About the Model
BreastOncoPredict is a clinical decision support tool based on ensemble machine learning methods for predicting breast cancer recurrence and mortality at 2- and 10-year horizons.
Methodology
The model uses a stacking ensemble architecture that combines four base learners — Random Forest, XGBoost, LightGBM, and Support Vector Machine — with a Logistic Regression meta-learner. This approach leverages the complementary strengths of diverse algorithms to achieve robust predictions.
Input Features
16 clinical & hematological features
Base Learners
Random Forest
Bagging
XGBoost
Boosting
LightGBM
Boosting
SVM
Kernel
Meta-Learner
Logistic Regression
Stacking
2-Year Prediction
10-Year Prediction
Model Performance
Area Under the ROC Curve (AUC) across molecular subtypes:
| Subtype | AUC (2yr) | Sensitivity (2yr) | Specificity (2yr) | AUC (10yr) | Sensitivity (10yr) | Specificity (10yr) |
|---|---|---|---|---|---|---|
| General | 0.859 | 67.0% | 88.0% | 0.814 | 59.0% | 86.0% |
| HR+ (Luminal) | 0.862 | 58.0% | 95.0% | 0.804 | 59.4% | 84.3% |
| HER2+ | 0.892 | 69.0% | 96.0% | 0.831 | 66.7% | 83.9% |
| TNBC | 0.834 | 66.0% | 84.1% | 0.829 | 80.6% | 68.8% |
Reference
Moreira PH, et al. "An Ensemble-Based Machine Learning Approach to Predict 2- and 10-Year Breast Cancer Recurrence or Mortality based on Clinical, Pathological, and Hemogram Data at Diagnosis." ICESP/LIM24, Faculdade de Medicina da Universidade de São Paulo, Brazil.