MA08.05 High Concordance in Scoring PD-L1 in NSCLC Using the Roche Digital Pathology uPath system and PD-L1 Assessment Algorithm

Abstract

Accurate and reproducible assessment of PD-L1 expression is crucial in guiding the use of anti PD-1/PD-L1 immunomodulatory therapy. Interpretation can be challenging and both intra- and inter-pathologist discordance has been shown to be potentially significant in affecting clinical outcomes. Digital pathology and algorithmic assessment tools provide an opportunity to increase accuracy and concordance when making this important assessment in non-small cell lung cancer (NSCLC). Sections of 198 NSCLCs, (84 EBUS aspirates or small tissue biopsies and 114 resections) immunolabelled for PD-L1 using the Roche-Ventana SP263 antibody were scanned with a Roche DP200 digital image scanner and analysed using uPath slide viewer and the Roche PD-L1 interpretative algorithm. Sections were assessed by three ‘blinded’ pathologists for expression of PD-L1 which was given as an absolute percentage, the tumour proportion score (TPS) in four ways; (1) unassisted (by pathologist using conventional microscopy), (2) automated whole slide (unsupervised uPath assessment of whole slide), (3) automated annotated (uPath assessment of slide annotated by pathologist) and (4) assisted (by pathologist using digital image assisted by uPath assessment of annotated slide). The last method of assessment was repeated after a six-week ‘washout’ period. Intraclass correlation co-efficient (ICC) and Cohen’s Kappa (for clinical categories of <1%, 1-49% or ≥50% TPS) were used to compare concordance of scoring assessment methods.Table 1Comparison of TPS scoring methodsTPS assessment methods - Full CohortICCKappap-valueAssisted vs unassisted0.9730.856<0.0001Assisted vs automated whole slide0.9110.475<0.0001Assisted vs automated annotated0.9530.598<0.0001TPS assessment methods - small biopsies/EBUS onlyAssisted vs automated annotated0.9520.836<0.0001 Open table in a new tab Intra-pathologist concordance between repeated assisted reads was very high (ICC 0.991 K 0.989 p<0.0001) as was inter-pathologist concordance for assisted interpretations (ICC 0.986 K 0.910 p<0.0001 respectively) Automated algorithmic assessment guided by pathologists is an accurate approach that helps produce consistent and reliable interpretation of PD-L1 expression in NSCLC by reducing intra- and inter- observer discordance. Digital pathology and automated algorithms are powerful tools that can help optimise the predictive power of PD-L1 expression as a predictor of response to immunomodulatory therapy.