Aakash Desai, Associate Director, Phase 1 and Precision Oncology Program at UAB O’Neal Comprehensive Cancer Center, shared on LinkedIn:
“PD-L1 <1% vs 1-10% is one of the most important lines we draw in first-line NSCLC. It’s also one of the hardest to read by eye.
A new study at ESMO – European Society for Medical Oncology (poster 313P, October 25) 2026 compared AI and pathologist PD-L1 scoring on 993 resected NSCLC specimens from Japan’s LC-SCRUM-Advantage/MRD study:
- The AI and the pathologist put tumors in the same TPS category 73% of the time (κ: 0.58).
- Agreement was lowest in tumors the pathologist called negative (<1%).
- 211 tumors were “negative” by the pathologist but ≥1% by AI. Of those, 97% (204) landed in the 1-10% band.
So the disagreement sits almost entirely at the cutoff, and that cutoff changes the first-line regimen.
The tool (Lunit SCOPE PD-L1) is research-use only. Surgical specimens aren’t the small biopsies we usually score in advanced disease, and the abstract can’t tell us which read better predicts benefit from immunotherapy.
Open question for my pathology and thoracic colleagues: when AI and the pathologist split on either side of 1%, whose read should drive treatment?
Definitely an interesting direction for PDL1 scoring to be made more objective.”
Lunit Highlights AI-Powered Pathology Insights Across Cancer Types at ESMO 2026.
You can also read: ESMO 2026 Lung Cancer: 10 Abstracts to Watch.
