Modelling of multisegmental osteoporotic vertebral compression fracture using machine learning to analyse and predict risk factors.

Modelling of multisegmental osteoporotic vertebral compression fracture using machine learning to analyse and predict risk factors. | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Modelling of multisegmental osteoporotic vertebral compression fracture using machine learning to analyse and predict risk factors. Xuetao Zhu, Ling Zhang, Yuanqiang Zhang, Weibing Si, Dejian Liu, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7295247/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background The aim is to investigate the risk factors for re-fracture of multisegmental osteoporotic vertebral compression fractures (OVCF) and to construct a clinical prediction model for their occurrence using machine learning algorithms. Objective The aim is to investigate the risk factors for re-fracture of multisegmental osteoporotic vertebral compression fractures (OVCF) and to construct a clinical prediction model for their occurrence using machine learning algorithms. Methods The CT images were imported into Slice-O-Matic software, and the fat infiltration rate of paraspinal and lumbar major muscles, paraspinal muscle mass, and lumbar major muscle mass were measured for each patient. The screening process was conducted through a multifaceted approach encompassing between-group comparison analysis, LASSO regression, and multivariate logistic regression. To ensure the robustness of the models, a total of 13 machine learning algorithms were employed in their construction. The prediction performance of each model was evaluated and the optimal model was selected through accuracy, recall, precision, F1 score and area under the receiver operating characteristic curve (AUROC). Finally, the models were interpreted using SHAP analysis to elucidate the importance of the features in the models and their impact on the direction of prediction. Conclusion In this study, the fat infiltration rate of paravertebral muscle and total type I collagen amino-terminal extender peptide were identified as potential risk factors for the development of multisegmental OVCF. A prediction model for the occurrence of multisegmental OVCF was constructed by the XGBoost model, and the model was evaluated to have good predictive performance. SHAP analysis was utilised to enhance the interpretability of the model, thereby demonstrating the importance of paraspinal muscle fat infiltration rate and total amino-terminal-propeptide of type I collagen in the prediction of the model. Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 1. Introduction As the global population ages, the prevalence of osteoporosis continues to increase. One of the most common complications of osteoporosis is osteoporotic vertebral compression fracture (OVCF), the incidence of which has increased annually. According to the latest research data, the prevalence of osteoporotic vertebral compression fractures (OVCF) in the elderly population is as high as 20–30% in some developed countries 1 – 3 . In China, the incidence of osteoporotic vertebral compression fractures (OVCF) is increasing rapidly as the economy develops and the population ages. Studies have shown that OVCF patients have significantly lower quality of life scores than the general population, particularly with regard to physical functioning, mental health, and social participation. Gradual loss of the ability to care for oneself due to long-term pain and impaired mobility can place a heavy burden on society and families, as patients require more care and support from their families and society 4 . A multisegmental spinal fracture is a fracture of two or more vertebrae at the same time. The pathogenesis of multisegmental spinal fractures is more complex than that of single-segment spinal fractures, as it involves the continuity and stability of multiple vertebrae. Furthermore, multisegmental fractures may be accompanied by neurological injuries, such as spinal cord compression and nerve root involvement, which increases the difficulty and risk of treatment. OVCF was previously recognized as a skeletal disorder and most studies have focused on the effects on bone density 5 , 6 . Recent studies have shown that sarcopenia is more prevalent in patients with osteoporotic vertebral compression fracture (OVCF) than in patients without OVCF, and that OVCF is also significantly more prevalent in patients with sarcopenia 7 , 8 . A decrease in paraspinal muscle (PM) density is one of the early clinical manifestations of sarcopenia and has gradually become more widely recognised in relation to OVCF. The paravertebral muscles, which are mainly composed of the multifidus and the erector spinae muscles, are located on both sides of the spine. They are important for spinal stability and motor function 9 , 10 . The paraspinal muscles maintain the biomechanical balance of the spine through compensatory mechanisms. However, when the muscle tissue density decreases to the compensatory threshold, a double pathological effect occurs: the overall stability of the spine decreases significantly and the stress load on specific segments increases 11 . This leads to the occurrence of osteoporotic vertebral compression fracture 12 . Thus, there is a strong correlation between paraspinal muscles and OVCF. However, no studies have investigated the role of paraspinal muscles in the development of multisegmental OVCF. In recent years, machine learning algorithms have become deeply integrated into the medical field. Machine learning is a subcategory of artificial intelligence, and its core mechanism involves constructing computational models that can autonomously evolve and continuously optimise diagnostic prediction models based on large amounts of medical data. There are multiple ways in which artificial intelligence can be combined with osteoporosis research 13 – 16 . However, there are no prediction models for multisegmental OVCF in existing studies. Therefore, we employed 13 machine learning methods to develop predictive models. The aim of this study was therefore to collect a comprehensive set of CT and clinical indicators in order to identify the most sensitive indicators of multi-stage osteoporotic refracture, and to build a predictive model of OVCF through machine learning. This will enable clinicians to take targeted preventive and therapeutic measures for high-risk patients. Figure 1 shows the detailed workflow. 2. Methods 2.1 Patients : In this study, data were collected from a total of 1,632 patients with osteoporotic vertebral compression fractures (OVCF) attending three tertiary-level hospitals in Shandong Province between January 2015 and October 2022. A total of 298 patients who met the inclusion and exclusion criteria were screened. The selected patients had a complete medical history and imaging and test data. Inclusion criteria: (1)A discharge diagnosis including osteoporosis, spinal compression fracture, or osteoporotic vertebral compression fracture;(2) Bone densitometry findings indicating decreased bone mass or osteoporosis;(3) Imaging reports indicating new vertebral compression fractures of the spine. Exclusion criteria: (1) Inadequate medical history or imaging results to determine whether the vertebrae in the fractured segment of the spine are newly fractured;(2) A history of primary or secondary spinal tumours, spinal tuberculosis or spinal infection.