A nomogram for predicting residual low back pain after percutaneous kyphoplasty in osteoporotic vertebral compression fractures.

UNLABELLED

To establish a risk prediction model for residual low back pain after percutaneous kyphoplasty (PKP) for osteoporotic vertebral compression fractures. We used retrospective data for model construction and evaluated the model using internal validation and temporal external validation and finally concluded that the model had good predictive performance.

INTRODUCTION

The cause of residual low back pain in patients with osteoporotic vertebral compression fractures (OVCFs) after PKP remains highly controversial, and our goal was to investigate the most likely cause and to develop a novel nomogram for the prediction of residual low back pain and to evaluate the predictive performance of the model.

METHODS

The clinical data of 281 patients with OVCFs who underwent PKP at our hospital from July 2019 to July 2020 were reviewed. The optimal logistic regression model was determined by lasso regression for multivariate analysis, thus constructing a nomogram. Bootstrap was used to perfomance the internal validation; receiver operating characteristic (ROC) curve, calibration curve, and decision curve analysis (DCA) were used to assess the predictive performance and clinical utility of the model, respectively. Temporal external validation of the model was also performed using retrospective data from 126 patients who underwent PKP at our hospital from January 2021 to October 2021.

RESULTS

Lasso regression cross-validation showed that the variables with non-zero coefficients were the number of surgical vertebrae, preoperative bone mineral density (pre-BMD), smoking history, thoracolumbar fascia injury (TLFI), intraoperative facet joint injury (FJI), and postoperative incomplete cementing of the fracture line (ICFL). The above factors were included in the multivariate analysis and showed that the pre-BMD, smoking history, TLFI, FJI, and ICFL were independent risk factors for residual low back pain (P < 0.05). The ROC and calibration curve of the original model and temporal external validation indicated a good predictive power of the model. The DCA curve suggested that the model has good clinical practicability.

CONCLUSION

The risk prediction model has good predictive performance and clinical practicability, which can provide a certain basis for clinical decision-making in patients with OVCFs.