Abstract 6623: Spatial organization of tumor-infiltrating lymphocytes (TILs) is prognostic for survival in triple negative breast cancer (TNBC)

Abstract

Abstract Introduction: TILs quantification is a predictor of response to neoadjuvant chemotherapy and prognostic for long term outcomes in TNBC. TILs are typically quantified using standard hematoxylin/eosin (H&E) slides manually by trained pathologists; limitations include interobserver variability and need for cutoffs that aid in clinical decision making. Computational approaches utilizing H&E slides have potential to improve reproducibility and refine predictive/prognostic utility by evaluation of additional metrics beyond quantification. Methods: H&E whole slide images digitized at 20x magnification from patients with stage I-III TNBC (ER/PR ≤10%, HER2 negative) treated with neo/adjuvant chemotherapy at two institutions were utilized. All cell nuclei were automatically segmented using a deep learning model (Hover-Net) and classified as TIL or non-TIL based on morphological features. Features related to TIL density, spatial distribution, and morphological features were extracted. The top 3 features determined by least absolute shrinkage and selection operator were used to train a Cox proportional hazards regression model (SpaTILs) that assigned a recurrence free survival (RFS) event risk score to each patient. First quartile risk score in the training cohort was used as a cutoff for SpaTILs high vs low risk. Model performance for prediction of RFS and overall survival (OS) was evaluated in the testing cohort by Kaplan Meier method, with SpaTILs high vs low risk groups compared by log rank test and Cox regression analysis. Results: In the training cohort (n=26) and testing cohort (n=62), median age was 51 and 53 years, 77% and 36% were node positive, and 81% and 45% received neoadjuvant chemotherapy, respectively. Median follow up in testing and training cohorts was 2.3 and 6.8 y, respectively. In the testing cohort, 31% and 69% of patients were categorized as SpaTILs high and low risk, respectively. Baseline characteristics were not significantly different between SpaTILs high and low risk patients. In the SpaTILs high and low risk groups, 5y RFS was 53% and 90%, respectively (HR 4.85 [95% CI: 1.62-14.52], p=0.002) and 5y OS was 78% and 98%, respectively (HR 13.91 [95% CI 1.67-115.66], p=0.001). SpaTIL high risk remained a significant predictor for RFS and OS after adjusting for T stage and nodal status: HR 7.25 (95% CI 2.14-24.39), p=0.001 for RFS; HR 4.80 (95% CI 1.31-131.06), p=0.028 for OS. Conclusion: The computational model based on TILs density, spatial, and morphological features (SpaTILs) was independently prognostic for RFS and OS in early stage TNBC. This model is being validated in larger TNBC cohorts and warrants prospective evaluation in clinical trials. Citation Format: Sahil H. Patel, Germán Corredor, Rachel Yoder, Cristian Barrera, Miluska Castillo, Luis Bernabe, Joshua Staley, Shane Stecklein, Satish E. Viswanath, Carlos A. Castaneda, Priyanka Sharma, Anant Madabhushi. Spatial organization of tumor-infiltrating lymphocytes (TILs) is prognostic for survival in triple negative breast cancer (TNBC) [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 6623.