(3) History of surgery involving the implantation of internal spinal fixation devices;(4) Vertebral compression fractures that are clearly caused by other diseases or violent trauma;(5) Unclear or unavailable CT images. 2.2 Clinical Data Collection In this study, we obtained the study data through the relevant information systems of three tertiary hospitals in Shandong Province. Various types of clinical data were collected from 298 patients using a retrospective study method and strict inclusion and exclusion criteria. This data covered basic patient information as well as the results of several laboratory and imaging examinations. The basic information included name, gender, hospital number and age. The laboratory tests included erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), parathyroid hormone (PTH), total blood pressure (TBP) and body mass index (BMI). The following were also collected: 25-hydroxy vitamin D (25-OH-vitD), N-terminal and midcourse osteocalcin (N-MIDOS), β-collagen degradation products (β-CrossLaps), total type I collagen amino-terminal extender peptide, and β-collagen degradation products. Collagen amino-terminal propeptide of type I collagen (PINP), blood calcium (Ca), blood phosphorus (P) and blood magnesium (Mg); albumin (ALB); creatinine (Cr); prealbumin (PAB); bile acid (BA); and cystatin C (Cys-C) were collected. Bone mineral density (BMD) and spinal CT images were also collected. BMD and CT images of the spine were collected. In addition, information on the patient's past history of OVCF and underlying diseases was collected. 2.3 Image Acquisition Spinal CT scanning using a 64-channel spiral CT scanner was carried out on all of the patients included in this study, and the resulting data were saved in DICOM format. To ensure data quality, data containing noise or motion artefacts was excluded. The specific parameters used for CT scanning were as follows: electron tube current of 300 mA, electron tube voltage adjusted within the range of 70–140 kV, and slice thickness of 0.4 mm. 2.4 Image Processing Computed tomography (CT) is becoming an increasingly popular research tool for studying aspects of skeletal muscle biology in vivo. Skeletal muscle fat content can be quantified using X-ray attenuation properties, and radiation attenuation values (in Hounsfield units, or HU) are negatively correlated with the degree of muscle degeneration. Pathological changes in the paraspinal muscles most often occur between the L4 and L5 vertebrae. To best represent the degree of degeneration of the paraspinal muscles, we selected an axial image of the middle plane of the L4-L5 vertebrae from each patient's CT scan. We then manually sketched the extent of the paraspinal muscles and the psoas major muscle (PMM) in this cross-section. The software automatically recognised the muscles and fat, and the cross-sectional areas of the muscle and intermuscular fat were output. Spinal CT images from 298 patients were analysed using SliceOmatic version 5.The specific operations are briefly described below. As shown in Fig. 2 , the axial images of the middle plane of the L4-L5 vertebral body were first selected. The ranges of the paravertebral muscles and the lumbar major muscle were then sketched out. The partition calculation was carried out by setting the appropriate threshold ranges so that the muscular and fatty tissues could be automatically identified. The threshold ranges were set to -29 HU to 150 HU for muscular tissues and − 190 HU to -30 HU for fatty tissues. The software automatically calculates the cross-sectional area (CSA) and average CT value of corresponding tissues and outputs these as paraspinal muscle CSA (PMCSA), paraspinal interosseous fat CSA (PMFCSA) and fat CSA (FCSA). The outputs were the paraspinal muscle cross-sectional area (PMCSA), the paraspinal intermuscular fat cross-sectional area (PMFCSA), the psoas major muscle cross-sectional area (PMMCSA), the psoas major intermuscular fat cross-sectional area (PMMFCSA), the paravertebral muscle density (PMD) and the psoas major muscle density (PMMD). The fat infiltration ratio was also calculated. The fat infiltration ratio (FIR), the paraspinal muscle fat infiltration ratio (PMFIR), and the psoas major muscle fat infiltration ratio (PMMFIR) were calculated as follows: PMFIR = PMFCSA/(PMCSA + PMFCSA) × 100%; PMMFIR = PMMFCSA/(PMMCSA + PMMFCSA) × 100%. 2.5 Descriptive Statistical Analysis The Kolmogorov–Smirnov test was used to test the numerical variables for conformity to a normal distribution. Variables that conformed to a normal distribution were presented as mean ± standard deviation (mean ± SD), while those that did not conform to a normal distribution were presented as median and interquartile range (median, IQR). Categorical variables were presented as counts and percentages (n,%). The t-test was used for numerical variables that conformed to a normal distribution and were consistent with the chi-squared test. The non-parametric rank sum test was used for other numerical variables and the chi-squared test for categorical variables. Variables that were statistically significant after LASSO regression screening were included in multifactor logistic regression analysis. The odds ratio (OR) and 95% confidence interval (CI) were used as indicators of effect. Risk factors for the occurrence of multisegmental osteoporotic vertebral compression fracture (OVCF) were identified based on the significance of the differences and the regression coefficients of the analysed results. 2.6 Predictive Model Construction、Performance Evaluation and Interpretability After screening the risk factors, the original dataset was divided using the random sampling method in a 7:3 ratio. Seventy per cent of the data was used to construct the model and 30 per cent was used to validate it internally. Thirteen commonly used machine learning algorithms were used to construct the prediction model, including Decision Tree, Random Forest, AdaBoost, GBDT, CatBoost, KNN, Extra Trees, ANN, SVM, XGBoost, LightGBM, Simple Bayes and Logistic Regression. The prediction model was then evaluated using the test set. The neural network, SVM, XGBoost, LightGBM, Simple Bayes and Logistic Regression algorithms were used to construct the prediction model on the training set data, and the model was then evaluated using the test set. When evaluating the performance of machine learning models, we use important metrics such as accuracy, precision, recall, the F1 score and the AUROC to show how well the models perform in different areas. The machine learning model's prediction results are finally interpreted using the SHAP (Shapley Additive Explanations of Additive Features) method. 