Abstract
There are few tools support postmastectomy radiotherapy (PMRT) decisions for triple-negative breast cancer (TNBC) patients. This study aimed to develop a whole slide image (WSI)-based risk score to identify PMRT beneficiaries. We analyzed 608 TNBC patients treated at West China Hospital, Sichuan University (WCH) and 182 patients from The Cancer Genome Atlas (TCGA), diagnosed between 2011 and 2019. Utilizing features extracted from hematoxylin and eosin-stained WSI and clinicopathological characteristics, an image score and a clinicopathological score for each patient were computed via eXtreme Gradient Boosting, combined into a comprehensive risk score. Multivariate Cox proportional hazards regression analyses revealed a higher image score strongly associated with worse 5-year invasive disease-free survival (iDFS) (HR = 8.80, P < 0.001), overall survival (OS) (HR = 9.02, P < 0.001), and locoregional recurrence-free survival (LRFS) (HR = 8.39, P = 0.003). The integrated risk score outperformed clinicopathological score in prognostic accuracy, as evidenced by the area under the curve (WCH validation cohort, iDFS: 0.805 vs. 0.756; OS: 0.874 vs. 0.807; LRFS: 0.957 vs. 0.870; TCGA cohort, iDFS: 0.683 vs. 0.598; OS: 0.685 vs. 0.617). Importantly, PMRT significantly improved OS in "high-risk" patients identified by the risk score, but not in "low-risk" patients in both WCH and TCGA cohorts. These findings suggest that the WSI-based risk score shows promise as a tool for prognostic assessment and guiding PMRT decision-making in TNBC patients.