Abstract
PURPOSE: The programmed death ligand 1 (PD-L1) SP142 assay with a 1% immune cell (IC) cutoff is approved for the selection of advanced triple-negative breast cancer (TNBC) patients for atezolizumab treatment. We aimed to evaluate the interobserver concordance of PD-L1 scoring and inter-assay variability of various PD-L1 assays in TNBC. METHODS: Thirty patients with primary TNBC were selected, and SP142, SP263, 22C3, and E1L3N assays were performed. PD-L1 staining in ICs and tumor cells (TCs) was scored by 10 pathologists who were blinded to the assay. The interobserver concordance among pathologists and the inter-assay variability of the four PD-L1 assays were analyzed. For SP142, the intraobserver concordance among the six pathologists was analyzed after training. RESULTS: The adjusted means of PD-L1 IC scoring ranged from 6.2% to 12.9% for the four assays; the intraclass correlations showed moderate (0.584-0.649) reader concordance. The PD-L1 IC scoring with a 1% cutoff resulted in identical scoring in 40.0%-66.7% of cases and a poor to moderate agreement (Fleiss κ statistic [FKS] = 0.345-0.534) for the four assays. The SP142 assay had the widest range of positive rate (56.5%-100.0%), lowest number of cases with identical scoring, and lowest FKS at 1% cutoff. Pairwise comparison of adjusted means showed significantly decreased PD-L1 staining in SP142 compared with the other assays in both ICs and TCs. As for the intraobserver concordance in the SP142 assay, the overall percent agreement was 87.8% with a 1% IC cutoff. After training, the proportion of cases with identical scoring at a 1% IC cutoff increased to 70.0%; the FKS also increased to 0.610. CONCLUSION: The concordance of PD-L1 IC scoring among pathologists was low, at the 1% cutoff for the SP142 assay without training. SP142 showed the lowest PD-L1 expression in both IC and TC.