Development and validation of a nomogram for predicting the probability of new vertebral compression fractures after vertebral augmentation of osteoporotic vertebral compression fractures

Abstract

Introduction: New vertebral compression fractures (NVCFs) are adverse events after vertebral augmentation of osteoporotic vertebral compression fractures (OVCFs). Predicting the risk of vertebral compression fractures (VCFs) accurately after surgery is still a significant challenge for spinal surgeons. The aim of our study was to identify risk factors of NCVFs after vertebral augmentation of OVCFs and develop a nomogram.

Methods: We retrospectively reviewed the medical records of patients with OVCFs who underwent percutaneous vertebroplasty (PVP) or percutaneous kyphoplasty (PKP). Patients were divided into the NVCFs group and control group, base on the patients with or without NVCFs within 2 years follow-up period after surgery. A training cohort of 403 patients diagnosed in our hospital from June 2014 to December 2016 was used for model development. The independent predictive factors of postoperative VCFs were determined by least absolute shrinkage and selection operator (LASSO) logistic regression, univariate analysis and multivariate logistic regression analysis. We provided a nomogram for predicting the risk of NVCFs based on independent predictive factors and used the receiver operating characteristic curve (ROC), calibration curve, and decision curve analyses (DCA) to evaluated the prognostic performance. After internal validation, the nomogram was further evaluated in a validation cohort of 159 patients included between January 2017 and June 2018.

Results: Of the 403 patients in the training cohort, 49(12.16%) were NVCFs at an average of 16.7 (1 to 23) months within the 2 years follow-up period. Of the 159 patients in the validation cohort, 17(10.69%) were NVCFs at an average of 8.7 (1 to 15) months within the 2 years follow-up period. In the training cohort, the proportions of elderly patients older than 80 years were 32.65 and 13.56% in the NVCFs and control group, respectively (p = 0.003). The percentages of patients with previous fracture history were 26.53 and 12.71% in the NVCFs and control group, respectively (p = 0.010). The volume of bone cement were 4.43 ± 0.88 mL and 4.02 ± 1.13 mL in the NVCFs and Control group, respectively (p = 0.014). The differences have statistical significance in the bone cement leakage, bone cement dispersion, contact with endplate, anti-osteoporotic treatment, post-op Cobb angle and Cobb angle restoration characteristics between the two groups. The model was established by multivariate logistic regression analysis to obtain independent predictors. In the training and validation cohort, the AUC of the nomogram were 0.882 (95% confidence interval (CI), 0.824-0.940) and 0.869 (95% CI: 0.811-0.927), respectively. The C index of the nomogram was 0.886 in the training cohort and 0.893 in the validation cohort, demonstrating good discrimination. In the training and validation cohort, the optimal calibration curves demonstrated the coincidence between prediction and actual status, and the decision curve analysis demonstrated that the full model had the highest clinical net benefit across the entire range of threshold probabilities.

Conclusion: A nomogram for predicting NVCFs after vertebral augmentation was established and validated. For patients evaluated by this model with predictive high risk of developing postoperative VCFs, postoperative management strategies such as enhance osteoporosis-related health education and management should be considered.

Keywords: New vertebral compression fractures; Nomogram; Osteoporosis vertebral compression fractures; Percutaneous kyphoplasty; Percutaneous vertebroplasty; Vertebral augmentation.

Fig. 1

Predictor feature selection using the LASSO logistic regression model. A Te tuning parameter(λ) was determined in the LASSO model by using a tenfold crossvalidation and a minimum criterion. B The LASSO coefficient profile plot of 20 parameters was generated against the log (lambda) sequence. A vertical line was drawn at the corresponding value with Fig. A using cross-validation, and 12 nonzero parameters were selected. Notes: 1:Age, 2:Population, 3:Hypertension, 4: Fractures, 5:Bone cement dosage, 6:Anti-osteoporosis treatment, 7:Bone cement leakage, 8:Bone cement dispersion, 9:Contact between bone cement and endplate, 10:AVHRR, 11:Post-op Cobb angle, 12:Cobb angle restoration

Fig. 2

The developed nomogram of new vertebral compression fractures after vertebral augmentation. Mark the values at each factor axis, acquire the corresponding points at the points axis, and sum up the points of all factors. Mark the total points on the total point axis and draw a perpendicular line towards the risk of leakage axis. The value on the bottom line gives the probability of the cement leakage. Note: The nomogram of postoperative VCFs was developed in training cohort, including age, bone cement dosage, bone cement leakage, bone cement dispersion, contact between bone cement and endplate, and available anti-osteoporosis treatment

Fig. 3

The receiver operating characteristic curves with corresponding area under the curves of nomogram and independent predictors in training cohort (A) and validation cohort (D). Calibration of the nomogram in training cohort (B) and validation cohort (E). The lines in the figure represent the apparent value, the bias corrected value, and ideal value. The apparent and the bias corrected values are close to each other, which means the nomogram has a good predictive performance. Decision curve analysis for nomogram prediction of risk of postoperative VCFs in training cohort (C) and validation cohort (F). The y-axis shows the net benefit: x-axis shows the threshold probability. The blue line represents the net benefit of our nomogram. The oblique gray line indicates the hypothesis that all patients had positive VCFs. The black horizontal line represents the hypothesis that no patients had positive VCFs