Comparison of Immune Response Assessment in Colon Cancer by Immunoscore (Automated Digital Pathology) and Pathologist Visual Scoring

Simple Summary The immune response to colon cancer (CC) is highly variable among patients and is clinically relevant. In this study, we compared the immune response assessment for early-stage CC, as measured by Immunoscore (IS), to pathologist visual scoring of the CD3+ and CD8+ T-cell densities at the tumor site (T-score). The objectives were to determine the inter-observer agreement between pathologists and the concordance between the two methods. Agreement between pathologists was minimal to weak. Moreover, a weak concordance between the two methods was observed, leading to misclassification of 48% of cases by pathologist scoring. Due to the high level of immune infiltrate heterogeneity resulting in disagreement of interpretation among pathologists, IS is unlikely to be reproduced via non-standardized methods. Abstract Adjunction of immune response into the TNM classification system improves the prediction of colon cancer (CC) prognosis. However, immune response measurements have not been used as robust biomarkers of pathology in clinical practice until the introduction of Immunoscore (IS), a standardized assay based on automated artificial intelligence assisted digital pathology. The strong prognostic impact of the immune response, as assessed by IS, has been widely validated and IS can help to refine treatment decision making in early CC. In this study, we compared pathologist visual scoring to IS. Four pathologists evaluated tumor specimens from 50 early-stage CC patients and classified the CD3+ and CD8+ T-cell densities at the tumor site (T-score) into 2 (High/Low) categories. Individual and overall pathologist scoring of immune response (before and after training for immune response assessment) were compared to the reference IS (High/Low). Pathologists’ disagreement with the reference IS was observed in almost half of the cases (48%) and training only slightly improved the accuracy of pathologists’ classification. Agreement among pathologists was minimal with a Kappa of 0.34 and 0.57 before and after training, respectively. The standardized IS assay outperformed expert pathologist assessment in the clinical setting.