AI Could Improve Accuracy of PD-L1 Scoring for NSCLC

Why It Matters

Standardized AI scoring reduces diagnostic variability that directly influences immunotherapy selection, potentially improving patient outcomes and streamlining oncology workflows.

Key Takeaways

• AI matched pathologists across all PD‑L1 assays.
• AI outperformed on SP142 assay with higher correlation.
• Agreement highest at 50% TPS cutoff, crucial treatment threshold.
• Study used 80 slides, 24 pathologists’ scores as reference.
• Faster, standardized scoring could streamline immunotherapy decisions.

Pulse Analysis

The expression of programmed death‑ligand 1 (PD‑L1) on tumor cells is a pivotal biomarker that determines eligibility for checkpoint inhibitor therapy in non‑small cell lung cancer (NSCLC). Historically, four commercial immunohistochemistry assays have been deployed, each tied to a specific drug, creating a fragmented testing landscape. Manual interpretation by pathologists introduces inter‑observer variability that can shift a patient’s tumor proportion score (TPS) across the 50 % threshold, directly influencing whether they receive monotherapy or a chemo‑immunotherapy combo. Reducing this variability has been a long‑standing goal for oncologists and health systems alike.

The recent European Lung Cancer Congress abstract presented data from 80 archived slides evaluated by an AI platform against consensus scores from 24 pathologists. Across all assays, the algorithm achieved non‑inferior overall agreement, with overall percent agreement ranging from 94.3 % to 100 % at the clinically relevant 50 % cutoff. Notably, the SP142 assay—traditionally the most discordant—showed a 0.192 improvement in intraclass correlation, suggesting the AI can resolve assay‑specific challenges. These results pave the way for prospective trials that compare AI‑guided scoring with standard practice in predicting immunotherapy response.

Adopting AI‑driven PD‑L1 scoring could accelerate turnaround times, lower labor costs, and standardize results across community hospitals that lack access to multiple assays. However, integration will require regulatory clearance, validation in diverse patient populations, and alignment with reimbursement frameworks. Companies such as PathAI, already funding the study, are positioning themselves to supply turnkey solutions to pathology labs. If clinical outcomes confirm the algorithm’s predictive advantage, we may see a shift toward digital companion diagnostics that reinforce personalized oncology and reshape the economics of lung‑cancer care.