Abstract 2492: Enhanced patient selection for anti-PD-L1 treatment in metastatic NSCLC with quantitative continuous scoring of PD-L1

Abstract Background: Anti-PD-L1 therapy has demonstrated clinical activity in patients with metastatic non-small cell lung cancer (mNSCLC). However, only subgroups of patients respond and their identification via PD-L1 as a biomarker remains imperfect. PD-L1 expression is commonly assessed by pathologist tumor cell (TC) scoring of immunohistochemically (IHC) stained tissue. We developed a system for digitally scoring PD-L1 in IHC (PD-L1 QCS), which demonstrated robust scoring across studies [1]. Here, we present a comparison of PD-L1 QCS against manual scoring of PD-L1 (SP263 assay, Ventana) in the MYSTIC clinical trial. Methods: PD-L1 QCS on digitized whole slide images (WSI) comprises two deep learning models, enabling segmentation of single TCs followed by PD-L1 expression quantification via their optical density (OD). Positive cells are classified based on an OD threshold, allowing robust digital calculation of the TC percentage [1]. The analysis included 502 WSI from the MYSTIC trial (NCT02453282), representing 256 patients treated with anti-PD-L1 therapy and 246 treated with chemotherapy as standard-of-care (SoC) [2]. First, an optimal cut-point was determined by optimization against outcome, classifying patients with ≥0.575% TC as biomarker positive (BM+). Next, the approach was compared against manual TC scoring at 1%, 25% and 50% cut-off. Results: In durvalumab treated patients, median overall survival (mOS) in the PD-L1 QCS BM+ subgroup (prevalence 54.3%) was 12.1 months longer than in the BM- subgroup (19.9m vs. 7.8m, HR=0.45, CI [0.33, 0.60]). Analogous comparison of subgroups based on manually scored TC proportion at 1% (prev. 75.0%), 25% (prev. 42.6%) or 50% (prev. 29.7%) cut-points yielded a mOS difference of 7.8m (HR=0.52, CI [0.38, 0.72]), 8.3m (HR=0.61, CI [0.45, 0.82]) and 11.0m (HR=0.55, CI [0.40, 0.77]) respectively. Comparing durvalumab treatment against SoC within the PD-L1 QCS BM+ subgroup yielded a HR of 0.62 (CI [0.46, 0.82], log rank p=0.0008, 7.4m mOS delta). In comparison, a HR of 0.69 (CI [0.46,1.02], p=0.0642, 8.4m mOS delta) was obtained for manual TC scoring at 50%. Conclusion: We compared a computational pathology approach for continuous PD-L1 scoring for the selection of mNSCLC patients for anti-PD-L1 treatment against established manual scoring. Our results suggest that PD-L1 QCS has the potential to identify a larger patient subgroup that retains benefit from anti-PD-L1 treatment and more precisely identifies non-responders. References: 1. Lesniak, Jan, et al. "Quantitative computational assessment of PD-L1 enables robust patient selection for biomarker-informed anti-PD-L1 treatment of NSCLC patients." J. Immunother. Cancer, Vol. 10., 2022. 2. Rizvi NA, et al. "Durvalumab with or without tremelimumab vs standard chemotherapy in first-line treatment of mNSCLC: the MYSTIC phase 3 randomized clinical trial." JAMA Oncology. 2020;6.5:661-674. Citation Format: Jan Martin Lesniak, Markus Schick, Thomas Kunzke, Federico Pollastri, Juan Pedro Vigueras-Guillén, Harald Hessel, Susanne Haneder, Pallavi Sontakke, Karma DaCosta, Regina Alleze, Hadassah Sade, J Carl Barrett, Günter Schmidt, Ross Stewart. Enhanced patient selection for anti-PD-L1 treatment in metastatic NSCLC with quantitative continuous scoring of PD-L1 [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 2492.