Modelling of multisegmental osteoporotic vertebral compression fracture using machine learning to analyse and predict risk factors

To identify independent risk factors for multisegmental osteoporotic vertebral compression fractures (OVCF) and develop an interpretable machine-learning model for individualized risk prediction. This multicenter retrospective study included 298 eligible patients selected from 1,632 patients with OVCF treated at three tertiary hospitals. Clinical characteristics, laboratory parameters, and CT-derived muscle measurements were obtained. LASSO and multivariable logistic regression analyses were used to identify independent predictors. Thirteen machine-learning algorithms were developed and internally validated using a 7:3 training-validation split. The model performance was evaluated using the AUROC, accuracy, precision, recall, and F1 score. SHAP analysis was performed to improve the model interpretability. In this study, the fat infiltration rate of paraspinal muscle and total type I collagen amino-terminal extender peptide were identified as potential risk factors for the development of multisegmental OVCF. A prediction model for the occurrence of multisegmental OVCF was constructed by the NB model, and the model was evaluated to have good predictive performance. SHAP analysis was utilised to enhance the interpretability of the model, thereby demonstrating the importance of paraspinal muscle fat infiltration rate and total amino-terminal-propeptide of type I collagen in the prediction of the model. PMFIR and T-P1NP are independent predictors of multisegmental OVCF. The proposed interpretable Naïve Bayes model demonstrated favorable predictive performance and may facilitate the early identification of high-risk patients and support individualized clinical decision-making. Future multicenter prospective studies are warranted to externally validate and further optimize this model.