Five-miRNA signature for breast cancer detection in a northern Mexico border cohort.

Breast cancer remains the leading malignancy diagnosed in women worldwide and is of particular concern in low- and middle-income countries, such as Mexico, where mortality rates are strongly associated with late diagnosis. While the diagnostic performance of circulating miRNA expression panels has been evaluated globally, their application in the Mexican population remains largely unexplored. This study evaluated the discriminatory capacity of a circulating 5-miRNA expression panel in a cohort of Mexican patients with confirmed histopathological diagnoses (11 breast cancer, 23 reference individuals). Expression levels of miR-21, miR-155, miR-195, miR-222-3p, and miR-10b were quantified by RT-qPCR, and their discriminatory capacity was assessed using individual ROC curves and machine learning algorithms. Individually, the miRNAs showed moderate discriminatory capacity, but a multivariate integration using Random Forest achieved the best individual model performance (AUC = 0.727). The final soft-voting ensemble model reported an AUC of 0.690 (sensitivity of 72.7% and specificity of 64.7%), successfully mitigating individual classifier biases and stabilizing predictions in a limited-sample scenario. These findings serve as an exploratory proof-of-concept for combining this miRNA expression panel with machine learning strategies in the Mexican population; however, its moderate performance strictly necessitates multicentric validation in a larger cohort before any clinical utility can be established.