The Prediction Model of Refracture in Elderly Patients with Osteoporotic Vertebral Compression Fracture was Constructed Based on Artificial Intelligence Method

Background Osteoporotic vertebral compression fractures (OVCF) in elderly patients frequently lead to debilitating subsequent fractures. At present, clinicians lack validated tools to identify individuals at highest risk within this vulnerable population. This study was designed to develop an artificial intelligence-based model for predicting refracture risk specifically among elderly OVCF patients in China. Methods Hospitalization records of elderly OVCF patients (2022–2024) from three hospitals affiliated with Zunyi Medical University formed the study dataset. After developing seven machine learning models with hyperparameter optimization and 5-fold cross-validation, we evaluated performance using standard metrics-sensitivity, specificity, accuracy, and AUROC. The optimal model was subsequently used to determine the primary predictors of refracture risk. Results The elastic net model provided the most accurate predictions, achieving training and testing AUROCs of 0.735 (95% CI: 0.690–0.778) and 0.711 (95% CI: 0.632–0.775), respectively. It highlighted five key clinical factors associated with refracture: diminished lumbar bone mineral density (OR = 0.930, 95% CI: 0.613–0.999), previous fracture (OR = 1.081, 95% CI: 1.001–1.564), lower cement injection volume (OR = 0.975, 95% CI: 0.658–0.999), elevated serum total protein (OR = 1.038, 95% CI: 1.001–1.394), and performance of percutaneous vertebroplasty (OR = 1.041, 95% CI: 1.001–1.306). Conclusions In elderly OVCF patients, the elastic net model effectively predicted refracture risk.Key predictors included lumbar bone mineral density, fracture history, cement injection volume, serum total protein, and percutaneous vertebroplasty. Addressing these factors could meaningfully reduce refracture incidence and improve long-term outcomes. Clinical trial number: not applicable.