Machine learning-based prognostic modeling and surgical value analysis of de novo metastatic invasive ductal carcinoma of the breast.

Abstract

Whether primary lesion surgery improves survival in patients with de novo metastatic breast cancer (dnMBC) is inconclusive. We aimed to establish a prognostic prediction model for patients with de novo metastatic breast invasive ductal carcinoma (dnMBIDC) based on machine learning algorithms and to investigate the value of primary site surgery. The data used in our study were obtained from the Surveillance, Epidemiology, and End Results database (SEER, 2010-2021) and the First Affiliated Hospital of Nanchang University (1st-NCUH, June 2013-June 2023). We used COX regression analysis to identify prognostic factors. We divided patients into training and validation groups and constructed Extreme Gradient Boosting (XGBoost) prognostic prediction model. In addition, we used propensity score matching (PSM), K-M survival analysis, and COX regression analysis to explore the survival benefit of patients undergoing primary lesion surgery. A total of 13,383 patients were enrolled, with 13,326 from SEER and 57 from 1st-NCUH. The results showed that XGboost had good predictive ability (training set C-index = 0.726, 1 year AUC = 0.788, 3 year AUC = 0.774, 5 year AUC = 0.774; validation set C-index = 0.723, 1 year AUC = 0.785.1, 3 year AUC = 0.770, 5 year AUC = 0.764), which has better predictive power than the Coxph model. We used Shiny-Web to make our model easily available. Furthermore, we found that surgery was associated with a better prognosis in dnMBIDC patients. Based on the XGboost, we can accurately predict the survival of dnMBIDC patients, which can provide a reference for clinicians to treat patients. In addition, surgery may bring survival benefits to dnMBIDC patients.