Paravertebral Muscle Fat Infiltration and Other Imaging Predictors of Cement Nonunion Following Percutaneous Kyphoplasty: A Multivariable Modeling Study

Objective To identify risk factors for postoperative cement nonunion in patients with OVCF and to develop prediction models for clinical risk stratification. Methods A retrospective cohort of 237 patients with primary OVCF was analyzed, including 168 patients without cement nonunion and 69 with cement nonunion. Group comparisons were performed using the chi-square or Fisher’s exact test for categorical variables and the Mann–Whitney U test for continuous variables after normality testing. Multivariable logistic regression was performed using nine candidate predictors, with events-per-variable and variance inflation factors assessed. To enhance robustness, L1- and L2-regularized logistic regression models were developed and compared with support vector machine (SVM), random forest, and gradient boosting. Model performance was evaluated using pooled predictions from stratified 10-fold cross-validation. Calibration was assessed using calibration plots and the Brier score. Results Multivariable logistic regression identified lower vertebral body CT value (OR 0.04, 95% CI 0.007–0.25), greater Cobb angle change (OR 2.62, 95% CI 1.16–5.94), cement leakage (OR 14.33, 95% CI 2.11–97.24), and higher paravertebral muscle fat infiltration grade (OR 16.56, 95% CI 3.51–78.09) as independent risk factors for cement nonunion. Calibration improved with regularization (Brier score: traditional 0.065; L2 0.056). All models showed high internal discrimination in stratified 10-fold cross-validation; SVM achieved the highest internal AUC (99.0%), while L2-regularized logistic regression showed balanced calibration and interpretability. Conclusion Four independent risk factors for cement nonunion were identified. The models showed promising internal performance, but external validation is required before clinical application.