3. Results 3.1Comparative Results of Patient Data Groups General information was compared between patients in the single-segment and multi-segment groups (Table 1). As can be seen from the table, the only significant difference in the baseline data of the two groups was in the history of previous OVCF (P=0.036), while the history of the underlying disease was not significantly different between the two groups (P>0.1). A comparison of the two groups' admission laboratory indicators, examination results and muscle measurement parameters (Table 2) revealed significant differences in a variety of factors. To further evaluate the predictive value of these indicators for OVCF with multiple segments, ROC curves were plotted (Fig 3) and the area under the curve (AUC) was calculated. The AUC values for each indicator were then ranked according to their magnitude. 3.2 LASSO Regression The collected data were analysed through the LASSO regression method, selecting statistically significant variables in the between-group comparison analysis by applying the LASSO regression model, and selecting the optimal regularization parameter λ through the cross-validation method(Fig.4A). The regression coefficients of each variable were obtained by substituting the optimal λ values into the LASSO regression and analysing it on the training set. Following a thorough analysis of the regression coefficients, the risk factors with significant effects on multisegmental OVCF were identified(Fig.4B). 3.3 Multivariate Logistic Regression The variables identified by LASSO regression were included in a multivariate logistic regression analysis. This analysis showed that T-P1NT levels and the rate of fat infiltration in the paravertebral muscles significantly affected the recurrence of multisegmental osteoporotic vertebral compression fractures (Table.3). Consequently, total type I collagen amino-terminal prolongation peptide (T-PINT) and paravertebral muscle fat infiltration rate (PMFIR) have been identified as independent risk factors for the development of multisegmental OVCF. 3.4 Machine Learning Predictive Model Construction and Evaluation In this study, total type I collagen amino-terminal extender peptide (T-PINT) and paravertebral muscle fat infiltration rate (PMFIR) were selected for the construction of a model and its application to 13 machine learning algorithms. The selection of algorithms included decision trees, random forests, AdaBoost, GBDT, CatBoost, KNN, Extra Trees, ANN, SVM, XGBoost, LightGBM, Simple Bayes, and Logistic Regression. The calculation of each algorithm's accuracy, recall, precision, F1 score and AUROC is conducted independently. Following a thorough evaluation, it was determined that the XGBoost model demonstrated the most effective performance and was thus identified as the optimal machine learning algorithm for predicting multi-segment OVCF(Fig.5). The XGBoost model performed the best with an accuracy of 0.944, precision of 0.945 and AUROC of 0.958. 3.5 SHAP Analysis of Predictive Models In this study, SHAP analysis was employed to facilitate a more intuitive interpretation of the XGBoost prediction model at both the aggregate and individual levels, respectively. At the aggregate level, the SHAP values for each risk factor were calculated to quantify the degree of dependence of the model on different risk factors for prediction. The most significant features were identified as total type I collagen amino-terminal extender peptide and paraspinal muscle fat infiltration rate(Fig.6 A.B). At the same time, we label the two samples as 'A' and 'B'. We then interpret the results separately using waterfall and force diagrams(Fig.6 C-F). 4. Discussion As the population ages and the average life expectancy increases, the incidence of osteoporosis is on the rise. Among the complications caused by osteoporosis, OVCF is particularly prevalent and a major health concern worldwide 16 . The occurrence of osteoporotic vertebral compression fracture (OVCF) causes long-term back pain in the elderly 17 . This not only seriously affects their mobility, but also leads to limitations in their daily activities, greatly reducing their quality of life. To treat the disease and relieve the pain, patients must incur significant medical expenses, which undoubtedly places a heavy financial burden on their families. Complications such as re-fracture and spinal cord compression are more common with multi-segment OVCF than with single-segment OVCF 18 . Surgical treatment requires intervention on multiple vertebrae and is therefore more difficult. It also carries a higher risk of complications and refracture. Current clinical studies on multisegmental osteoporotic vertebral compression fracture (OVCF) mostly focus on treatment and prognosis, with little analysis of risk factors. In this study, we employed statistical methods and multifactorial logistic regression to identify total amino-terminal-propeptide of type I collagen and paravertebral muscle fat infiltration rate as risk factors for the development of multisegmental OVCF. These factors were then incorporated into the construction of a prediction model. Bone turnover markers are important indicators that reflect the state of bone metabolism 19 . They assist in the diagnosis, differential diagnosis, treatment and evaluation of the efficacy of osteoporosis 20 . Under normal physiological conditions, the human skeleton maintains a dynamic balance of bone turnover. A number of bone metabolites reflecting the activity of osteoblasts and osteoclasts are produced in this process, and these are also known as bone turnover markers. Previous studies have reported that re-fracture is influenced by various risk factors. In the present study, the focus was on type I collagen amino-terminal extender peptide and paravertebral muscle fat infiltration rate as the main influences on re-fracture in osteoporotic vertebral compression fracture (OVCF) 21 . In diagnosing and treating osteoporosis, bone turnover markers play a very important role 22 – 24 . Changes in their levels in the patient's serum can be used to assess treatment effectiveness and predict osteoporotic fractures 25 . Currently, type I collagen amino-terminal extender peptide is generally recommended internationally as the preferred bone formation marker for evaluating the bone formation process, as well as for diagnosing and studying related diseases 24 . Some studies have shown that type I collagen amino-terminal extender peptide is an independent risk factor for re-fracture after PKP in patients with OVCF 26 . It has also been demonstrated that, in patients with osteoporotic vertebral compression fractures who experience delayed fracture healing following vertebroplasty, the expression level of serum type I collagen amino-terminal elongation peptide is increased in the early postoperative period 27 . There is also a certain correlation