Tumor Cell Proportion Assessment in Advanced Non-Squamous Non-Small Cell Lung Cancer Tissue Samples in Real-World...

Background/Objectives: Identification of driver gene alterations helps determine first-line treatment for non-squamous non-small cell lung cancer (NSCLC). Precise assessment of tumor cell proportion is critical for accurate detection of gene alterations. ASTRAL was a multicenter, prospective, observational study to investigate the agreement in tumor cell proportion assessments between different raters. Methods: Tissues collected in daily clinical practice from patients with advanced NSCLC were used. Raters included local pathologists, a Central Pathology Committee (CPC), and an artificial intelligence (AI) algorithm. Hematoxylin and eosin-stained slides were assessed by local pathologists, and digitized images of those slides were assessed by the CPC and the AI algorithm. The primary endpoint was agreement in assessment of tumor cell proportion between local pathologists and the CPC, as determined using the intraclass correlation coefficient (ICC). Secondary endpoints included agreement between the AI algorithm and local pathologists or the CPC. Results: Tissue samples from 204 patients were assessed. The ICC for local pathologists vs. the CPC showed poor to moderate agreement (0.588 [95% confidence interval (CI) 0.483-0.674]). The AI algorithm showed moderate agreement with the CPC (ICC 0.652 [95% CI 0.548-0.733]), and poor to moderate agreement with local pathologists (ICC 0.465 [95% CI 0.279-0.604]). Conclusions: The ICC for the AI algorithm vs. the CPC was numerically highest among the rater pairs, indicating a level of usefulness for the algorithm. Continued efforts are needed to ensure the accurate estimation of tumor cell proportion. Integration of AI algorithms in real-world practice may contribute to this.