AI pathology scoring flags longer survival on pembrolizumab in metastatic triple-negative breast cancer

An artificial intelligence platform that reads digital pathology slides identified a subgroup of patients with metastatic triple-negative breast cancer who lived longer without disease progression on pembrolizumab, according to a retrospective study published in Breast Cancer Research. In the analysis of 49 patients whose tumors tested positive for PD-L1 by standard pathology, an AI-based assessment of PD-L1 flagged cases with significantly longer progression-free survival, while assessment by pathologists did not reach statistical significance.

The researchers analyzed samples from patients treated with pembrolizumab between 2021 and 2024, all with a PD-L1 combined positive score of at least 10. Tumor specimens were classified as pre-treatment or post-treatment depending on prior systemic therapy exposure: 23 patients contributed a pre-treatment specimen only, 24 a post-treatment specimen only, and 2 contributed both, so the samples were not serially paired. Immune phenotypes and immune cell densities were quantified from hematoxylin-eosin and PD-L1-stained whole slide images using the Lunit SCOPE IO platform for H&E slides and Lunit SCOPE uIHC for PD-L1 immunohistochemistry, with PD-L1 status evaluated independently by the AI and by pathologists.

The two assessment methods diverged sharply. AI-based PD-L1 positivity was observed in 51.5% of samples, compared with 74.2% by pathologists. Over a median follow-up of 13.2 months, AI-positive cases had a median progression-free survival of 8.8 months versus 6.7 months for AI-negative cases, a hazard ratio of 0.50 (95% CI 0.27–0.94; p = 0.031). Pathologist-assessed PD-L1 positivity showed a median of 7.2 versus 7.9 months, with a hazard ratio of 0.63 (95% CI 0.38–1.04) that did not reach significance (p = 0.072). The association was strongest in pre-treatment samples, where AI-based positivity corresponded to a median progression-free survival of 7.2 versus 4.4 months (HR 0.29, 95% CI 0.12–0.71; p = 0.006); no such association appeared in post-treatment samples.

The survival benefit tied to AI-based positivity emerged after the first six months of treatment, with a hazard ratio of 0.30 beyond six months compared with 0.73 before, a pattern the authors note is consistent with the delayed separation typically seen with immune checkpoint blockade. The study also examined the immune composition of PD-L1-positive immune cells, finding that macrophages accounted for a higher proportion in post-treatment than in pre-treatment samples with a combined positive score of at least 10 (34% versus 16%; p = 0.0044). Post-treatment PD-L1 positivity, which was accompanied by this increased proportion of PD-L1-positive macrophages, showed no association with longer progression-free survival.

Pembrolizumab plus chemotherapy already improves survival in PD-L1-positive metastatic triple-negative breast cancer, but responses remain heterogeneous and robust predictive biomarkers are lacking, which motivated the quantitative approach. The authors emphasize that the findings are hypothesis-generating and require validation in an independent cohort before AI-based scoring can be considered for patient selection in this setting. As a single-arm retrospective analysis of 49 patients with a short median follow-up, the study cannot establish that AI scoring improves clinical outcomes, and the apparent advantage may reflect differences in sample timing and cohort composition. Nevertheless, the results suggest that fully automated quantification of PD-L1 expression and tumor microenvironment features on routine slides could, if validated, sharpen the currently coarse PD-L1 cutoff used to guide immunotherapy decisions in metastatic triple-negative breast cancer.

Subject of Research: AI-based digital pathology assessment of PD-L1 and the tumor microenvironment as a predictor of pembrolizumab response in metastatic triple-negative breast cancer

Article Title: Artificial intelligence-based tumor microenvironment and PD-L1 analysis to predict pembrolizumab response in metastatic triple-negative breast cancer

Article References: Kim, J. H., Cha, Y. J., Shin, S.-J., Bae, S. J., Ahn, S. G., Hwang, S., Lee, S., Ahn, C. H., Pereira, S., Ryu, H. J., Jeong, J., & Koo, J. S. (2026). Artificial intelligence-based tumor microenvironment and PD-L1 analysis to predict pembrolizumab response in metastatic triple-negative breast cancer. Breast Cancer Research. https://doi.org/10.1186/s13058-026-02401-w

Image Credits: AI Generated

DOI: 10.1186/s13058-026-02401-w

Keywords: triple-negative breast cancer, pembrolizumab, PD-L1, artificial intelligence, digital pathology, tumor microenvironment, progression-free survival, biomarkers, immunotherapy, Lunit SCOPE, macrophages, retrospective study