Development and validation of a risk prediction model for adjacent vertebral fracture following percutaneous vertebroplasty

Aging populations drive rising osteoporotic vertebral compression fracture (OVCF) cases, with percutaneous vertebroplasty (PVP) as the first-line therapy. However, PVP frequently causes adjacent vertebral fracture (AVF), whose risk factors remain unclear and prediction tools are lacking, impeding targeted prevention and personalized patient care. The aim of this study is to develop and validate a predictive model for AVF following PVP. We conducted a multicenter retrospective study involving two tertiary hospitals. OVCF patients undergoing PVP (2021.01–2022.12) were enrolled and randomly divided into training and internal validation subsets at a 7:3 ratio for model development and internal validation. An independent cohort (2023.01–2024.01) from the same two tertiary hospitals was included for temporal external validation. Patients were stratified into AVF and AVF-free groups based on postoperative outcomes. Univariate analysis and least absolute shrinkage and selection operator (LASSO) regression were performed in RStudio to identify significant AVF risk factors. A nomogram was constructed, internally validated, and externally validated to assess predictive performance, generalizability, and stability. Of the total 639 enrolled patients, 67 (10.5%) suffered postoperative AVF within the 2-year follow-up. This included 46 cases from the derivation cohort and 21 cases from the external validation cohort. LASSO regression revealed that cement leakage into the intervertebral space, wedge angle of the injured vertebra >15°, and no anti-osteoporosis therapy were significant risk factors for AVF after PVP in patients with OVCF. A nomogram prediction model incorporating these factors demonstrated excellent discrimination (C-index: 0.971). In the training cohort, the AUC was 0.9707 (95% CI 0.9537–0.9876); in the internal validation cohort, 0.8985 (95% CI 0.7807–1); and in the external validation cohort, 0.9391 (95% CI 0.9016–0.9767). The Calibration curves showed favorable agreement between predicted and observed outcomes. Internal validation and external validation confirmed good consistency between predicted and actual outcomes. The most important risk factors affecting AVF after PVP were cement leakage into the intervertebral space, vertebral wedge angle >15°, and no postoperative anti-osteoporosis therapy, which are statistically significant ( P < 0.05). All these predictors were integrated into our novel prediction model, which has potential predictive value and may serve as a reference for preventing AVF after PVP.