Development and validation of machine learning models for predicting the risk of refracture after percutaneous kyphoplasty in OVCF patients.

Abstract

This study aims to identify the risk factors for adjacent vertebral refracture after PKP (Percutoneous Kyphoplasty) and to develop a predictive model using the multiple machine learning models to evaluate the impact of various clinical features on postoperative outcomes. A retrospective analysis was conducted on 3,942 OVCF patients who underwent PKP between 2018 and 2023. Patients were classified into non-refracture (Group A) and refracture (Group B) groups. Univariate and multivariate logistic regression analyses were used to identify independent risk factors. The multiple machine learning models were constructed and validated to predict refracture risk. Among the patients, 424 (10.75%) experienced adjacent vertebral refracture. Independent risk factors include bone mineral density (BMD), preoperative AVH, the anterior vertebral height restoration rate (AVHRR), and osteoporosis treatment (P < 0.05). The Balanced Bagging model achieved an accuracy of 96.58%, sensitivity of 94.12%, specificity of 96.88%, and an F1 score of approximately 0.8556 in predicting postoperative refracture. The risk of adjacent vertebral refracture after KP is associated with AVHRR, BMD, and osteoporosis treatment. Through a comparison of multiple models, the optimal predictive model, the Balanced Bagging model, was identified. This will aid clinicians in implementing personalized postoperative management, improving patient outcomes, and reducing the likelihood of subsequent surgeries.