Abstract 2484: Association of artificial intelligence (AI)-based HER2 scoring with clinical outcomes to first-line trastuzumab plus chemotherapy in HER2-positive metastatic gastric cancer (mGC)

Abstract

Abstract Background: Although trastuzumab plus chemotherapy is the treatment of choice for HER2-positive mGC, not all patients respond to the therapy. To date, there is no reliable and efficient biomarker that can predict the response to trastuzumab-based treatment. Methods: Clinical data and pathology slides from HER2-positive mGC patients treated with first-line trastuzumab and capecitabine/5-fluorouracil plus cisplatin were retrospectively collected. HER2 positivity was confirmed using standard methods with immunohistochemistry (IHC) and/or in situ hybridization. An AI-based model was applied to a single HER2 IHC-stained whole slide image (WSI) of each patient to identify and quantify the tumor cells with HER2 IHC staining intensity categories of 0, 1+, 2+, or 3+ on the slide. The proportion of tumor cells with 3+ staining intensity among total tumor cells (h3 percentage) in each slide was calculated and analyzed in relation to clinical outcomes. Results: A total of 160 cases were included in this analysis. Overall, the median progression-free survival (PFS) was 9.2 mo, and the overall survival (OS) was 18.9 mo. By standard IHC HER2 grading, 107 (66.9%) 3+ and 53 (33.1%) 2+ cases were included. The median AI-estimated h3 percentage in these patients was 14.8% (range 0 - 99.5%). The patients with their best tumor response of complete (CR) or partial response (PR) (44.4%) exhibited significantly higher h3 percentages (mean h3 percentage 46.6% in CR/PR vs. 29.4% in stable or progressive disease, p = 0.018). The h3 percentage exhibited strong associations as a continuous variable with longer PFS (hazard ratio [HR] 0.94 [95% confidence interval (CI) 0.90 - 0.99] for 10% change of h3 percentages, p = 0.015) or OS (HR 0.93 [95% CI 0.89 - 0.98] for 10% change of h3 percentages, p = 0.004). Notably, patients having a proportion of 3+ tumor cells at ≥ 90% (27/160, 16.9%) showed particularly longer PFS (median PFS 21.4 mo in ≥ 90% vs. 8.5mo in < 90%, p = 0.006) and OS (median OS 40.3 mo in ≥ 90% vs. 16.6 mo in < 90%, p = 0.004). Conclusion: AI-based models interpreting WSI can provide additive predictive information to HER2-positive mGC patients treated with trastuzumab plus chemotherapy. Citation Format: Minsu Kang, Yoojoo Lim, Hyeon Jeong Oh, Ji Ae Lee, Jin Won Kim, Ji-Won Kim, Sanghoon Song, Heon Song, Keun-Wook Lee. Association of artificial intelligence (AI)-based HER2 scoring with clinical outcomes to first-line trastuzumab plus chemotherapy in HER2-positive metastatic gastric cancer (mGC) [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 2484.