Clinical Report: A Predictive Machine Learning Framework for Osteoporotic Fragility Fracture Risk
Overview
This study presents a machine learning framework for predicting osteoporotic fragility fractures using clinical data from hospitalized patients in Southern China. The LightGBM model achieved a training set ROC-AUC of 0.903 and a test set ROC-AUC of 0.840.
Background
Osteoporotic fragility fractures are a significant public health concern, leading to increased hospitalization and economic burden. Accurate prediction of fracture risk is essential for timely intervention. Traditional methods like FRAX have limitations.
Data Highlights
Model
Training Set ROC-AUC
Test Set ROC-AUC
LightGBM
0.903
0.840
Key Findings
1125 hospitalized patients with osteoporotic fractures were analyzed.
73 clinical, laboratory, and radiological features were collected for analysis.
LightGBM outperformed other machine learning algorithms in predictive accuracy.
Age, Hepatitis C virus IgG antibody, and serum sodium were identified as significant risk factors based on model analysis.
SHAP analysis was utilized to explain the model's predictions.
Clinical Implications
Identifying key risk factors can facilitate targeted interventions for at-risk individuals.
Conclusion
The developed predictive model demonstrates strong predictive ability for identifying patients at risk for osteoporotic fragility fractures.
