Breast cancer survival prediction using machine learning and multimodal data for personalized care plan

Area of Science:

• Oncology
• Biostatistics
• Machine Learning

Background:

• Accurate breast cancer survival prediction is crucial for personalized treatment planning.
• Existing models may not fully capture complex, time-dependent patient and tumor characteristics.

Purpose of the Study:

• To develop and validate time-stratified breast cancer survival prediction models using machine learning.
• To identify key predictors of long-term survival using explainable AI techniques.

Main Methods:

• Utilized multimodal data from 3,476 breast cancer patients with 43 features.
• Applied six machine learning algorithms (XGBoost, Logistic Regression, Random Forest, Decision Tree, AdaBoost, Multilayer Perceptron) for survival rate prediction.
• Employed Shapley Additive exPlanations (SHAP) for model interpretability and feature importance analysis.

Main Results:

• XGBoost demonstrated superior performance with over 90% accuracy.
• Top predictors for longer survival included lower DCIS, lower Ki-67, lower lymph node/tumor grade, higher age, and longer breastfeeding duration.
• SHAP analysis provided detailed insights into factors influencing survival across different time points.

Conclusions:

• Machine learning, particularly XGBoost, offers a powerful tool for accurate breast cancer survival prediction.
• Explainable AI (SHAP) reveals critical clinical, molecular, and demographic factors that can guide personalized treatment strategies.
• The study provides valuable insights for improving patient care and outcomes in breast cancer management.