INTRODUCTION
Osteoporosis (OP) is primarily characterized by the deterioration of bone microstructure and a reduction in overall bone mass, resulting in heightened bone fragility and an increased susceptibility to fractures [1]. With the aging population and continuous improvements in life expectancy, the incidence of OP is on the rise. Osteoporotic vertebral compression fracture (OVCF), a severe complication of OP, accounts for approximately onefourth of osteoporotic fractures and significantly contributes to morbidity and increased mortality rates [2]. OVCF typically results in prolonged back pain, significant restriction of daily activities, diminished quality of life, and imposes substantial medical and economic burdens on patients [3].
Treatment options for OVCF patients include both traditional conservative and surgical approaches. However, conservative methods have notable drawbacks, such as persistent back pain, extended treatment duration, and the risk of complications like pressure ulcers, respiratory and urinary tract infections, and constipation due to prolonged bed rest [4]. Therefore, percutaneous vertebral augmentation (PVA), which includes percutaneous vertebroplasty (PVP) and percutaneous kyphoplasty (PKP), has gained extensive traction in clinical practice for surgically managing OVCF. This approach swiftly alleviates back pain and facilitates the restoration of vertebral height [5]. Nonetheless, various complications have been reported in OVCF patients following PVA procedures, including adjacent vertebral fractures (AVFs), postoperative vertebral height loss, increased local kyphotic angle, and cement leakage, all potentially leading to pulmonary embolism or severe neurological dysfunction [6,7]. Studies have reported the incidence of AVF following PVA ranging from 7.9% to 21.6%, with AVF often occurring earlier than distant vertebral fractures [8-14]. Several independent risk factors associated with AVF following PVA have been identified, such as gender, bone mineral density (BMD), age, number of operated vertebrae, and history of diabetes [3,13,15]. While some studies have established AVF risk factors through controlled investigations, predicting the risk of refracture for individual patients remains challenging [16-18]. Others have integrated multiple independent risk factors and established predictive models to assess the risk of refracture events [19-21]. However, these predictive models and scoring systems face challenges in personalized calculation of individual fracture risk and visual representation of multiple risk factors.
Nomograms, widely employed mathematical tools, enable clinicians to gauge the likelihood of clinical events and tailor treatment strategies, thereby optimizing proactive posttreatment care measures [22]. However, the present nomogram predictive models for AVF were derived from single-center data, which made them susceptible to overfitting and posed challenges to their generalizability [23]. Consequently, their broader adoption in clinical practice has been impeded. Moreover, current research has lacked sufficient focus on postoperative rehabilitation and care, with a scarcity of clinical evidence guiding patients in their postoperative recovery, highlighting the urgent need for improvement in this area.
This study has utilized multicenter data from OVCF patients undergoing PVA treatment to comprehensively identify independent risk factors associated with AVF following PVA. Additionally, it has integrated considerations of postoperative care factors, aiming to establish a robust and generalizable nomogram predictive model. The model not only supports spine surgeons with preoperative decision-making and postoperative management strategies but also offers personalized prognosis prediction and rehabilitation guidance for OVCF patients.
MATERIALS AND METHODS
1. Study Design
This research was conducted and reported following the guidelines of the TRIPOD (Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis) [24]. We performed a multicenter retrospective study utilizing anonymized case databases from 3 participating medical centers (Center 1, 180 cases; Center 2, 155 cases; Center 3, 148 cases) (Fig. 1). The flowchart illustrates the basic design of this study, outlining the patient selection and variable collection processes (Fig. 1).
2. Patient Information
This study included patients who received PVA treatment for OVCF between October 2017 and February 2022. All cases completed at least 24 months of follow-up after PVA or confirmed AVF within 24 months post-PVA. This study was approved by the ethics committees of the aforementioned 3 hospitals. All methods were performed in accordance with the relevant guidelines and regulations, specifically in line with the Declaration of Helsinki. All patients included in the study signed written informed consent upon admission.
