An interpretable machine learning model for predicting brain metastasis in breast cancer

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Breast cancer is a leading global malignancy.
• Brain metastasis significantly worsens breast cancer patient prognosis.
• Accurate prediction of brain metastasis is crucial for effective clinical management.

Purpose of the Study:

• To develop and validate a machine learning model for predicting breast cancer brain metastasis risk.
• To create a clinically applicable tool to assist in patient management and early detection.
• To enhance personalized risk stratification for breast cancer patients.

Main Methods:

• Utilized logistic regression for variable screening and eight machine learning algorithms for model construction.
• Trained and validated the model on large patient cohorts (N=154,193 training, N=66,084 internal validation, N=765 external validation).
• Employed AUC, AUPRC, DCA, calibration curves, and SHAP analysis for performance evaluation and interpretability.

Main Results:

• Identified higher tumor grade, advanced T/N stage, and PR positivity as key risk factors.
• Radiotherapy, chemotherapy, surgery, and specific subtypes (HR+/HER2-, HER2+) were protective factors.
• The XGBoost model demonstrated high predictive performance (AUCs up to 0.99) and clinical utility via a web calculator.

Conclusions:

• Developed an interpretable and deployable XGBoost model for breast cancer brain metastasis risk prediction.
• The web-based calculator facilitates clinical application for personalized risk stratification and early screening.
• The model aids in optimizing resource allocation and improving management strategies for breast cancer brain metastasis.