| Home | Magazines | Editorial Board | Instruction | Subscribe Guide | Archive | Advertising | Template | Guestbook | Help |
| YANG Fei,JIANG Long,WANG Qiang.Risk factors and random forest model for intravertebral bone defect after posterior transpedicular internal fixation of acute thoracolumbar fractures[J].Chinese Journal of Spine and Spinal Cord,2026,(9):1066-1075. |
| Risk factors and random forest model for intravertebral bone defect after posterior transpedicular internal fixation of acute thoracolumbar fractures |
| Received:May 20, 2026 Revised:August 10, 2026 |
| English Keywords:Acute thoracolumbar fracture Posterior transpedicular internal fixation Intravertebral bone defect Random forest Risk factors |
| Fund:江苏大学临床医学科技发展基金项目 (编号:JLY2021043);无锡市卫生和计划生育委员会卫生科研项目(编号:Q202060) |
|
| Hits: 33 |
| Download times: 0 |
| English Abstract: |
| 【Abstract】 Objectives: To investigate the risk factors for intravertebral bone defect after posterior transpedicular internal fixation in patients with acute thoracolumbar fractures, to construct a random forest prediction model, and to evaluate its predictive performance and conduct internal validation. Methods: A total of 260 patients with acute single-level thoracolumbar fractures who were admitted to our hospital and underwent posterior transpedicular internal fixation between June 2020 and December 2023 were retrospectively enrolled. The patients were followed up for 12 months after surgery and were divided into a bone-defect group and a non-bone-defect group according to follow-up CT results. The collected data included general patient information, bone mineral density T-score, fracture morphological characteristics, endplate–intervertebral disc complex injury, surgical strategy, and reduction-related parameters. Variables were screened using least absolute shrinkage and selection operator(LASSO) regression, and multivariable logistic regression was used to analyze the independent factors associated with the development of intravertebral bone defect at 12 months after surgery. Logistic regression and random forest models were constructed. Receiver operating characteristic(ROC) curves were plotted to obtain the area under the curve(AUC) for comparing model performance, and the model with the best performance was selected. Internal validation of model performance was performed by obtaining the corrected C-index using bootstrap resampling repeated 1000 times, and calibration curves and decision curve analysis were used to evaluate model calibration and clinical benefit. In addition, 104 patients with acute single-level thoracolumbar fractures who underwent posterior transpedicular internal fixation at our hospital between January 2024 and February 2025 were selected for temporal external validation. Results: Among the 260 patients used for model construction, 78 developed intravertebral bone defect, with an incidence of 30.00%. After LASSO screening, multivariable logistic regression showed that age, bone mineral density T-score, preoperative anterior height compression ratio of the injured vertebra, endplate-intervertebral disc complex injury, intraoperative overdistraction, and bone grafting through the injured vertebra were independent factors associated with postoperative intravertebral bone defect(P<0.05). The random forest model showed that, according to the mean decrease in accuracy, the variables ranked as follows in terms of their relative contribution to its predictive performance: bone mineral density T-score, bone grafting through the injured vertebra, endplate-intervertebral disc complex injury, preoperative anterior height compression ratio of the injured vertebra, age, and intraoperative overdistraction. The ROC curves showed that the AUC of the random forest model was 0.932(95%CI: 0.900-0.963), which was higher than that of the logistic regression model(0.883, 95%CI: 0.838-0.928), and the difference was statistically significant by the DeLong test(P=0.001). Internal validation showed that the corrected C-index of the random forest model was 0.939, suggesting good model discrimination. The calibration curve showed a high degree of overlap between the ideal and actual probability curves, with a mean absolute error of 0.021 and a Brier score of 0.088. The decision curve showed that the random forest model provided good clinical net benefit when the threshold probability was between 0.05 and 0.96. In addition, temporal external validation showed that the random forest model had a sensitivity of 0.800, a specificity of 0.851, a positive predictive value of 0.686, and a negative predictive value of 0.913. Conclusions: Intravertebral bone defect after internal fixation in patients with acute thoracolumbar fractures is closely associated with advanced age, poor bone quality, and severe vertebral compression, and the random forest model constructed on this basis shows good predictive performance. |
| View Full Text View/Add Comment Download reader |
| Close |
|
|
|
|
|