Predicting osteoporotic fractures post-vertebroplasty: a machine learning approach with a web-based calculator

Area of Science:

• Spinal Surgery
• Medical Imaging
• Machine Learning

Background:

• Osteoporotic vertebral compression fractures (OVCF) pose a significant risk after percutaneous vertebroplasty (PVP).
• Accurate prediction of re-fracture risk is crucial for patient management and personalized treatment strategies.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting new OVCF risk in patients post-PVP.
• To create a user-friendly, web-based calculator for practical clinical application of the predictive model.

Main Methods:

• Retrospective analysis of patients undergoing PVP between June 2016 and June 2018.
• Independent variables screened using Boruta, modeled with 9 algorithms; performance assessed via ROC-AUC and DCA.
• Model interpretability analyzed using SHAP; deployment via a web calculator.

Main Results:

• Support Vector Machine (SVM) model demonstrated the best performance with an AUC of 0.77 in both training and validation groups.
• Clinical decision curve analysis (DCA) indicated the model's benefit for patients in both training and test sets.
• A web calculator based on the SHAP-enhanced SVM model is available for clinical risk assessment.

Conclusions:

• The SVM-based ML model effectively predicts new-onset OVCF risk following PVP.
• The developed web calculator serves as a practical tool for clinical decision-making.
• This work advances personalized care in spinal surgery by providing a risk prediction tool.