with prognostic indexes, which can be used as an additional indicator to predict the occurrence of delayed fracture healing in OVCF patients following surgery. The paravertebral muscles are primarily made up of the multifidus and the erector spinae muscles. These muscles play an important role in maintaining spinal stability and movement. Past studies have found a strong association between muscle mass and osteoporosis, and muscle mass can be used to predict osteoporosis to some extent 28 , 29 . When these relationships were considered in the context of osteoporosis and fracture risk, it was found that muscle and fat were associated with each other with regard to fracture risk in patients with osteoporosis. In a study by Cheng et al., it was demonstrated that low paraspinal muscle lean mass at the L4 level was an independent predictor of adjacent vertebral fracture in patients treated with PKP for OVCF 30 . The severity of OVCF was found to be correlated with the degree of fat infiltration in the paraspinal muscles; higher levels of infiltration were associated with greater vertebral compression 31 . These studies further demonstrate the strong association between paravertebral muscles and OVCF. Based on this, this study collected CT images of patients with OVCF and processed them using software to identify and output the paraspinal muscle fat infiltration rate and density, and the lumbar major muscle fat infiltration rate and density. Ultimately, it was determined that an elevated paraspinal muscle fat infiltration rate is a risk factor for multisegmental OVCF. In recent years, machine learning has become widely used in various medical disciplines. It can help us to predict disease trends, screen high-risk groups, and guide prevention strategies, all of which play an important role in accelerating the development of biomedicine. Machine learning algorithms can produce the best model predictions with the fewest risk factor models. They can also find optimal subsets of features, significantly improving model utility 32 , 33 . With machine learning algorithms, we can accurately map various features to specific classifications. This mapping operation allows us to make predictions that are useful in practice. In our study, the total type I collagen amino-terminal extension peptide and the paravertebral muscle fat infiltration rate were selected for inclusion in the prediction model using decision trees, random forests, AdaBoost, GBDT, CatBoost, KNN, Extra Trees, ANN, SVM, XGBoost, LightGBM, the plain Bayesian algorithm and logistic regression. Thirteen commonly used machine learning algorithms were employed to construct the prediction model. By comparing the predictive efficacy of each algorithm, XGBoost was selected to build the model for multisegmental OVCF due to its high probability of correct prediction and good differentiation of patients with the condition. Machine learning algorithms can capture nonlinear, high-dimensional relationships between predictors and provide an accurate, personalised approach to predicting the re-fracture of osteoporotic vertebral fractures. In our study, we constructed a machine learning model and a logistic regression (LR) model, finding that the machine learning model had higher area under the curve (AUC) values compared to the LR model. The strength of this multicentre study is that it included OVCF patients from three tertiary hospitals, enhancing the applicability of the results. This study innovatively proposed that the fat infiltration rate of the paravertebral muscle is closely related to the occurrence of multisegmental OVCF. Thirteen machine learning models were used in the modelling process. After comparing the predictive performance of each model, XGBoost was chosen to construct a prediction model for multisegmental OVCF, demonstrating the value of studying XGBoost in this context. SHAP analysis improved the interpretability of the prediction model, enabling clinicians to make more informed choices about whether to trust and adopt the model's predictions. However, there are limitations to this study. As a retrospective study, selective bias may occur. The sample size is relatively small and the evidence is insufficient. A large number of samples are needed for further study. Although we evaluated a large number of variables, the influence of confounding factors that were not included due to poorly documented patient histories cannot be ignored. Furthermore, the model's performance has not been validated by an external dataset. To address these limitations, subsequent prospective multicentre cohort studies are required to incorporate a more comprehensive set of variables in order to build an optimisation model and thus verify the generalisation efficacy of the current findings. 5. Conclusion This study identified multiple factors as potential risk indicators for multisegmental osteoporotic vertebral compression fracture (OVCF). Through LASSO regression and multifactorial logistic regression, the fat infiltration rate of the paravertebral muscle and the total type I collagen amino-terminal extension peptide were selected as predictive indexes to be included in the machine learning model. The XGBoost model was ultimately chosen as it demonstrated excellent predictive performance. Meanwhile, SHAP analysis visualised and interpreted the model, indicating that the fat infiltration rate of the paravertebral muscle was the most important feature for prediction. This provides clinicians with sufficient information to understand and apply the model. Declarations Ethics Approval and Consent to Participate This study was approved by the ethics committee of Qilu Hospital. Informed consent was obtained from all participants. Competing Interests The authors declare no competing interests. Acknowledgements This work was supported partly by the National Natural Science Foundation of China (82402758). Author’s contributions Xuetao Zhu designed the experiments., Dejian Liu and Ling Zhang acquired the data. Yuanqiang Zhang performed the statistical analyses. Aibo Song and Weibing Si edited the manuscript. All authors drafted and reviewed the manuscript.. Data Availability The datasets generated and/or analyzed during the current study are available from the corresponding authors on reasonable request. References Ensrud KE, Blackwell TL, Fink HA, et al. What Proportion of Incident Radiographic Vertebral Fractures in Older Men Is Clinically Diagnosed and Vice Versa: A Prospective Study. J Bone Miner Res 2016; 31 (8): 1500–3. Fink HA, Milavetz DL, Palermo L, et al. What proportion of incident radiographic vertebral deformities is clinically diagnosed and vice versa? J Bone Miner Res 2005; 20 (7): 1216–22. Middleton ET, Steel SA. Routine versus targeted vertebral fracture assessment for the detection of vertebral fractures. Osteoporos Int 2008; 19 (8): 1167–73. Wang O, Hu Y, Gong S, et al. A survey of outcomes and management of patients post fragility fractures in China. Osteoporos Int 2015; 26 (11): 