab and for standard-of-care (in controlled studies), and then estimates of response rate, prevalence, and receiver operating characteristics (ROC) analysis were performed over various CPS cutpoints. CPS distribution by response, tumor type, and line of therapy were also assessed.ResultsThere were 3769 treated patients with available PD-L1 CPS (pembrolizumab, n=2678; standard-of-care, n=1091). The area under the ROC curve for ORR was 0.63 (95% CI, 0.61–0.66) for pembrolizumab and 0.48 (95% CI, 0.43–0.53) for standard-of-care when a positive association was evaluated between CPS and ORR (figure 1); individual cutpoints of 1, 10, 20, and 50 were examined (table 1). Figure 2 shows a boxplot of CPS distribution for response in pembrolizumab-treated patients.Abstract 282 Table 1Response Rates and Sensitivity at Individual CPS Cutpoints for Pembrolizumab-Treated PatientsAbstract 282 Figure 1ROC analysis of PD-L1 CPS for pembrolizumab versus standard-of-care therapyAbstract 282 Figure 2Boxplot of PD-L1 CPS distribution for responders versus nonresponders in pembrolizumab-treated patients by tumor type and line of therapy in order of descending median CPSConclusionsThis retrospective, exploratory pan-tumor analysis demonstrates that CPS is an effective scoring method for measuring PD-L1 expression and can be used as a predictive biomarker to identify patients likely to respond to pembrolizumab monotherapy. CPS demonstrated enrichment of response to pembrolizumab monotherapy across most, but not all, tumor types, including some tumor types for which efficacy favors pembrolizumab regardless of PD-L1 expression, and for which a companion diagnostic is therefore not needed. In the randomized studies, CPS did not show a consistent association with ORR for standard-of-care therapy.