To develop a machine learning-based predictive model for accurately assessing the risk of bone metastasis in breast cancer patients.
Approach:
Data Source: Utilized the Surveillance, Epidemiology, and End Results database for model development.
Model Development: Conducted univariate and multivariate logistic regression analyses to identify key predictive variables, followed by constructing eight machine learning algorithms.
Model Evaluation: Employed 10-fold cross-validation for hyperparameter optimization and evaluated model performance using AUC, AUPRC, decision curve analysis, and calibration curves.
Interpretability: Applied SHAP analysis to enhance model interpretability and developed a web-based calculator for clinical application.
Key Findings:
Majority of patients were aged over 50 years, female, with HR+/HER2− subtype, and low incidence of bone metastasis.
Key independent risk factors identified included age >50 years, higher tumor grade, advanced T stage, N stage, clinical stage, HR-/HER2- subtype, radiotherapy, and married status.
The LGB model achieved an AUC of 0.98 in training and internal validation sets, and 0.91 in the external validation set.
SHAP analysis indicated surgery as the primary protective factor, with advanced N stage associated with increased bone metastasis risk.
Interpretation:
The developed LGB model and web-based calculator provide tools for assessing the risk of bone metastasis in breast cancer patients.
Limitations:
Potential selection bias due to reliance on public databases.
Limited generalizability and external validity due to lack of diverse clinical settings.
Conclusion:
The study presents a transparent machine learning model that can aid in the early detection and management of bone metastasis in breast cancer patients.