2631–40. Zvekic-Svorcan J, Aleksic J, Jankovic T, et al. Capture the vertebral fracture: Risk factors as a prediction. J Back Musculoskelet Rehabil 2019; 32 (2): 269–76. Ji C, Rong Y, Wang J, et al. Risk Factors for Refracture following Primary Osteoporotic Vertebral Compression Fractures. Pain Physician 2021; 24 (3): E335–e40. Wang WF, Lin CW, Xie CN, et al. The association between sarcopenia and osteoporotic vertebral compression refractures. Osteoporos Int 2019; 30 (12): 2459–67. Lidar S, Salame K, Chua M, et al. Sarcopenia Is an Independent Risk Factor for Subsequent Osteoporotic Vertebral Fractures Following Percutaneous Cement Augmentation in Elderly Patients. J Clin Med 2022; 11 (19). Takahashi S, Hoshino M, Ohyama S, et al. Relationship of back muscle and knee extensors with the compensatory mechanism of sagittal alignment in a community-dwelling elderly population. Sci Rep 2021; 11 (1): 2179. Banno T, Arima H, Hasegawa T, et al. The Effect of Paravertebral Muscle on the Maintenance of Upright Posture in Patients With Adult Spinal Deformity. Spine Deform 2019; 7 (1): 125–31. Lee DG, Bae JH. Fatty infiltration of the multifidus muscle independently increases osteoporotic vertebral compression fracture risk. BMC Musculoskelet Disord 2023; 24 (1): 508. Wang M, Liu J, Chen X, Wang X, Chen X. Muscle quality and spine fractures. J Cachexia Sarcopenia Muscle 2022; 13 (2): 1426–8. Zhu X, Liu D, Liu L, et al. Fully Automatic Deep Learning Model for Spine Refracture in Patients with OVCF: A Multi-Center Study. Orthop Surg 2024; 16 (8): 2052–65. Smets J, Shevroja E, Hügle T, Leslie WD, Hans D. Machine Learning Solutions for Osteoporosis-A Review. J Bone Miner Res 2021; 36 (5): 833–51. Sebro R, Elmahdy M. Machine Learning for Opportunistic Screening for Osteoporosis and Osteopenia Using Knee CT Scans. Can Assoc Radiol J 2023; 74 (4): 676–87. Ferizi U, Honig S, Chang G. Artificial intelligence, osteoporosis and fragility fractures. Curr Opin Rheumatol 2019; 31 (4): 368–75. Patil S, Rawall S, Singh D, et al. Surgical patterns in osteoporotic vertebral compression fractures. Eur Spine J 2013; 22 (4): 883–91. Tseng YY, Yang TC, Tu PH, Lo YL, Yang ST. Repeated and multiple new vertebral compression fractures after percutaneous transpedicular vertebroplasty. Spine (Phila Pa 1976) 2009; 34 (18): 1917–22. Schini M, Vilaca T, Gossiel F, Salam S, Eastell R. Bone Turnover Markers: Basic Biology to Clinical Applications. Endocr Rev 2023; 44 (3): 417–73. Brown JP, Don-Wauchope A, Douville P, Albert C, Vasikaran SD. Current use of bone turnover markers in the management of osteoporosis. Clin Biochem 2022; 109-110 : 1–10. Zhang J, Hu Y, Cai W. Bone metabolism factors in predicting the risk of osteoporosis fracture in the elderly. BMC Musculoskelet Disord 2024; 25 (1): 442. Bhattoa HP, Vasikaran S, Trifonidi I, et al. Update on the role of bone turnover markers in the diagnosis and management of osteoporosis: a consensus paper from The European Society for Clinical and Economic Aspects of Osteoporosis, Osteoarthritis and Musculoskeletal Diseases (ESCEO), International Osteoporosis Foundation (IOF), and International Federation of Clinical Chemistry and Laboratory Medicine (IFCC). Osteoporos Int 2025; 36 (4): 579–608. Tan RZ, Loh TP, Vasikaran S. Bone turnover marker monitoring in osteoporosis treatment response. Eur J Endocrinol 2020; 183 (1): C5–c7. Eastell R, Szulc P. Use of bone turnover markers in postmenopausal osteoporosis. Lancet Diabetes Endocrinol 2017; 5 (11): 908–23. Di Medio L, Brandi ML. Advances in bone turnover markers. Adv Clin Chem 2021; 105 : 101–40. Li YX, Guo DQ, Zhang SC, et al. Risk factor analysis for re-collapse of cemented vertebrae after percutaneous vertebroplasty (PVP) or percutaneous kyphoplasty (PKP). Int Orthop 2018; 42 (9): 2131–9. Cauley JA, Chalhoub D, Kassem AM, Fuleihan Gel H. Geographic and ethnic disparities in osteoporotic fractures. Nat Rev Endocrinol 2014; 10 (6): 338–51. Yang S, Center JR, Eisman JA, Nguyen TV. Association between fat mass, lean mass, and bone loss: the Dubbo Osteoporosis Epidemiology Study. Osteoporos Int 2015; 26 (4): 1381–6. Rosen CJ, Klibanski A. Bone, fat, and body composition: evolving concepts in the pathogenesis of osteoporosis. Am J Med 2009; 122 (5): 409–14. Cheng Y, Yang H, Hai Y, et al. Low paraspinal lean muscle mass is an independent predictor of adjacent vertebral compression fractures after percutaneous kyphoplasty: A propensity score-matched case-control study. Front Surg 2022; 9 : 965332. Huang W, Cai XH, Li YR, et al. The association between paraspinal muscle degeneration and osteoporotic vertebral compression fracture severity in postmenopausal women. J Back Musculoskelet Rehabil 2023; 36 (2): 323–9. Wu EQ, Hu D, Deng PY, et al. Nonparametric Bayesian Prior Inducing Deep Network for Automatic Detection of Cognitive Status. IEEE Trans Cybern 2021; 51 (11): 5483–96. Wu EQ, Zhou M, Hu D, et al. Self-Paced Dynamic Infinite Mixture Model for Fatigue Evaluation of Pilots' Brains. IEEE Trans Cybern 2022; 52 (7): 5623–38. Tables Table1 General information of OVCF patients characterization single fracture group Multiple fracture group （n=181） (n=117) P Gender (n,%) Male Female 34(18.78) 147(81.22) 17(14.53) 100(85.47) 0.341 Pre-existing OVCF history (n,%) YES NO 53(29.28) 128(70.72) 48(41.03) 69(58.97) 0.036 Diseases of the cardiovascular system (%) YES NO 68(37.57) 113(62.43) 40(34.19) 77(65.81) 0.553 Diabetes (n,%) YES NO 19(10.50) 162(89.50) 8(6.84) 109(93.16) 0.282 Respiratory diseases (n,%) YES NO 8(4.42) 173(95.58) 8(6.84) 109(93.16) 0.366 Rheumatologic diseases (n,%) YES NO 4(2.21) 177(97.79) 5(4.27) 112(95.73) 0.503 Age (median,IQR) 69(63.5,77.0) 69(64.0,76.0) 0.957 Table.2 Imaging and laboratory characteristics of OVCF patients characterization single fracture group （n=181) Multiple fracture group (n=117) P Erythrocyte sedimentation rate(mm/h) (median,IQR) 26(14.0,41.0) 23(13.0,36.5) 0.432 C-reactive protein (mg/L)( median,IQR) 4.230(1.9,11.9) 3.230(1.2,10.1) 0.074 parathyroid hormone (pg/ml)( median,IQR) 35.810(26.4,47.2) 32.950(24.3,43.5) 0.086 25-hydroxy vitamin D (ng/ml)(median,IQR) 17.500(13.3,24.0) 21.950(16.2,29.7) 0.000 Amino-terminal and midcourse osteocalcin (ng/ml) (median,IQR) 16.510(12.1,22.0) 19.740(13.9,26.0) 0.004 β-crosslaps (ng/ml) (median,IQR) 0.716(0.5,0.9) 0.810(0.6,1.1) 0.007 T-PINT(ng/ml) (median,IQR) 58.480(41.2,77.6) 70.04(54.4,93.7) 0.000 Calcium (mmol/L)(median,IQR) 2.280(2.2,2.4) 2.300(2.2,2.4) 0.323 Phosphorus (mmol/L) (mean±SD) 1.18±0.18 1.22±0.18 0.034 Magnesium (mmol/L) (mean±SD) 0.91±0.07 0.90±0.08 0.411 Albumin (g/L)(median,IQR) 41.400(38.1,44.3) 40.500(37.6,43.3) 0.057 Creatinine (umol/L)(median,IQR) 58.000(52.0,70.0) 56.000(50.0,66.7) 0.051 Prealbumin (g/L) (median,IQR) 22.600(19.2,30.9) 22.300(18.3,32.1) 0.511 Bile acid (umol/L)(median,IQR) 4.100(2.5,7.3) 4.100(2.7,7.0) 0.839 Cystatin C (mg/L)(median,IQR) 0.910(0.8,1.1) 0.940(0.8,1.1) 0.51 Bone mineral density (median,IQR) -3.000(-3.8, -2.2) -3.300(-4.3, -2.4) 0.030 PMD(HU)(median,IQR) 28.450(22.0,33.8) 