Computational pathology-informed immune biomarker for trastuzumab benefit in HER2+ breast cancer: Validation in NSABP B-41 clinical trial.

595 Background: Trastuzumab, a targeted therapy, considerably improves survival outcomes in HER2+ breast cancer patients (pts). However, identifying pts most likely to benefit from trastuzumab and those who might avoid it, remains a challenge. Data from the NSABP B-41 randomized clinical trial which evaluated lapatinib-containing regimens against trastuzumab, each given concurrently with chemotherapy in the neoadjuvant setting, holds potential for computational pathology to uncover treatment response patterns. In this study, we present DeSTIL (Density and Spatial Architecture of Tumor Infiltrating Lymphocytes [TILs]), a predictive biomarker derived from Hematoxylin and Eosin (H&E) images, to evaluate the immune microenvironment and identify HER2+ pts that benefit from trastuzumab. Methods: Digitized and quality-controlled H&E slides from HER2+ pts in The Cancer Genome Atlas (TCGA; n=175) were used to develop DeSTIL and was independently validated on NSABP B-41 (n=221) pts. Following nuclei segmentation, TILs were identified, and their density and spatial architecture features were quantified. A lasso-regularized Cox proportional hazards model was used to select features and compute a continuous DeSTIL risk score, which was then dichotomized at the median into DeSTIL-positive and DeSTIL-negative groups in the training set. The locked model was validated on NSABP B-41 to predict event-free survival (EFS) in pts receiving neoadjuvant chemotherapy with trastuzumab, lapatinib, or a combination of trastuzumab and lapatinib. Within the DeSTIL-stratified groups, treatment-specific progression was analyzed to evaluate the potential benefit of trastuzumab-based therapies. Results: Among 221 HER2+ pts from the NSABP B-41 trial, 61 pts (28%) were classified as DeSTIL-positive and 160 (72%) as DeSTIL-negative, based on the median training threshold. DeSTIL-positive pts demonstrated a significant benefit with the trastuzumab-alone arm compared to the combination regimen (HR=0.09, 95% CI=0.01-0.77, p=0.0061) (interaction term p=0.024) and against lapatinib plus the combination regimen (HR = 0.11, 95% CI = 0.01-0.9, p = 0.01) (interaction term p = 0.05). In contrast, no significant benefit was observed in DeSTIL-negative pts when comparing trastuzumab to the combination regimen (HR=1.33, 95% CI=0.47-3.75, p=0.5840) or to lapatinib plus the combination regimen (HR=1.01, 95% CI=0.45-2.30, p=0.9701). Conclusions: DeSTIL, a biomarker based on the density and spatial architecture of TILs, may help identify HER2+ pts more likely to benefit from trastuzumab. Further validation through prospective trials is warranted. Additionally, this biomarker offers a practical framework for comparing HER2-targeted therapies, with the potential to minimize unnecessary treatments, reducing associated cardiotoxicity and financial burden.