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

Background The aim is to investigate the risk factors for re-fracture of multisegmental osteoporotic vertebral compression fractures (OVCF) and to construct a clinical prediction model for their occurrence using machine learning algorithms. Objective The aim is to investigate the risk factors for re-fracture of multisegmental osteoporotic vertebral compression fractures (OVCF) and to construct a clinical prediction model for their occurrence using machine learning algorithms. Methods The CT images were imported into Slice-O-Matic software, and the fat infiltration rate of paraspinal and lumbar major muscles, paraspinal muscle mass, and lumbar major muscle mass were measured for each patient. The screening process was conducted through a multifaceted approach encompassing between-group comparison analysis, LASSO regression, and multivariate logistic regression. To ensure the robustness of the models, a total of 13 machine learning algorithms were employed in their construction. The prediction performance of each model was evaluated and the optimal model was selected through accuracy, recall, precision, F1 score and area under the receiver operating characteristic curve (AUROC). Finally, the models were interpreted using SHAP analysis to elucidate the importance of the features in the models and their impact on the direction of prediction. Conclusion In this study, the fat infiltration rate of paravertebral 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 XGBoost 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.