3. Inclusion and Exclusion Criteria
Patients meeting the following criteria were included in the study: (1) diagnosed with OP, meeting at least one of the following 2 conditions: (a) dual-energy x-ray absorptiometry (DXA) results of the lumbar spine or hip indicating a T-score ≤-2.5; (b) DXA results of the lumbar spine or hip indicating a T-score between -1.0 and -2.5 combined with fragility fractures; (2) history of low-energy trauma; (3) patients with a visual analogue scale score for back pain >6, and corresponding magnetic resonance imaging (MRI) confirmation of new vertebral fractures from T5 to L5; (4) received PVP or PKP treatment.
Exclusion criteria for patients: (1) incomplete medical or imaging records; (2) previous spinal surgery other than PVA; (3) preoperative symptoms of spinal cord compression or radicular injury; (4) patients who underwent PVA due to tumors, hemangiomas, infections, or symptomatic Schmorl’s nodes; (5) presence of vertebral burst fractures; (6) diagnosis of central nervous system diseases such as dementia or stroke before or during follow-up after PVA; (7) history of violent trauma after PVA.
4. Data Collection and Main Outcome Determination
All variables included in the study were categorized into baseline data, surgery-related factors, and postoperative care factors (Table 1). The radiological assessment criteria and surgical procedures involved in this study were recorded in the Supplementary Materials.
Criteria for incident vertebral fractures determination, meeting any one of the following conditions: (1) Follow-up x-ray comparison with baseline imaging showing a greater than 15% decrease in the anterior, central, or posterior height of the vertebral body adjacent to the surgical site [25]; (2) MRI confirmation of a fracture in the vertebral body adjacent to the surgical site.
5. Establishment and Evaluation of the Nomogram Predictive Model
This study will use R ver. 4.2.2 (R Foundation for Statistical Computing, Vienna, Austria) to develop a nomogram prediction model and generate evaluation charts. Least absolute shrinkage and selection operator (LASSO) regression, via the “glmnet” package, will identify key predictors, and multivariable logistic regression using “glm” will build the model.
The “rms” package will create the nomogram and calibration curve (CC) to assess prediction accuracy. Decision curve analysis (DCA), plotted with “rmda,” will evaluate the model’s clinical benefit, while the “pROC” package will calculate the receiver operating characteristic (ROC) curve and area under the curve (AUC) to measure performance. The AUC, CC, and DCA will provide a comprehensive evaluation of the model.
6. Statistical Analysis
Statistical data were presented as mean±standard deviation for continuous variables and as frequencies for categorical variables. For the comparison of continuous variables, the Student t-test was used. Categorical variables were compared using the chi-square test or Fisher exact test. A p-value less than 0.05 was considered statistically significant.
RESULTS
1. Patient Demographics
This study included 483 patients with a total of 546 fractured vertebrae. The average age was 74.0±9.5 years, with 85 males (17.57%) and 398 females (82.43%). Within 2 years post-PVA, 52 patients (10.76%) experienced adjacent vertebral refractures, while 431 patients (89.24%) did not. The database cohort was randomized into groups, with 387 patients in the training set and 96 patients in the validation set. Baseline data comparison between the groups is shown in Table 1.
2. LASSO Regression Analysis Summary
The data from the training set were analyzed using LASSO regression (Fig. 2A and B). Fig. 2B illustrates the shrinkage of the coefficients of all independent variables towards zero as the penalty coefficient increases, ultimately resulting in all coefficients being reduced to zero. Fig. 2A shows that when the predictive model includes 15 variables, the binomial deviance reaches its minimum. The selected features with non-zero coefficients include: age, BMD, history of diabetes, height, thoracolumbar junction fractures, total number of fractured vertebrae, intravertebral vacuum cleft sign, recovery rate of local kyphosis angle in fractured vertebrae, distribution of bone cement in the coronal plane, relationship between bone cement and endplate, bone cement leakage into the intervertebral disc, long-term steroid use, history of anti-OP medication injection, regular aerobic exercise, and postoperative use of thoracolumbar braces.
3. Construction of the Nomogram Model
The non-zero coefficient variables selected by the LASSO regression were included in the multivariate logistic regression analysis, resulting in 7 independent influencing factors: BMD, history of diabetes, total number of fractured vertebrae, IVC sign, recovery rate of local kyphosis angle in fractured vertebrae, regular aerobic exercise, and postoperative use of lumbar brace. These factors were used to construct the predictive model (Tables 2 and 3, Fig. 3).
