Programmed death ligand-1 (PD-L1) clone 22C3 expression in resected colorectal cancer as companion diagnostics for immune checkpoint inhibitor therapy: A comparison study and inter-rater agreement evaluation across proposed cut-offs and predictive (TPS, CPS and IC) scores

Background: Expression of programmed death ligand-1 (PD-L1) guides the use of immune checkpoint inhibitors (ICI) in several cancers. In colorectal cancer (CRC), ICI are only approved for metastatic CRC, while several studies suggest high efficacy even in operable CRC. The aim of this study was to investigate the inter-rater agreement of PD-L1 as a companion diagnostic marker.

Methods: Specimens from resected stage I-III CRC (n = 166 tumors) were stained with PD-L1 22C3 clone. PD-L1 expression was scored by two pathologists as tumor proportion score (TPS), combined positive score (CPS) and immune cell score (IC). Inter-rater agreement was tested using three different agreement coefficients.

Results: Raw scores of the two pathologists had 'good' to 'excellent' correlation. Spearman's rho for TPS=0.917 (95 %CI 0.839-0.995), for CPS=0.776 (95 %CI 0.726-0.826) and IC=0.818 (95 %CI 0.761-0.875). For TPS, kappa (κ)-agreements for both the ≥1 % and ≥10 % cutoffs had excellent correlation. For CPS the ≥1 % and ≥10 % cutoffs demonstrated κ=0.32 (95 %CI 0.12-0.51) and κ=0.36 (95 %CI 0.25-0.48) respectively. Cutoffs for IC showed κ=0.53 (95 %CI 0.18-0.79) for the ≥1 % cutoff, and κ=0.61 (95 %CI 0.48-0.73) for the ≥10 % cutoff. Gwet's agreement coefficient (AC₁) showed higher agreement coefficients than κ-values for most, but not all cut-offs.

Conclusion: Agreement for PD-L1 was good to excellent for raw scores. Agreement variation across several criteria and cut-offs suggests the need for more robust criteria for PD-L1 as a companion diagnostic marker.