Mortality Prediction Modeling for Patients with Breast Cancer Based on Explainable Machine Learning

Area of Science:

• Oncology
• Data Science
• Medical Informatics

Background:

• Breast cancer is a leading cause of death in women globally.
• Strategic interventions are crucial to reduce breast cancer mortality rates.
• Real-world data analysis is vital for developing effective predictive models.

Purpose of the Study:

• To develop a predictive classification model for breast cancer mortality.
• To utilize real-world clinical data and machine learning for mortality prediction.
• To enhance model interpretability using explainable AI methods.

Main Methods:

• Analysis of 11,286 breast cancer patients' data from the National Cancer Center.
• Application of machine learning models, including extreme gradient boosting (XGB), on 31 clinical features.
• Utilizing SHapley Additive exPlanations (SHAP) for model interpretability.

Main Results:

• The XGB model demonstrated high discriminatory power (AUC 0.8722, specificity 0.9472).
• Key predictors identified include metastasis, age, N stage, T stage, radiation treatment, and Ki-67.
• The XGB model maintained strong performance (AUC 0.8518, specificity 0.9766) even after excluding patients with secondary cancers.

Conclusions:

• Developed models accurately predict breast cancer mortality using South Korean real-world data.
• Explainable AI (SHAP) confirmed the clinical applicability and interpretability of the predictive models.
• The findings support the use of AI in personalized breast cancer management and mortality risk assessment.