by Christos Evangelou, MSc, PhD – Medical Writer and Editor
According to results presented at the European Lung Cancer Congress (ELCC) 2025 by investigator Dr. Martin Reck, Head of the Department of Thoracic Oncology at Lung Clinic Grosshansdorf in Germany, artificial intelligence (AI)-based digital pathology scoring of PD-L1 expression can identify 4% more patients with PD-L1 tumor cell expression ≥50% compared to manual scoring.1 This finding suggests that digital pathology could enhance patient selection for adjuvant immunotherapy in non-small cell lung cancer (NSCLC) following complete resection and chemotherapy.
“An exploratory assay demonstrated that digital pathology increased the identification of patients with PD-L1 ≥50% compared to manual scoring,” noted Dr. Reck in an interview with Pathology News. “Patients identified by digital pathology exhibited comparable efficacy benefits. In real-world settings, an approved digital pathology algorithm could potentially identify more patients eligible for adjuvant atezolizumab.”
Study Rationale
Immunotherapy has revolutionized NSCLC treatment, and PD-L1 expression is a biomarker for patient selection. The IMpower010 trial was the first to demonstrate significant disease-free survival (DFS) improvement with adjuvant atezolizumab versus best supportive care in patients with NSCLC after complete resection and adjuvant chemotherapy.2
The greatest benefit from adjuvant immunotherapy occurred in patients with stage II-IIIA NSCLC and PD-L1 tumor cell (TC) expression ≥50%. However, PD-L1 assessment faces several challenges, including intratumoral heterogeneity, interobserver variability, and differences in scoring systems used in different assays.
To address PD-L1 scoring challenges, researchers developed an AI model for PD-L1 measurement (AIM-PD-L1). In an exploratory analysis, the researchers compared manual and digital pathology scoring of PD-L1 expression to assess whether automated scoring could identify patients likely to benefit from atezolizumab.
Methodology
IMpower010 is a global, multicenter, open-label, randomized phase 3 study. Eligible patients had completely resected stage IB-IIIA NSCLC, ECOG performance status 0-1, and underwent lobectomy with tumor tissue available for PD-L1 analysis.2 Patients received 1-4 cycles of cisplatin-based chemotherapy before randomization (1:1) to either atezolizumab (1200 mg every 3 weeks for 16 cycles or 1 year) or best supportive care.2
The primary endpoint was investigator-assessed DFS, which was tested hierarchically across different populations. Secondary endpoints included overall survival (OS) in the intention-to-treat population and DFS in the population with PD-L1 TC ≥50% stage II-IIIA disease.2
For this exploratory analysis, researchers compared manual and AIM-PD-L1 scoring on SP263-stained TC PD-L1 samples.1 The AIM-PD-L1 system quantified PD-L1 TC expression for continuous digital scoring (range 0%–100%), identified tumor and stromal tissue regions, made automated cell and tissue predictions, and computed PD-L1 scores for each whole slide image.1
Key Findings
According to findings from the exploratory analysis presented at ELCC 2025, digital pathology can increase the proportion of patients eligible for neoadjuvant immunotherapy. Digital pathology scoring using AIM-PD-L1 identified 30 additional patients (4%) with PD-L1 TC ≥50% compared with manual scoring. Positive, negative, and overall percent agreements between digital and manual scoring were all approximately 91%.1
“Digital pathology has the potential to significantly increase the identification of patients with high PD-L1 expression, thereby allowing more patients to benefit from adjuvant atezolizumab therapy,” Dr. Reck stated.
DFS and OS benefits were similar between digital pathology and manual scores at the ≥50% TC cutoff.1 Hazard ratio for DFS was 0.46 (95% confidence interval [CI], 0.31–0.69; P = 0.00012) with digital pathology and 0.45 (95% CI, 0.29–0.69; P = 0.00016) with manual scoring. Similarly, the hazard ratio for OS was 0.49 (95% CI, 0.30–0.80; P = 0.00330) with digital pathology and 0.44 (95% CI, 0.26–0.74; P = 0.00150) with manual scoring. Patients who were PD-L1-positive by digital pathology (regardless of manual scoring results) showed similar clinical outcomes with atezolizumab compared to those who were positive by manual scoring.1
Addressing Challenges in PD-L1 Assessment
According to Dr. Reck, a significant advantage of digital pathology is its potential to address some of the challenges associated with PD-L1 assessment. “Digital pathology is able to quantify each cell across the whole slide image with high reproducibility,” he said. Digital pathology algorithms may improve pathologist reproducibility and accuracy when introduced as an AI-assisted tool by guiding pathologists in their assessment, particularly at difficult cut points, Dr. Reck added.
He further explained how this technology tackles intratumoral heterogeneity:
“Intratumoral heterogeneity is an important factor in the decision-making process for determining patient eligibility based on PD-L1 assessment. As a scalable and quantitative tool, digital pathology may help to better characterize intratumoral heterogeneity.”
Dr. Reck added that because the digital pathology algorithm uses the whole slide image rather than just selected areas, it may be less biased by reducing intratumoral heterogeneity as well as interobserver variability, providing a more accurate representation of PD-L1 expression across the tumor.
Barriers to Implementation
Dr. Reck clarified that currently, AIM-PD-L1 can be used only for research and is not approved for diagnostic procedures. When asked about the timeline for clinical implementation, he explained that validation and submission to regulatory authorities are needed to become available for use in clinical practice.
“It remains difficult to figure out an accurate timeline related to multiple registrational steps; however, I would assume that the first tools might be available in the upcoming years,” Dr. Reck added.
Commenting on potential barriers to implementing AI-based PD-L1 scoring in routine clinical practice, Dr. Reck stated that to gain broad acceptance, clinicians and pathologists need to trust the accuracy of AI-generated scores and understand the rationale behind AI’s decisions.
He also noted that it is important to ensure seamless integration into clinical workflows and compatibility with existing systems, adding that significant investment in infrastructure and ongoing maintenance may be required.
“While still in the early phases, access and adoption of slide scanning and image management and analysis platforms are growing in clinical practice, and this trend is expected to continue as the body of evidence for the benefits of digital pathology grows,” Dr. Reck said.
Looking Ahead
Dr. Reck emphasized that the potential applications of digital pathology extend beyond PD-L1 assessment.
“A range of analyses are in progress to assess the capability of digital pathology to identify patients with oncogenic driver mutations and to evaluate immune markers using H&E slides,” he said.
Dr. Reck also foresees a broader shift toward AI-assisted precision oncology, in which digital pathology can transform multiple aspects of biomarker testing in lung cancer and potentially other malignancies.
References
- Reck M, Felip E, Altorki N, et al. IMpower010: Digital pathology vs manual SP263 PD-L1 tumour cell scoring on whole imaging slides from patients with stage II-IIIA PD-L1 TC ≥50% NSCLC. Presented at: European Lung Cancer Congress (ELCC) 2025; March 26-29, 2025; Paris, France.
- Felip E, Altorki N, Zhou C, et al. Overall survival with adjuvant atezolizumab after chemotherapy in resected stage II-IIIA non-small-cell lung cancer (IMpower010): a randomised, multicentre, open-label, phase III trial. Ann Oncol. 2023;34(10):907-919. doi:10.1016/j.annonc.2023.07.001
No audio available for this article yet.
No quiz available for this article yet.