4. Performance Evaluation of the Nomogram Predictive Model
The AUC for the training set was 89.21% (95% confidence interval [CI], 83.86%–94.56%), and for the validation set, it was 98.33% (95% CI, 95.61%–100%) (Fig. 4). The CC is shown in Fig. 5, with the x-axis representing the predicted probability of AVF occurrence and the y-axis representing the actual AVF outcomes. The closer the curve is to the diagonal dashed line, the more accurate the predictions. The Hosmer-Lemeshow test results showed that the p-value for the training set was 0.838, and for the validation set, it was 0.839, both significantly greater than 0.05, indicating no significant difference between the predicted and actual values and demonstrating good model fit. A comprehensive evaluation using multiple testing methods revealed that the predictive model established in this study exhibited excellent performance in both the training and validation sets.
The DCA curve (Fig. 6) of this prediction model indicates that across the entire range of AVF risk from 0 to 1, using this model can yield additional net benefits compared to the strategy of intervening with all patients or not intervening at all. This represents the broad applicability of this prediction model. For example, if a patient’s individual threshold probability for AVF is 60% (meaning they would opt for intervention treatment when their estimated AVF probability exceeds 60%), the net benefit is approximately 0.28. This indicates an advantage over strategies that either treat all patients or treat none, making the use of the nomogram-guided treatment decision more beneficial.
DISCUSSION
As a common complication of OP, OVCF affects a significant number of individuals over 50 years old, with estimates ranging from 30% to 50% [26]. Although PVA technology provides rapid pain relief and corrects spinal deformities, long-term clinical practice had revealed persistently high rates of AVFs following PVA, particularly within 2 years postprocedure [27]. Recurrent OVCFs contributed to increased economic burden, disability rates, and mortality among patients. However, there is currently no widely adopted nomogram predictive model for post-PVA AVF that visualizes both risk and protective factors.
To our knowledge, this study broke new ground by developing and validating a concise and practical nomogram predictive model based on a multicenter database of OVCF patients. Compared to previous studies on post-vertebroplasty refracture prediction models, this study did not suffer from the limitation of single-source data, thereby avoiding the issue of overfitting that occurred with single-center data [27-29]. Additionally, during the model performance validation, we found that despite using data from multiple centers and not employing internal resampling methods, the results of the ROC curves and calibration plots still demonstrated that this nomogram prediction model possessed excellent predictive performance. This indicated that our prediction model had exceptional generalizability.
The results of this study showed that among the 483 patients followed up for 2 years, there were 52 cases of AVF (10.76%). Previous studies from Asia and Europe reported varying rates of AVF post-PVA ranging from 7.9% to 21.6%, which aligned with our findings [10,11,30]. In our study cohort, except for 2 female patients aged 47, all other patients were over 50 years old, consistent with multicenter studies in North America where OVCF predominantly occurred in individuals over 50 years of age [31]. Regarding gender, there were 85 male patients (17.6%) and 398 female patients (82.4%) among those who experienced OVCF in our cohort. Therefore, for individuals aged over 50 years old, especially women, raising awareness about OP prevention and treatment was crucial to reducing the incidence of OVCF and improving overall quality of life in the elderly population.
Among the numerous risk factors for OVCF, postmenopausal status and advanced age were frequently mentioned, and these factors were usually closely related to low BMD. In addition, reduced bone mass leading to decreased biomechanical strength and increased brittleness of the bones was also a primary factor accelerating the occurrence of fragility fractures [32]. This study showed that low BMD was an independent risk factor for AVF and significantly contributed to its occurrence, indicating that clinicians should not overlook bone density screening in order to reduce the risk of AVF. In our LASSO regression analysis, anti-OP treatment emerged as a variable with a non-zero coefficient, suggesting that it could reduce the incidence of AVF. However, the final logistic regression analysis revealed that anti-OP treatment was not an independent protective factor. This did not imply that anti-OP treatment was ineffective. We considered patients who received bisphosphonates, denosumab, or parathyroid hormone treatment as having undergone anti-OP treatment, but did not quantify the treatment duration. Some patients who did not consistently receive anti-OP treatment might have confounded these results. Moreover, anti-OP treatment, as one of the effective methods for rapidly improving BMD, informed patients that adhering to standardized anti-OP treatment could effectively reduce the incidence of AVF.
