Poster
An Empirical Framework for Validating Artificial Intelligence–Derived PD-L1 Positivity Predictions Applied to Urothelial Carcinoma
SITC 2019
Study Background
• Assessing programmed death ligand 1 (PD-L1) immunohistochemistry (IHC) expression plays an important role in identifying patients likely to benefit from anti–programmed death-1/PD-L1 therapies in advanced cancer, including urothelial carcinoma (UC)
• Studies have shown moderate-to-strong interobserver agreement for pathologist assessment of PD-L1 expression on tumor cells, with moderate-to-poor concordance for immune cell scoring1–3
• Thus, conventional pathologist estimation of whole-slide image scores is a suboptimal approach to obtain reference data for the evaluation of the performance of image-analysis algorithms, especially
for immune cell scoring
• Studies have shown moderate-to-strong interobserver agreement for pathologist assessment of PD-L1 expression on tumor cells, with moderate-to-poor concordance for immune cell scoring1–3
• Thus, conventional pathologist estimation of whole-slide image scores is a suboptimal approach to obtain reference data for the evaluation of the performance of image-analysis algorithms, especially
for immune cell scoring
Authors
Beck et al.