21.780(14.0,28.9) 0.000 PMMD(HU)(median,IQR) 41.840(37.8,45.4) 38.940(34.3,43.2) 0.001 PMFIR(%) (median,IQR) 25.663(19.9,35.7) 37.014(30.6,46.2) 0.000 PMMFIR (%)(median,IQR) 2.7(1.5,5.2) 3.908(2.3,6.2) 0.004 Table.3 Risk factors of multi-level OVCF determined by multivariable logistic regression Variable Regression coefficient P OR 95%CI lower limit Upper Limit constant -4.665 0.00 0.009 0.002 0.042 β-CROSS 0.41 0.41 1.508 0.565 4.024 T-P1NT 0.011 0.05 1.011 1 1.023 BMD -0.226 0.06 0.798 0.628 1.013 PMFIR 0.07 0.00 1.073 1.044 1.103 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7295247","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":509324877,"identity":"332749a6-a429-4e48-99b1-0635bea9867d","order_by":0,"name":"Xuetao Zhu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAzElEQVRIiWNgGAWjYBAC+/7G5sc/Kmzq29gbiNRiIHG4zZjhTBpjP88BYrUwpDdIM7YdZpw5I4FILeYMBxuMC9sOMxvcfLzxBkONTTRBLZbNjQ2PZ5xLZzO4nVZswXAsLbeBoJ4DBxsMeMqseQxu55hJMDYcJkZLYoMEDxuzhMHNM0RqMQBqkeZpczaQnMFDpBbJGQfbDGecSUvg5wH6JYEYv/Dztz9+8KHCJoGN/fDGGx9qbIjwC7IjJRJIUQ7RQqqOUTAKRsEoGBkAALY+RH9Hk5C4AAAAAElFTkSuQmCC","orcid":"","institution":"The Affiliated Suzhou Hospital of Nanjing Medical University","correspondingAuthor":true,"prefix":"","firstName":"Xuetao","middleName":"","lastName":"Zhu","suffix":""},{"id":509324878,"identity":"ff23c4db-4958-4c88-b902-bd8a5794a7fe","order_by":1,"name":"Ling Zhang","email":"","orcid":"","institution":"Shandong Medical College","correspondingAuthor":false,"prefix":"","firstName":"Ling","middleName":"","lastName":"Zhang","suffix":""},{"id":509324880,"identity":"a155aaf9-e658-4faa-b126-b0348f515ce3","order_by":2,"name":"Yuanqiang Zhang","email":"","orcid":"","institution":"Qilu Hospital of Shandong University","correspondingAuthor":false,"prefix":"","firstName":"Yuanqiang","middleName":"","lastName":"Zhang","suffix":""},{"id":509324882,"identity":"bdba1579-0c33-40d0-94c0-44d9fef9c29b","order_by":3,"name":"Weibing Si","email":"","orcid":"","institution":"The Affiliated Suzhou Hospital of Nanjing Medical University","correspondingAuthor":false,"prefix":"","firstName":"Weibing","middleName":"","lastName":"Si","suffix":""},{"id":509324884,"identity":"5f2612e8-8fc7-477d-8af2-9bb98cfac24d","order_by":4,"name":"Dejian Liu","email":"","orcid":"","institution":"Qilu Hospital of Shandong University","correspondingAuthor":false,"prefix":"","firstName":"Dejian","middleName":"","lastName":"Liu","suffix":""},{"id":509324886,"identity":"87fd8323-5809-44f6-9635-85b0424e7e71","order_by":5,"name":"Aibo Song","email":"","orcid":"","institution":"The Affiliated Suzhou Hospital of Nanjing Medical University","correspondingAuthor":false,"prefix":"","firstName":"Aibo","middleName":"","lastName":"Song","suffix":""}],"badges":[],"createdAt":"2025-08-05 01:53:21","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7295247/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7295247/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":90485171,"identity":"58532348-85a8-4001-a7e4-0734cfa4eb9b","added_by":"auto","created_at":"2025-09-03 08:45:31","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":237314,"visible":true,"origin":"","legend":"\u003cp\u003eOverall flow chart.\u003c/p\u003e","description":"","filename":"Figure.1.tif.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7295247/v1/f0b89dc863a9a472033d0c4f.jpg"},{"id":90485178,"identity":"e244aa14-cc29-4cc0-bbd6-d39f3cc7a100","added_by":"auto","created_at":"2025-09-03 08:45:31","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":121320,"visible":true,"origin":"","legend":"\u003cp\u003eMuscle and fat delineation in paraspinal muscles and psoas major muscles at the L4-L5 vertebral midplane（red for muscle，yellow for fat）.\u003c/p\u003e","description":"","filename":"Figure.2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7295247/v1/862df99ceb90046702801de3.jpg"},{"id":90486183,"identity":"3c78bd76-94ed-4829-80b7-1b7d60384e32","added_by":"auto","created_at":"2025-09-03 08:53:31","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":53194,"visible":true,"origin":"","legend":"\u003cp\u003eROC of indicators.\u003c/p\u003e","description":"","filename":"Figure.3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7295247/v1/6c2e749e92b9bf84bc3469b9.jpg"},{"id":90485179,"identity":"b4852417-ac86-4241-9d25-a049f3f98be0","added_by":"auto","created_at":"2025-09-03 08:45:31","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":579630,"visible":true,"origin":"","legend":"\u003cp\u003eResults of data analysis by LASSO regression method. A. LASSO regression mean squared error plot of λ values. B. LASSO regression λ value coefficient plot.\u003c/p\u003e","description":"","filename":"Figure.4.tif.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7295247/v1/5220f41a825b77d0f8126c05.jpg"},{"id":90485172,"identity":"1f1bbf58-39cb-4f99-83b2-1ecbe2d752b8","added_by":"auto","created_at":"2025-09-03 08:45:31","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":30496,"visible":true,"origin":"","legend":"\u003cp\u003eMachine Learning Predictive Modeling Results.\u003c/p\u003e","description":"","filename":"Figure.5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7295247/v1/9c705dcd58b51ef025040ef9.jpg"},{"id":90485180,"identity":"459b7075-b6ad-4be0-93b1-b8ff18036a4c","added_by":"auto","created_at":"2025-09-03 08:45:31","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":954843,"visible":true,"origin":"","legend":"\u003cp\u003eOverall and individual level SHAP analysis of the XGBoost prediction model. A.SHAP bar plot of XGBoost. B. SHAP beeswarm plot of XGBoost. C. SHAP waterfall diagram for A. \u0026nbsp;D. SHAP waterfall diagram for B. \u0026nbsp;E. SHAP force diagram of A. F. SHAP force diagram of B.\u003c/p\u003e","description":"","filename":"Figure.6.tif.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7295247/v1/31705905ff250196838ca2bd.jpg"},{"id":100942509,"identity":"9d0a2669-4b6c-4e73-9746-6e474efde401","added_by":"auto","created_at":"2026-01-23 05:10:35","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2827508,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7295247/v1/15678e6b-41e5-47f9-b8f7-edd639fb3978.