Factors related to postoperative care were the most easily overlooked in clinical practice and previous research. The results of this study showed that regular aerobic exercise and wearing a lumbar brace after PVA were significant independent protective factors. This provided strong support for clinicians in guiding postoperative care for OVCF patients. Mechanistically, after suffering from OVCF, even with PVA treatment, most patients maintained low levels of physical activity due to psychological and physiological factors. Consequently, patients experienced a sharp decline in exercise capacity. As the function of key limb and trunk muscles declined, patients’ coordination and balance also decreased, ultimately leading to progressive bone density loss and an increased risk of accidental falls [33]. Regular aerobic exercise and the use of a lumbar brace provided additional support to the trunk muscles, reducing the load on the spine itself.
Diabetes, as a disease affecting bone metabolism, was evidenced to impair normal bone microstructure, with diabetic patients showing significantly lower trabecular bone scores compared to the general population. While previous studies sometimes found higher BMD in diabetic patients compared to normal populations, the risk of fractures in diabetic populations was significantly increased [34]. In conjunction with the results of this study, a history of diabetes was an independent risk factor for AVF, and based on regression coefficients, its impact on the endpoint event was considerable. This underscored the importance for clinicians to focus on controlling diabetes symptoms and implementing long-term medication plans. A national cohort study from Korea highlighted that the appropriate use of metformin in diabetes management could minimize the fracture risk in diabetic patients [35].
Multiple old vertebral fractures were often closely associated with poor bone quality and alterations in the biomechanical load transmission lines of the spine. In these patients, even with effective anti-OP drug treatment, recurrent OVCF could occur more easily [36]. Additionally, recent finite element analysis demonstrated that multiple old fractures led to changes in the biomechanical load transmission lines of normal vertebrae, increasing local stress on the vertebrae significantly and significantly increasing the risk of AVF [37].
In this study, the presence of IVC sign increased the incidence of AVF by 3.81 times compared to cases without IVC sign, likely due to the morphology of bone cement filling within the vertebrae. When IVC sign are present in the fractured vertebra, injected bone cement tends to flow first into larger voids in lowpressure areas, leading to increased volume and chunky filling of the cement within the fractured vertebra [38]. This results in inadequate bonding with the trabecular bone inside the vertebra, causing stress concentration and altering local stress transmission patterns. This observation was thoroughly validated in finite element analysis [39]. Similarly, in some OVCF patients with pronounced local kyphosis, clinicians may use techniques such as PKP or physical reduction methods to achieve better cosmetic correction by correcting the vertebra to a greater angle. However, restoring a larger local kyphotic angle often involves injecting a larger volume of bone cement, significantly altering the stiffness of the locally reinforced vertebra, thereby increasing the susceptibility of adjacent vertebrae to AVF [30].
This study had several strengths and limitations. Firstly, to our knowledge, it was the first nomogram predictive model established based on multicenter data of OVCF patients, which provided better generalizability than single-center models. Secondly, beyond baseline and surgery-related data, this study particularly included postoperative health care factors, offering more personalized guidance for the postoperative care of OVCF patients. However, as a retrospective study, it inevitably led to the exclusion of some patients, which could have introduced bias. Future studies that integrate prospective and retrospective approaches were hoped to further enhance the reliability of the predictive model.
CONCLUSION
The results of this study demonstrated that low BMD, a history of diabetes, multiple vertebral fractures, IVC sign in the fractured vertebra, and excessive local kyphotic angle restoration of the fractured vertebra were independent risk factors for recurrent AVF post-PVA. Postoperative guidance emphasizing regular aerobic exercise and thoracolumbar brace significantly reduced the risk of recurrent AVF. Based on multicenter OVCF data, this study developed a nomogram predictive model, showing good performance and reliability in model performance assessment and validation. These results provided further support for clinical diagnosis, treatment, and rehabilitation care of OVCF.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
Print
