Abstract 7658: Artificial intelligence-powered spatial analysis of tumor-infiltrating lymphocytes as a prognostic biomarker for pembrolizumab plus chemotherapy in recurrent or metastatic head and neck squamous cell carcinoma

Abstract Background: Programmed cell death ligand 1 (PD-L1) expression alone has limited predictive value for pembrolizumab plus 5-fluorouracil and platinum (PFP) therapy in recurrent or metastatic head and neck squamous cell carcinoma (HNSCC). An artificial intelligence (AI) -powered spatial tumor-infiltrating lymphocyte (TIL) analyzer may provide independent prognostic information. Methods: This study included 63 patients with locally incurable HNSCC who received first-line PFP at Samsung Medical Center. Patients were classified into inflamed phenotype (IP) or non-inflamed phenotype (NIP) based on the 5% proportion threshold of inflamed 1 mm2-sized grids over total grids within hematoxylin and eosin-stained whole-slide images (WSIs) analyzed using Lunit SCOPE IO. We evaluated the prognostic value of inflamed status and densities of intratumoral TILs (iTILs), stromal TILs (sTILs), and tumor microenvironment TILs (tTILs). Results: Patients were predominantly male (78%). The median age at PFP initiation was 57 years. The most common primary tumor site was oral cavity (49. 2%), followed by oropharynx (16%) and hypopharynx (10%). Of the 37 patients with available PD-L1 combined positive score (CPS), 2 (5%), 23 (62%), and 12 (32%) had CPS values of 1, 1-19, and ≥20, respectively. The overall response rate was 54%, with 67% in the CPS ≥20 subgroup and 48% in the CPS 20 subgroup. The median progression-free survival (PFS) was 7. 2 months, with 10. 8 months for patients with CPS ≥20 and 5. 5 months for patients with CPS 20. The median overall survival (OS) was 18. 6 months. Among 62 patients with available WSIs, 47 (74%) and 15 (26%) patients were categorized as having an IP and NIP, respectively. Oral cavity cancer was enriched with IP (57. 4% of all IP cases vs 26. 7% of all NIP cases), while all four paranasal sinus tumors had NIP. The overall response rate was 57% in the IP group and 44% in the NIP group. Patients with IP had significantly longer PFS and OS than patients with NIP, both in unadjusted (hazard ratio HR for PFS, 0. 4; 95% confidence interval CI, 0. 21-0. 78; P = 0. 007; HR for OS, 0. 43; 95% CI, 0. 2-0. 95; P = 0. 036) and PD-L1 CPS-adjusted analyses (HR for PFS, 0. 41; 95% CI, 0. 21-0. 79; P = 0. 008; HR for OS, 0. 4; 95% CI, 0. 18-0. 89; P = 0. 025). High densities (≥25th percentile) of iTILs, sTILs, and tTILs were all significantly associated with prolonged PFS after CPS adjustment (P = 0. 001, 0. 001, and 0. 012). High density of iTILs was also significantly associated with improved OS (P = 0. 001 after CPS adjustment), while there was no significant association between OS and high density of sTILs or tTILs (0. 077 and 0. 085, respectively, after CPS adjustment). Conclusion: An AI-driven spatial TIL analyzer might serve as a simple and independent prognostic biomarker in HNSCC patients receiving pembrolizumab plus chemotherapy. Citation Format: Sehhoon Park, Junghoon Shin, Chan-Young Ock, Chang Ho Ahn, Hyun Ae Jung, Se-Hoon Lee, Myung-Ju Ahn. Artificial intelligence-powered spatial analysis of tumor-infiltrating lymphocytes as a prognostic biomarker for pembrolizumab plus chemotherapy in recurrent or metastatic head and neck squamous cell carcinoma 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 (6Suppl): Abstract nr 7658.