Impact of Tumor Location on Predicting Early-Stage Breast Cancer Patient Survivability Using Explainable Machine...

Area of Science:

• Oncology
• Machine Learning
• Biostatistics

Background:

• Early-stage breast cancer survivability prediction is crucial for treatment planning.
• Explainable machine learning (ML) models offer potential for enhanced prognostic insights.
• Understanding feature importance in survival prediction can refine clinical decision-making.

Purpose of the Study:

• To investigate the impact of tumor quadrant location on 5-year early-stage breast cancer survivability prediction.
• To utilize explainable ML models, including Shapley Additive Explanations (SHAP), for feature importance analysis.
• To identify significant factors influencing patient outcomes in early-stage breast cancer.

Main Methods:

• Trained six ML models (Xtreme Gradient Boosting, Random Forest, Logistic Regression, Decision Tree, Support Vector Machine, AdaBoost) on data from 401 early-stage breast cancer patients.
• Evaluated models using performance metrics such as AUC-ROC and AUC-PR.
• Employed feature importance, coefficient effect size, and SHAP values for model interpretation, focusing on tumor quadrant variables.

Main Results:

• The extreme gradient boosting model achieved superior performance with AUC-ROC of 0.98 and AUC-PR of 0.97.
• Tumor quadrant location, particularly upper outer and miscellaneous sites, emerged as a top predictive feature for breast cancer survivability.
• SHAP analysis confirmed the significant influence of tumor locations on survival outcomes.

Conclusions:

• Explainable ML models effectively predict 5-year early-stage breast cancer survivability.
• Tumor quadrant location is identified as an independent prognostic factor for breast cancer.
• SHAP value interpretation provides clinicians with actionable insights to refine treatment protocols and enhance patient outcomes.