Identification and validation of screening models for breast cancer with 3 serum miRNAs in an 11,349 samples mixed cohort.

Abstract

The study focuses on enhancing breast cancer (BC) prognosis through early detection, aiming to establish a non-invasive, clinically viable BC screening method using specific serum miRNA levels. Involving 11,349 participants across BC, 11 other cancer types, and control groups, the study identified serum biomarkers through feature selection and developed two BC screening models using six machine learning algorithms. These models underwent evaluation across test, internal, and external validation sets, assessing performance metrics like accuracy, sensitivity, specificity, and the area under the curve (AUC). Subgroup analysis was conducted to test model stability. Based on the three serum miRNA biomarkers (miR-1307-3p, miR-5100, and miR-4745-5p), a BC screening model, SM4BC3miR model, was developed. This model achieved AUC performances of 0.986, 0.986, and 0.939 on the test, internal, and external sets, respectively. Furthermore, the SSM4BC model, utilizing ratio scores of miR-1307-3p/miR-5100 and miR-4745-5p/miR-5100, showed AUCs of 0.973, 0.980, and 0.953, respectively. Subgroup analyses underscored both models' robustness and stability. This research introduced the SM4BC3miR and SSM4BC models, leveraging three specific serum miRNA biomarkers for breast cancer screening. Demonstrating high accuracy and stability, these models present a promising approach for early detection of breast cancer. However, their practical application and effectiveness in clinical settings remain to be further validated.