Interassay and interobserver comparability study of four programmed death-ligand 1 (PD-L1) immunohistochemistry assays in triple-negative breast cancer

Different immunohistochemical programmed death-ligand 1 (PD-L1) assays and scorings have been reported to yield variable results in triple-negative breast cancer (TNBC). We compared the analytical concordance and reproducibility of four clinically relevant PD-L1 assays assessing immune cell (IC) score, tumor proportion score (TPS), and combined positive score (CPS) in TNBC. Primary TNBC resection specimens (n = 104) were stained for PD-L1 using VENTANA SP142, VENTANA SP263, DAKO 22C3, and DAKO 28–8. PD-L1 expression was scored according to guidelines on virtual whole slide images by four trained readers.The mean PD-L1 positivity at IC-score ≥1% and CPS ≥1 ranged between 53% and 75% with the highest positivity for SP263 and comparable levels for 22C3, 28–8, and SP142. Inter-assay agreement was good between 28–8 and 22C3 across all scores and cut-offs (kappa 0.68–0.74) and for both assays with SP142 at IC-score ≥1% and CPS ≥1 (kappa 0.61–0.67). The agreement between SP263 and all other assays was substantially lower for all scores. Inter-reader agreement for each assay was good to excellent for IC-score ≥1% (kappa 0.73–0.78) and CPS ≥1 (kappa 0.68–0.74), fair to good for CPS ≥10 (kappa 0.52–0.67) and TPS ≥1% (kappa 0.53–0.72). The percentage of overlapping cases in the positive/negative category was >90% between IC-score ≥1% and CPS ≥1 but below when comparing IC-score ≥1% with CPS ≥10. We demonstrate an overall good inter-reader agreement for all PD-L1 assays in TNBC along with assay specific differences in positivity and concordances, which may aid to select the right test strategy in routine diagnostics.