Although mammography remains the cornerstone of breast cancer screening, its diagnostic performance is influenced by factors such as breast density and tumor characteristics. Blood-based biomarkers may therefore provide complementary biological information to imaging-based detection strategies. While circulating microRNAs (miRNAs) have emerged as promising minimally invasive biomarkers, their diagnostic performance and associations with clinicopathological characteristics remain incompletely understood. This multicenter study aimed to develop and validate a serum miRNA–based diagnostic model for breast cancer and to evaluate its associations with clinical stage, histological grade, and imaging-related factors.

In this multicenter observational case-control study, serum samples from 345 patients with breast cancer and 373 cancer-free controls were analyzed using next-generation sequencing–based miRNA profiling. After normalization and batch correction, 114 stably expressed miRNAs were selected for model construction. An ensemble machine learning model integrating five algorithms (Lasso, Ridge, support vector machine, Nu-SVM, and histogram-based gradient boosting) was developed using a training cohort (n = 569) and evaluated in a held-out test cohort (n = 149). Model performance was assessed using receiver operating characteristic curve analysis and the area under the curve (AUC) with 95% confidence intervals calculated by the DeLong method. Model-derived prediction scores were further evaluated in relation to clinicopathological and imaging-related variables.

The serum miRNA–based diagnostic model demonstrated promising discrimination between breast cancer cases and cancer-free controls, with an AUC of 0.875 (95% CI, 0.820–0.930) in the held-out test set. Model-derived prediction scores increased with advancing clinical stage and showed an ordered trend across histological grades. In exploratory analyses, prediction scores tended to increase across higher clinical T categories and were higher among patients with distant metastasis, whereas no clear differences were observed according to mammographic category, breast density, or intrinsic tumor subtype.

In this multicenter case-control study, a serum miRNA–based ensemble model demonstrated promising diagnostic discrimination for breast cancer. The model-derived prediction score was associated with clinical stage and histological grade, suggesting that circulating miRNA profiles may provide biologically relevant information beyond binary case-control discrimination. These findings support further investigation of serum miRNA profiling as a potential adjunctive diagnostic approach; however, prospective validation in representative screening populations is required before clinical implementation.

Not applicable.