36 Digital Whole Slide Image (WSI) scoring is equivalent to microscope glass slide scoring for evaluation of programmed death-ligand 1 (PD-L1) expression across multiple tumor indications

BackgroundThe COVID-19 pandemic brought a host of new challenges, including the immediate need for digital solutions addressing the lack of remote options available to pathologists in the field of immunohistochemistry (IHC)-based companion diagnostics for Programmed Death-Ligand 1 (PD-L1) expression evaluation in tumor tissues. Agilent Technologies, Inc. investigated concordance of PD-L1 expression results recorded by trained pathologists between stained glass slides and digital whole slide images (WSIs). Formalin-fixed, paraffin-embedded (FFPE) specimens of eleven tumor indications (table 1) were evaluated in this study. Specimens were stained using the qualitative IHC assay PD-L1 IHC 22C3 pharmDx on Autostainer Link 48 and scored using TPS (Tumor Proportion Score) or CPS (Combined Positive Score) algorithms at six validated cutoffs.1 The objective was to demonstrate equivalency between digital WSI and microscope glass slide scoring.MethodsThree Agilent-certified pathologists evaluated specimen PD-L1 expression level (positive/negative) using CPS and/or TPS at relevant cutoff(s) for each indication (table 1) using two scoring modalities for the same specimen sets: 1) light microscope, and, 2) digital monitor (WSI) with a minimum 14-day washout period between glass slide and WSI reads. WSIs were generated using Leica’s Aperio AT2 scanner and evaluated using Aperio ImageScope software (figure 1) on appropriate monitors (table 2). Concordance between specimen glass slide (reference condition) and WSI PD-L1 expression results was assessed per cutoff on pooled data from all applicable indications using negative percent agreement (NPA), positive percent agreement (PPA) and overall agreement (OA) with 95% two-sided percentile bootstrap confidence intervals (CI); the acceptance criteria for equivalency at each cutoff were set at CI lower-bounds (CILBs) ≥85%. Discordant comparisons with respect to specimen screening data generated prior to inclusion in the study were also analyzed where applicable.ResultsNPA/PPA/OA CILBs for the CPS ≥1, CPS ≥10, TPS ≥1%, and TPS ≥50% cutoffs were ≥85% (table 3). NPA and OA CILBs at CPS ≥20 and CPS ≥50 were ≥85%; PPA CILBs were 83.2% and 84.2%, respectively. Discordant comparisons analysis for CPS ≥20 and CPS ≥50 suggested that WSI is not more prone to discordances in PD-L1 expression level than glass slide scoring when compared to specimen screening data (tables 4 and 5).Abstract 36 Table 1Algorithm-cutoff pairs testedAbstract 36