Abstract PS3-10-30: A clinical model to predict pathologic complete response using early peripheral absolute lymphocyte dynamics in HER2+ breast cancer

Abstract Introduction: Immune activation and lymphocyte infiltration into the tumor microenvironment are known to predict treatment response in HER2-positive (HER2+) breast cancer. Our work has shown that dynamic changes in peripheral absolute lymphocyte count (ALC) following initial neoadjuvant therapy (NAT) are also associated with a higher likelihood of achieving pathologic complete response (pCR). In this study, we aimed to develop and validate a predictive model for pCR based on clinically obtained pre-treatment ALC and ALC following a single dose of NAT, with the goal of creating a tool for early response prediction in HER2+ breast cancer. Methods: We assembled a retrospective discovery cohort of patients with HER2+ breast cancer treated with trastuzumab-based NAT from the Oncoshare database, which integrates electronic medical records and California Cancer Registry data from patients diagnosed between 2000 and 2020 at Stanford University. Our non-overlapping validation cohort consisted of patients with HER2+ breast cancer who underwent NAT on clinical trials of HER2-targeted therapy and/or who were enrolled in prospective institutional tissue collection studies. In the discovery cohort, we constructed a multivariable logistic regression model to identify predictors of pCR, including baseline ALC, early ALC change, days between baseline and on-treatment ALC draws, age at diagnosis, estrogen receptor (ER) status, stage, grade, and self-reported race/ethnicity. Early ALC change was defined as 1 minus the ratio of cycle 1 ALC to baseline ALC. To develop an ALC-based risk grouping strategy, we used logistic regression to create a composite risk score from baseline ALC and early ALC change, followed by receiver operating characteristic curve analysis. We then applied grid search optimization to test combinations of cutoff points across the 10th to 90th percentiles of the risk score distribution and identified the pair of cutoffs that maximized area under the curve (AUC) when patients were stratified into three risk categories. Model performance was evaluated using AUC analysis in both the discovery and validation cohorts. Results: In the discovery cohort (n = 154), higher baseline ALC and greater early ALC decline were independently associated with increased odds of pCR. The composite ALC risk score, using baseline ALC and early ALC change, stratified patients into three groups: low (25% of patients, n = 39), intermediate (45%, n = 69), and high (30%, n = 46) probability of pCR. Observed pCR rates in these groups were 15%, 51%, and 76%, respectively. When this ALC-based risk grouping was integrated with the other covariates (including time interval between ALC draws, age, ER status, clinical stage, tumor grade, and race/ethnicity), the full multivariable model demonstrated strong predictive performance with an area under the curve (AUC) of 0.81 (95% CI 0.74-0.88). In the separate validation cohort (n = 83), the composite ALC risk score stratified patients into low (30% of patients, n=25), intermediate (48% of patients, n=40), and high (22% of patients, n=18) probability of pCR groups, with observed pCR rates of 12%, 63%, and 78%, respectively. The model showed similar discriminative ability in the validation cohort, with an AUC of 0.81 (0.71-0.90). Conclusion: Early ALC dynamics, with no other covariates, effectively stratify patients with HER2+ breast cancer based on their likelihood to achieve pCR after NAT. We developed and validated a clincal model using routinely collected laboratory values, which achieved strong predictive performance in both discovery and validation cohorts. This ALC-based tool offers a cost-effective and scalable alternative for early response prediction, with potential to inform treatment adaptation strategies in real time. Citation Format: C. Bergstrom, I. Luo, E. Kotler, M. Satoyoshi, S. Philip, C. Curtis, A. Kurian, S. Han, J. Caswell-Jin. A clinical model to predict pathologic complete response using early peripheral absolute lymphocyte dynamics in HER2+ breast cancer [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 PS3-10-30.