Traditional manual scoring of immunohistochemistry (IHC) faces limitations in evaluating novel biomarkers for precision oncology. Conventional companion diagnostics rely on subjective, ordinal scoring systems (e.g., 0, 1+, 2+, 3) which may limit their ability to capture more complex biological mechanisms. Conversely, computational pathology AI tools capture objective, quantitative measurements, many that cannot be achieved by eye alone. Because computational pathology scores cannot be predicted manually, transitioning from ordinal frameworks to automated, continuous mathematical models is a difficult paradigm shift for pathologists to trust. To establish the analytical evidence required to build trust, a Roche-sponsored study evaluated the reproducibility of the TROP2 (EPR20043) NSCLC RUO algorithm across 12 real-world laboratories in eight countries. To evaluate the pathologist workflow, centrally stained slides were provided to determine concordance of digital results, and unstained slides were provided to assess full device concordance. The former achieved an overall inter-site agreement of 100% and the latter workflow achieved an overall inter-site agreement rate of 94.1%, which increased to 99.8% when excluding borderline cases near the established threshold. The workshop will conclude with an interactive Q&A forum to discuss how the results may impact the future of AI-driven precision oncology.

Learning Objectives:

1. Describe the biological mechanism of TROP2 normalized membrane ratio (NMR) in NSCLC
2. Discuss analytical reproducibility & concordance rates of automated TROP2 NMR scoring across 12 diverse international labs
3. Understand pathologist & lab roles and how this study serves as a practical guide for computational pathology deployment