PD-L1 immunohistochemistry in gastric cancer: Comparison of combined positive score and tumor area positivity across 28-8, 22C3, and SP263 assays.

Abstract

2625 Background: The clinical application of programmed death-ligand 1 (PD-L1) immunohistochemistry is complicated by multiple available assays and different testing platforms, scoring algorithms, and cut-offs applied. This study assessed the analytical comparability of three commercially available PD-L1 assays (28-8, 22C3, and SP263) and two scoring algorithms used in gastric cancer (GC). Methods: Serial sections of 100 commercially procured GC samples, selected by the 28-8 assay to represent the dynamic range of PD-L1 expression, were stained with the three in vitrodiagnostic (IVD)-grade PD-L1 assays. Stained slides were blindly and independently evaluated by three trained pathologists for intra- and inter-reader assessment. Scoring was performed using the combined positive score (CPS) and tumor area positivity (TAP) methods, followed by statistical analysis. Digital image analysis (DIA) was used to objectively assess the technical performance of each assay by simulating CPS and TAP scoring methods using the HALO platform. Results: Comparable, specific, staining patterns were observed with the three PD-L1 assays. Pathologist assessment of PD-L1 positivity was reproducible in GC sample cohorts despite discernible variability in the observed staining intensity. When the same PD-L1 cut-offs were applied, inter- and intra-assay assessments of all three assays, using either CPS or TAP scoring methods, demonstrated moderate to almost-perfect (inter-assay Cohen’s kappa [κ] ranged from 0.47 to 0.83) and substantial to almost-perfect (intra-assay κ ranged from 0.77 to 1.00) agreement, respectively. Moreover, inter-pathologist evaluation showed a significant level of reproducibility (intraclass correlation coefficient (ICC) ≥0.92). DIA confirmed no difference in technical performance when specific digital algorithms were applied. Conclusions: This GC study highlights analytical concordance in PD-L1 testing among three major PD-L1 assays when TAP and CPS scoring algorithms are prospectively applied. Independently, DIA further supports the comparability of the technical performance of the assays. These observations support flexibility in cross-application of the different PD-L1 assays and scoring algorithms currently used to characterize PD-L1–positive GC samples.