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Modelling of multisegmental osteoporotic vertebral compression fracture using machine learning to analyse and predict risk factors.","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eAs the global population ages, the prevalence of osteoporosis continues to increase. One of the most common complications of osteoporosis is osteoporotic vertebral compression fracture (OVCF), the incidence of which has increased annually. According to the latest research data, the prevalence of osteoporotic vertebral compression fractures (OVCF) in the elderly population is as high as 20\u0026ndash;30% in some developed countries\u003csup\u003e\u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. In China, the incidence of osteoporotic vertebral compression fractures (OVCF) is increasing rapidly as the economy develops and the population ages. Studies have shown that OVCF patients have significantly lower quality of life scores than the general population, particularly with regard to physical functioning, mental health, and social participation. Gradual loss of the ability to care for oneself due to long-term pain and impaired mobility can place a heavy burden on society and families, as patients require more care and support from their families and society\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eA multisegmental spinal fracture is a fracture of two or more vertebrae at the same time. The pathogenesis of multisegmental spinal fractures is more complex than that of single-segment spinal fractures, as it involves the continuity and stability of multiple vertebrae. Furthermore, multisegmental fractures may be accompanied by neurological injuries, such as spinal cord compression and nerve root involvement, which increases the difficulty and risk of treatment. OVCF was previously recognized as a skeletal disorder and most studies have focused on the effects on bone density\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e,\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. Recent studies have shown that sarcopenia is more prevalent in patients with osteoporotic vertebral compression fracture (OVCF) than in patients without OVCF, and that OVCF is also significantly more prevalent in patients with sarcopenia\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e,\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. A decrease in paraspinal muscle (PM) density is one of the early clinical manifestations of sarcopenia and has gradually become more widely recognised in relation to OVCF. The paravertebral muscles, which are mainly composed of the multifidus and the erector spinae muscles, are located on both sides of the spine. They are important for spinal stability and motor function\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e,\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. The paraspinal muscles maintain the biomechanical balance of the spine through compensatory mechanisms. However, when the muscle tissue density decreases to the compensatory threshold, a double pathological effect occurs: the overall stability of the spine decreases significantly and the stress load on specific segments increases\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e. This leads to the occurrence of osteoporotic vertebral compression fracture\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. Thus, there is a strong correlation between paraspinal muscles and OVCF. However, no studies have investigated the role of paraspinal muscles in the development of multisegmental OVCF.\u003c/p\u003e\u003cp\u003eIn recent years, machine learning algorithms have become deeply integrated into the medical field. Machine learning is a subcategory of artificial intelligence, and its core mechanism involves constructing computational models that can autonomously evolve and continuously optimise diagnostic prediction models based on large amounts of medical data. There are multiple ways in which artificial intelligence can be combined with osteoporosis research\u003csup\u003e\u003cspan additionalcitationids=\"CR14 CR15\" citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. However, there are no prediction models for multisegmental OVCF in existing studies. Therefore, we employed 13 machine learning methods to develop predictive models. The aim of this study was therefore to collect a comprehensive set of CT and clinical indicators in order to identify the most sensitive indicators of multi-stage osteoporotic refracture, and to build a predictive model of OVCF through machine learning. This will enable clinicians to take targeted preventive and therapeutic measures for high-risk patients. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e shows the detailed workflow.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cp\u003e\u003cb\u003e2.1 Patients\u003c/b\u003e: In this study, data were collected from a total of 1,632 patients with osteoporotic vertebral compression fractures (OVCF) attending three tertiary-level hospitals in Shandong Province between January 2015 and October 2022. A total of 298 patients who met the inclusion and exclusion criteria were screened. The selected patients had a complete medical history and imaging and test data.\u003c/p\u003e\u003cp\u003eInclusion criteria:\u003c/p\u003e\u003cp\u003e(1)A discharge diagnosis including osteoporosis, spinal compression fracture, or osteoporotic vertebral compression fracture;(2) Bone densitometry findings indicating decreased bone mass or osteoporosis;(3) Imaging reports indicating new vertebral compression fractures of the spine.\u003c/p\u003e\u003cp\u003eExclusion criteria:\u003c/p\u003e\u003cp\u003e(1) Inadequate medical history or imaging results to determine whether the vertebrae in the fractured segment of the spine are newly fractured;(2) A history of primary or secondary spinal tumours, spinal tuberculosis or spinal infection.(3) History of surgery involving the implantation of internal spinal fixation devices;(4) Vertebral compression fractures that are clearly caused by other diseases or violent trauma;(5) Unclear or unavailable CT images.\u003c/p\u003e\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.2 Clinical Data Collection\u003c/h2\u003e\u003cp\u003eIn this study, we obtained the study data through the relevant information systems of three tertiary hospitals in Shandong Province. Various types of clinical data were collected from 298 patients using a retrospective study method and strict inclusion and exclusion criteria. This data covered basic patient information as well as the results of several laboratory and imaging examinations. The basic information included name, gender, hospital number and age. The laboratory tests included erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), parathyroid hormone (PTH), total blood pressure (TBP) and body mass index (BMI). The following were also collected: 25-hydroxy vitamin D (25-OH-vitD), N-terminal and midcourse osteocalcin (N-MIDOS), β-collagen degradation products (β-CrossLaps), total type I collagen amino-terminal extender peptide, and β-collagen degradation products. Collagen amino-terminal propeptide of type I collagen (PINP), blood calcium (Ca), blood phosphorus (P) and blood magnesium (Mg); albumin (ALB); creatinine (Cr); prealbumin (PAB); bile acid (BA); and cystatin C (Cys-C) were collected. Bone mineral density (BMD) and spinal CT images were also collected. BMD and CT images of the spine were collected. In addition, information on the patient's past history of OVCF and underlying diseases was collected.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.3 Image Acquisition\u003c/h2\u003e\u003cp\u003eSpinal CT scanning using a 64-channel spiral CT scanner was carried out on all of the patients included in this study, and the resulting data were saved in DICOM format. To ensure data quality, data containing noise or motion artefacts was excluded. The specific parameters used for CT scanning were as follows: electron tube current of 300 mA, electron tube voltage adjusted within the range of 70\u0026ndash;140 kV, and slice thickness of 0.4 mm.