Antibody-drug conjugate (ADC) after ADC: Predicting survival in patients with HER2-low metastatic breast cancer with lung metastasis using artificial intelligence (AI) and radiomics.

Abstract

1027 Background: With the approval of multiple antibody drug conjugates (ADCs) for HER2-low metastatic breast cancer (MBC), defining which patients (pts) will benefit from an ADC after ADC approach is essential. No reliable biomarker exists to predict overall survival (OS) in the ADC after ADC setting. This study investigates whether AI informed radiomic texture features derived from CT scans prior to ADC therapy in pts with lung metastases who received both trastruzumab deruxtecan (T-DXd) and sacituzmab govitecan (SG) are associated with OS in HER2-low MBC. Methods: A single-institution retrospective study was conducted including pts with HER2-low MBC treated with both T-DXd and SG at Winship Cancer Institute of Emory University between 2020 and 2024. Among 44 pts reviewed (median age, 55 years), 21 pts with evaluable lung lesions were included. Radiomic texture features characterizing tumor heterogeneity were extracted from baseline pre-treatment CT scans of lung metastases before ADC initiation. A least absolute shrinkage and selection operator Cox regression model was applied within a 50-fold cross-validation framework to identify prognostic features associated with OS. OS was defined as the interval from initiation of first ADC therapy to death from any cause and was censored at the date of last follow-up for patients who were alive. A radiomic risk score (RRS) was calculated as a linear combination of the selected features weighted by their corresponding coefficients. Pts were stratified into high- and low-risk groups based on the median RRS. Cox proportional hazards regression was used to evaluate associations with OS. Results: Among the total 44 pts, the median duration of response was 6.1 months for the first ADC and 2.7 months for the subsequent ADC. 59.1% received SG as ADC1. For the 21 pts with lung metastasis, mOS was 20.8 months (95% CI: 14.33–30.8). In univariable analysis, the RRS was significantly associated with OS (HR = 2.5, 95% CI: 1.3–4.8; P = 0.006). In multivariable analysis adjusting for clinicopathologic factors including race, ER status, number of prior lines of therapy before ADC treatment, and relapsed vs de novo metastatic disease, RRS remained the only variable independently associated with OS (HR = 2.6, 95% CI: 1.2–5.5; P = 0.01). Conclusions: These preliminary findings suggest that CT-derived radiomic features may serve as a non-invasive biomarker to identify pts at higher risk of poor OS in the ADC after ADC setting for HER2-low MBC, for whom we need better, novel therapeutics after resistance develops to initial ADC therapy. This radiomics approach has the potential to inform treatment sequencing and risk stratification. Future work will include prospective validation in a larger cohort which includes other sites of metastatic disease.