Inter-Rater Agreement in PD-L1 (22C3) Expression Evaluation in Colon Cancer.

Abstract

Programmed death-ligand 1 (PD-L1) is a predictive biomarker in various cancers. In colorectal cancer, PD-L1 expression evaluation is not part of the routine examination and standardized evaluation methods are lacking. This study aimed to examine inter-rater agreement in PD-L1 expression evaluation in patients with colon cancer (CC) using different manual scoring methods, and to compare the manual methods to a digital evaluation. Two pathologists evaluated 188 PD-L1-stained slides from patients with CC. The tumor proportion score (TPS) and immune cell score (ICS) were categorized as <1%, 1-9%, and ≥10% positivity and the combined positive score (CPS) as <1, 1-9, and ≥10 positivity. Inter-rater agreement was assessed using different cutoffs and consensus scores were compared with an in-house-developed digital algorithm quantifying PD-L1. Percentage agreement was 98.9% for TPS, 85.6% for ICS, and 88.8% for CPS. When dichotomized using 1% or 10% for ICS and 1 or 10 for CPS as cutoffs, the percentage agreement increased. Only 2 cases presented with a consensus TPS ≥1%, while 96.8% had a consensus ICS of <10%, and 98.4% a consensus CPS of <10. When the digital PD-L1 score was categorized into low and high, the highest agreement with the manual evaluations was reached using the third quartile as a cutoff. Conclusively, high inter-rater percentage agreements were obtained for all scoring methods, although kappa values varied. Results comparable to the manual evaluation were reached with a digital algorithm. A digital approach can potentially eliminate inter-rater variability. Clinically relevant cutoff values remain to be determined.