Abstract
BACKGROUND: Ductal carcinoma in situ (DCIS) carries a significant risk of postoperative upgrading to invasive breast cancer (IBC), yet existing prediction models lack validation in Asian populations. This study aimed to develop and validate a population-specific nomogram to preoperatively predict DCIS-to-IBC upgrading in Asian patients. METHODS: A multicenter retrospective cohort of 465 Asian women diagnosed with DCIS by core needle biopsy (2015-2021) was analyzed. Patients were randomly divided into training (n = 257), internal validation (n = 110), and external validation cohorts (n = 98). Predictors were selected via LASSO regression and multivariable logistic regression. Model performance was assessed using AUC, calibration curves, and decision curve analysis (DCA). An interactive online nomogram was developed for clinical application. RESULTS: Postoperative upgrading occurred in 49.46% (230/465) of patients. Four independent predictors were identified: palpable mass (OR = 2.55, p = 0.096), lesion palpability (OR = 2.58, p = 0.043), low nuclear grade (OR = 0.55, p = 0.098), and suspected invasion (OR = 6.59, p < 0.001). The nomogram demonstrated robust discrimination in the training cohort (AUC = 0.802, 95% CI 0.748-0.856), with maintained performance in internal validation (AUC = 0.753) and acceptable generalizability in external validation (AUC = 0.680). DCA confirmed clinical utility across risk thresholds. The dynamic nomogram ( https://duancl777.shinyapps.io/dynnomapp/ ) enabled real-time risk stratification. CONCLUSIONS: The DCIS-IBC Guide Board is the first Asian-specific model integrating clinicopathological predictors to identify high-risk DCIS patients. It facilitates personalized decisions, such as omitting sentinel lymph node biopsy while reducing overtreatment. Although external validation showed moderate performance, this tool addresses critical population heterogeneity and enhances preoperative risk assessment. Prospective multicenter studies are warranted to optimize generalizability and explore multimodal predictors.