Clinical Validation of AI-Powered PD-L1 Tumor Proportion Score Interpretation for Predicting Immune Checkpoint Inhibitor Response in NSCLC

Medical Laboratory Science Information Science Artificial Intelligence Medicine Oncology
Artificial Intelligence PD-L1 Tumor Proportion Score Non-Small Cell Lung Cancer Immune Checkpoint Inhibitors Clinical Validation

Clinical Validation of AI-based Interpretation of PD-L1 Tumor Proportion Score in Predicting Response to Immune Checkpoint Inhibitors in Non-small Cell Lung Cancer

In the field of tumor treatment and diagnosis, the assessment of PD-L1 (Programmed Death-Ligand 1) Tumor Proportion Score (TPS) is a critical task, especially in predicting the response to immune checkpoint inhibitors (ICIs) in non-small cell lung cancer (NSCLC). However, the evaluation of PD-L1 TPS by pathologists is limited by subjective factors, such as intra-observer/inter-observer variability and tumor heterogeneity. Recent studies have shown that applying artificial intelligence (AI) technology to provide quantitative biomarker assessment in pathological images suggests a promising future for AI in pathological diagnosis.

This research, conducted by Dr. Hyojin Kim, Dr. Seokhwi Kim, and other authors from Seoul National University Incheon Hospital, was published in the May 9, 2024, issue of JCO Precision Oncology. The study developed an AI analyzer to evaluate PD-L1 TPS and compared it with multiple board-certified pathologists to validate its performance in predicting clinical outcomes for patients with advanced NSCLC receiving single-agent ICI therapy.

Study Process

The AI analyzer was developed using 393,565 PD-L1-expressing tumor cells annotated by board-certified pathologists. These cells were derived from 802 whole-slide images stained with 22C3 PharmDx immunohistochemistry (IHC). The clinical performance of the AI analyzer was validated in an external cohort comprising 430 whole-slide images (WSIs) from NSCLC patients.

In the external cohort, three pathologists performed annotation work and compared their consensus TPS with the AI-based TPS. The study revealed significant findings: there was a strong positive correlation between the PD-L1 TPS assessed by the AI analyzer and that assessed by pathologists (Spearman coefficient of 0.925; p<0.001). The AI analyzer demonstrated clinical performance in predicting tumor response and progression-free survival (PFS) that was similar to, or even better than, that of the pathologists.

Study Results

The AI analyzer’s evaluation of PD-L1 TPS accurately predicted tumor response and PFS in patients with advanced NSCLC receiving ICI therapy. The study found that the AI analyzer performed better than pathologists in prognostication for patients in the TPS 1%-49% and TPS % groups.

Study Significance

The AI analyzer’s consistency with pathologists in evaluating TPS holds significant scientific and application value. This study not only highlights AI’s potential in supporting consistency among pathologists, but also showcases AI’s potential role in predicting ICI response in NSCLC.

Author Information and Contributions

The primary authors of the article are Dr. Hyojin Kim and Dr. Seokhwi Kim, both affiliated with Seoul National University Hospital. This project was supported by Seoul National University Incheon Hospital and Lunit Inc. All authors were responsible for various aspects of the study and provided final approval.

Conclusion

This study demonstrates AI’s significant role in precision cancer care and PD-L1 TPS interpretation, especially given the recognized challenges in accurate interpretation. AI provides an important auxiliary tool as pathology continues to move towards digital and highly accurate assessments. AI technology will further drive the development of precision oncology treatment.