Prediction of Lung Metastasis in Breast Cancer Patients Using Machine Learning Classifiers

Breast cancer is the most common cancer among women, and metastasis to the lung is associated with poor prognosis. Reliable biomarkers for predicting lung metastasis are urgently needed to improve early detection and clinical decision-making. This study used microarray data sets comprising gene expression profiles and clinical data from primary breast cancer patients who were followed up for lung metastasis outcomes. High-throughput screening combined with Venn diagram analysis was used to identify common candidate probes, and the least absolute shrinkage and selection operator method were used to select 11 genes for model development. Logistic regression was used to construct predictive models, and the final risk signature consisted of 10 candidate genes (CDK19, GLUD1, GTPBP4, HLCS, HYI, KCND3, MAP2K1, NMUR1, PRKD3, and SLC16A3). The model achieved strong performance in training and validation cohorts (areas under the curve >0.87) and generalized to the independent METABRIC data set (area under the curve = 0.706). Subset analyses restricted to patients with early-stage disease confirmed that the signature retained predictive value. Kaplan-Meier analyses showed that patients with high-risk scores had shorter lung metastasis-free survival, recurrence-free survival, and overall survival. Multivariate Cox analysis confirmed that the risk signature provided independent predictive information from clinical variables. In conclusion, the risk signature accurately identifies patients with breast cancer at risk of lung metastasis, enabling clinicians to better assess risk and tailor effective treatment strategies.