Clinical Report: Creation and assessment of a transparent machine learning predictive model for bone metastasis in breast cancer patients
Overview
This study developed a machine learning predictive model to assess the risk of bone metastasis in breast cancer patients. The model demonstrated high accuracy and was validated using real-world data.
Background
Breast cancer is a leading malignancy globally, with bone metastasis being the most common form of distant spread. Early prediction of bone metastasis is crucial for timely interventions.
Data Highlights
Model
AUC (Training Set)
AUC (Internal Validation)
AUC (External Validation)
AUPRC (Training Set)
AUPRC (Internal Validation)
AUPRC (External Validation)
LGB Model
0.98
0.98
0.91
0.96
0.79
0.87
Key Findings
The LGB model achieved an AUC of 0.98 in both the training and internal validation sets.
Key independent risk factors for bone metastasis included age >50 years, higher tumor grade, and advanced clinical stage.
SHAP analysis indicated surgery as a protective factor against bone metastasis.
Decision curve analysis showed net clinical benefit within the 0.1-0.8 threshold range.
The model was validated on 342 real-world cases from an independent hospital cohort.
Clinical Implications
The development of this machine learning model provides a tool for assessing the risk of bone metastasis in breast cancer patients.
Conclusion
The study presents a robust machine learning model for predicting bone metastasis in breast cancer, enhancing early detection and risk stratification.
Dana-Farber Cancer Institute's Dr. Jennifer Ligibel shared updated results from the BWEL study showing a weight loss intervention led to significantly better physical function, global physical and mental health, and symptom in patients with early breast cancer.
