Abstract PS2-08-26: Pathologist- and artificial intelligence-based TILs assessment in patients with early triple-negative breast cancer treated with neoadjuvant chemo-immunotherapy: real-world evidence from a nationwide cohort

Abstract Background: Neoadjuvant chemo-immunotherapy is the standard of care for patients with stage II-III triple negative breast cancer (TNBC). Real-world data on tumor-infiltrating lymphocytes (TILs) indicate a correlation of higher pre-treatment TILs levels with increased pCR rates. Digital pathology and artificial intelligence (AI) tools could provide value towards a more standardized and objective TILs assessment and additional spatial insights. In this study we aimed to assess the performance and prognostic value of both pathologist-based and digital pathology-based TILs scoring methods. Methods: The study population included a nationwide cohort of patients with early TNBC treated with neoadjuvant chemo-immunotherapy according to the KEYNOTE-522 regimen, in 20 Swedish hospitals between 2022-2024. TILs assessment on H&E-stained diagnostic biopsy tissue sections was performed i) manually (sTILs), based on the International Immuno-Oncology Working Group guidelines and ii) digitally (AI-TILs = TILs/Stromal cells). using the previously validated deep learning-based scoring algorithm HoverNet. Correlation with pCR was performed using univariate logistic regression model. Results: 337 patients were included in this nationwide cohort. 51.5% of the patients achieved pCR and 35% had RCB 2/3. Patients with baseline pathologist-based sTILs ≥30% were significantly associated with higher pCR rates and lower RCB score (chi-square test, p<0.0001). Both sTILs and AI-TILs were available for 84 patients and demonstrated a moderate correlation (Spearman’s rho = 0.59, p<0.0001). Both variables (continuous) were significantly associated with increased pCR rates (OR = 1.068, 95% CI: 1.04-1.11, p<0.001 for sTLs; OR = 1.072, 95% CI: 1.03-1.12, p<0.0001 for AI-based TILs). In non-lymphocyte predominant breast cancer (non-LPBC; sTILs <50%), patients with high AI-TILs (cut-off: median) were associated with higher likelihood of achieving pCR compared to those with low AI-TILs abundance (OR = 4.1; 95% CI: 1.3-14.1, p = 0.018) Conclusions: In this nationwide study, both pathologist- and AI-based TILs were predictive for pCR. AI-TILs assessment could potentially contribute to a better response discrimination in patients with non-LPBC treated with neoadjuvant chemo-immunotherapy. Validation in additional clinical cohorts is warranted. Citation Format: I. Zerdes, A. Papakonstantinou, B. Acs, N. Tsiknakis, S. Steen, E. Karlsson, X. Liu, K. Wang, S. Agartz, G. Manikis, J. Bergh, J. Hartman, T. Foukakis. Pathologist- and artificial intelligence-based TILs assessment in patients with early triple-negative breast cancer treated with neoadjuvant chemo-immunotherapy: real-world evidence from a nationwide cohort [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS2-08-26.