Abstract
INTRODUCTION: The aim of this project is to assess interobserver agreement for programmed death-ligand 1 (PD-L1) scoring on of non-small cell lung cancer (NSCLC) on cytological specimens in a large-scale multicenter study, by exploiting the cell block-derived tissue microarray (cbTMA) approach. METHODS: A total of 65 cell blocks (CB) diagnosed as NSCLC were retrospectively collected and selected for TMA preparation. Hematoxylin-eosin and PD-L1 stained slides were digitized and uploaded on a free web sharing platform. Participants were asked to provide PD-L1 assessment by using the clinically relevant cutoff of tumor proportion score (TPS) (<1%; 1%-49%; >50%). Interobserver agreement was calculated using Fleiss's κ. RESULTS: Of 65 CBs, 11 were deemed not suitable; therefore, an overall number of 54 cores were used for the preparation of four TMAs. A total of 1674 evaluations were provided by 31 cytopathologists from 21 different institutions in nine countries. The statistical analysis showed a moderate overall agreement (κ = 0.49). The highest agreement was achieved in the TPS >50% category (κ = 0.57); moderate agreement was observed in TPS <1% category (κ = 0.51) and the lowest κ value was obtained for TPS 1%-49% category (k = 0.32). CONCLUSIONS: The overall moderate agreement observed showed that there is still room for improvement in inter-pathologist agreement for PD-L1 evaluation on cytological samples, highlighting the need for standardization in sample preparation, focused training in PD-L1 evaluation on cytological material, and the integration of machine learning tools to improve interobserver consistency.