Development and validation of a predictive model for secondary vertebral compression fractures based on paravertebral muscles

Abstract

Abstract Purpose Develop a predictive model for secondary vertebral compression fractures (SVCF) following percutaneous vertebroplasty (PVP) or percutaneous kyphoplasty (PKP) in osteoporotic vertebral compression fracture (OVCF) patients. Methods Retrospective analysis of 229 OVCF patients treated with PVP or PKP from September 2020 to September 2021. SVCF occurrence within 2 years postoperatively categorized patients into training (n = 114) and validation (n = 115) sets. Model 1 and Model 2 were constructed using Lasso regression and random forest analysis. Model comparison involved the area under curve (AUC), calibration, decision curve analysis (DCA), and Akaike information criterion (AIC). Internal validation used 1000 Bootstrap iterations with tenfold cross-validation. Results presented through a Nomogram on a web platform. Results Among 229 PVP/PKP-treated OVCF patients, 40 (17.47%) experienced SVCF. Model 1 outperformed Model 2 in AUC, calibration, DCA, and AIC, making it the selected predictive model. Logistic regression identified surgery type, duration, spinal CT value, and erector spinae muscles' standardized functional cross-sectional area as predictors. Model 1 demonstrated AUC of 0.847 (95% CI 0.749–0.945) in training and 0.805 (95% CI 0.693–0.917) in validation. At a Youden index of 0.62, sensitivity and specificity were 0.74 and 0.88, respectively. Internal validation for the training set: accuracy 0.839, kappa coefficient 0.228, AUC 0.813. Hosmer-Lemeshow tests indicated good discriminative ability for Model 1 in both sets. Clinical decision curves and Nomogram accessible at https://sofarnomogram.shinyapps.io/DynNomapp/. Conclusion This predictive model, demonstrating favorable accuracy, effectively assesses SVCF risk in post-PVP/PKP OVCF patients in clinical practice.