Application Study of Interpretable Machine Learning Models for Predicting Postoperative Refracture After Vertebral Augmentation in Osteoporotic Vertebral Compression Fractures

<title>Abstract</title> <p>Background Osteoporotic vertebral compression fractures (OVCFs) are severe osteoporosis complications; vertebral augmentation is the preferred minimally invasive treatment, but postoperative refracture risk exists. Traditional logistic regression fails to accurately assess individual risks, while interpretable machine learning (ML) excels in high-dimensional data processing, with limited relevant studies. Purposes: To develop an interpretable ML prediction model for identifying risk factors of subsequent fractures after vertebral augmentation in patients with OVCFs. Methods A retrospective analysis was conducted on clinical data of 1,502 OVCF patients who underwent vertebral augmentation. Thirty-six characteristic indicators were extracted from electronic medical records and imaging systems. Six ML prediction models were constructed. Prediction performance was comprehensively evaluated using receiver operating characteristic (ROC) curves, accuracy, recall, F1 score, precision, calibration curves, and decision curve analysis. The optimal model was interpreted globally and locally via Shapley Additive exPlanations (SHAP) to analyze the contribution of key features. Results The 2-year post-operative subsequent fracture incidence in the study cohort was 9.65% (145 cases). After data preprocessing and model training, the extreme gradient boosting (XGBoost) model demonstrated the best performance on the test set. Calibration curve and decision curve analyses showed high consistency between predicted results and actual risks, with significant clinical net benefit. SHAP analysis identified nine key risk factors ranked by importance: age, bone cement leakage and types, history of osteoporosis, Previous history of fractures, bone mineral density, thoracolumbar fascitis, types of trauma, duration of surgery, and Braden score. Conclusions The XGBoost model combined with SHAP represents an effective tool for predicting subsequent fracture risk after vertebral augmentation in OVCF patients. Clinical application of this prediction model can assist clinicians in formulating individualized intervention strategies, thereby optimizing treatment protocols and post-operative management to reduce post-operative subsequent fracture incidence.</p>