ORAL01.01: A Prospective, Multi-Institutional Assessment of Four Assays for PD-L1 Expression in NSCLC by Immunohistochemistry: Topic: Pathology.

PD-L1 detection by IHC in pre-treatment NSCLC specimens enriches for patient response to anti-PD-1/PD-L1 therapies. Four assays have been registered with the FDA, two of which are cleared. The tests use four separate PD-L1 antibodies on two separate staining platforms and have their own scoring system. We compared the performance of four PD-L1 antibodies, including two FDA-cleared assays and two lab-derived tests (LDTs) using a pre-defined categorical scoring algorithm. Four serial histology from 90 archival NSCLCs were distributed to three sites that performed the following IHCs: 1) 28-8 antibody on Dako Link 48; 2) 22c3 antibody on Dako Link 48; 3) SP142 antibody on Ventana Benchmark (LDT mimicking the IUO test); and 4) E1L3N antibody on Leica Bond (LDT). Slides were scanned using Aperio scanner and thirteen pathologists scored digital images by estimating the percentage of malignant and immune cells expressing PD-L1. The SP142 Ventana assay was an outlier with a significantly lower mean score of PD-L1 expression by both tumor and immune cells (p<0.0001 compared to 28-8). Pairwise comparisons showed the 28-8 and E1L3N were not significantly different in either tumor or immune cell labeling, but that 22c3 showed a slight, but statistically significant reduction in tumor cell labeling only detectable when using an average score from 13 pathologists. Evaluation of ICC between antibodies, using the average of thirteen pathologists’ scores for tumor shows very high concordance between antibodies (0.971 excluding SP142). The average pathologist immune cell scoring is worse (ICC = 0.804 excluding SP142). In the real world, individual pathologist assessment is more important than the average. In scenario assessing individual pathologist discordance when evaluating >50% immunoreactive or positive tumor cells, 28-8, E1L3N and 22c3 were equivalent. The concordance between individual pathologists reads for tumor ranged from ICC of 0.83 to 0.88 for each antibody while the ICC from immune cells for each antibody ranged from 0.17 to 0.23. Fleiss Kappa calculation to determine concordance of individual pathologists around the 50% cut-point for each antibody averaged 0.749 while the average concordance around the 1% cut-point dropped to 0.537. The assay using the SP142 antibody is an outlier detecting significantly less tumor cell and immune cell PD-L1 expression. Pathologists show excellent concordance when scoring tumor cells stained with any antibody, but poor concordance for scoring immune cell staining. Further studies are needed to assess the predictive value of the cross use of assays/drugs.