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e2.4 Image Processing\u003c/h2\u003e\u003cp\u003eComputed tomography (CT) is becoming an increasingly popular research tool for studying aspects of skeletal muscle biology in vivo. Skeletal muscle fat content can be quantified using X-ray attenuation properties, and radiation attenuation values (in Hounsfield units, or HU) are negatively correlated with the degree of muscle degeneration. Pathological changes in the paraspinal muscles most often occur between the L4 and L5 vertebrae. To best represent the degree of degeneration of the paraspinal muscles, we selected an axial image of the middle plane of the L4-L5 vertebrae from each patient's CT scan. We then manually sketched the extent of the paraspinal muscles and the psoas major muscle (PMM) in this cross-section. The software automatically recognised the muscles and fat, and the cross-sectional areas of the muscle and intermuscular fat were output. Spinal CT images from 298 patients were analysed using SliceOmatic version 5.The specific operations are briefly described below. As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, the axial images of the middle plane of the L4-L5 vertebral body were first selected. The ranges of the paravertebral muscles and the lumbar major muscle were then sketched out. The partition calculation was carried out by setting the appropriate threshold ranges so that the muscular and fatty tissues could be automatically identified. The threshold ranges were set to -29 HU to 150 HU for muscular tissues and \u0026minus;\u0026thinsp;190 HU to -30 HU for fatty tissues. The software automatically calculates the cross-sectional area (CSA) and average CT value of corresponding tissues and outputs these as paraspinal muscle CSA (PMCSA), paraspinal interosseous fat CSA (PMFCSA) and fat CSA (FCSA). The outputs were the paraspinal muscle cross-sectional area (PMCSA), the paraspinal intermuscular fat cross-sectional area (PMFCSA), the psoas major muscle cross-sectional area (PMMCSA), the psoas major intermuscular fat cross-sectional area (PMMFCSA), the paravertebral muscle density (PMD) and the psoas major muscle density (PMMD). The fat infiltration ratio was also calculated. The fat infiltration ratio (FIR), the paraspinal muscle fat infiltration ratio (PMFIR), and the psoas major muscle fat infiltration ratio (PMMFIR) were calculated as follows: PMFIR\u0026thinsp;=\u0026thinsp;PMFCSA/(PMCSA\u0026thinsp;+\u0026thinsp;PMFCSA) \u0026times; 100%; PMMFIR\u0026thinsp;=\u0026thinsp;PMMFCSA/(PMMCSA\u0026thinsp;+\u0026thinsp;PMMFCSA) \u0026times; 100%.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\u003ch2\u003e2.5 Descriptive Statistical Analysis\u003c/h2\u003e\u003cp\u003eThe Kolmogorov\u0026ndash;Smirnov test was used to test the numerical variables for conformity to a normal distribution. Variables that conformed to a normal distribution were presented as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation (mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD), while those that did not conform to a normal distribution were presented as median and interquartile range (median, IQR). Categorical variables were presented as counts and percentages (n,%). The t-test was used for numerical variables that conformed to a normal distribution and were consistent with the chi-squared test. The non-parametric rank sum test was used for other numerical variables and the chi-squared test for categorical variables. Variables that were statistically significant after LASSO regression screening were included in multifactor logistic regression analysis. The odds ratio (OR) and 95% confidence interval (CI) were used as indicators of effect. Risk factors for the occurrence of multisegmental osteoporotic vertebral compression fracture (OVCF) were identified based on the significance of the differences and the regression coefficients of the analysed results.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e2.6 Predictive Model Construction、Performance Evaluation and Interpretability\u003c/h2\u003e\u003cp\u003eAfter screening the risk factors, the original dataset was divided using the random sampling method in a 7:3 ratio. Seventy per cent of the data was used to construct the model and 30 per cent was used to validate it internally. Thirteen commonly used machine learning algorithms were used to construct the prediction model, including Decision Tree, Random Forest, AdaBoost, GBDT, CatBoost, KNN, Extra Trees, ANN, SVM, XGBoost, LightGBM, Simple Bayes and Logistic Regression. The prediction model was then evaluated using the test set. The neural network, SVM, XGBoost, LightGBM, Simple Bayes and Logistic Regression algorithms were used to construct the prediction model on the training set data, and the model was then evaluated using the test set. When evaluating the performance of machine learning models, we use important metrics such as accuracy, precision, recall, the F1 score and the AUROC to show how well the models perform in different areas. The machine learning model's prediction results are finally interpreted using the SHAP (Shapley Additive Explanations of Additive Features) method.\u003c/p\u003e\u003c/div\u003e"},{"header":"3. Results","content":"\u003cp\u003e\u003cstrong\u003e3.1Comparative Results of Patient Data Groups\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGeneral information was compared between patients in the single-segment and multi-segment groups (Table 1). As can be seen from the table, the only significant difference in the baseline data of the two groups was in the history of previous OVCF (P=0.036), while the history of the underlying disease was not significantly different between the two groups (P\u0026gt;0.1).\u003c/p\u003e\n\u003cp\u003eA comparison of the two groups' admission laboratory indicators, examination results and muscle measurement parameters (Table 2) revealed significant differences in a variety of factors. To further evaluate the predictive value of these indicators for OVCF with multiple segments, ROC curves were plotted (Fig 3) and the area under the curve (AUC) was calculated.\u0026nbsp;The AUC values for each indicator were then ranked according to their magnitude.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2 LASSO Regression\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe collected data were analysed through the LASSO regression method, selecting statistically significant variables in the between-group comparison analysis by applying the LASSO regression model, and selecting the optimal regularization parameter λ through the cross-validation method(Fig.4A).\u0026nbsp;The regression coefficients of each variable were obtained by substituting the optimal λ values into the LASSO regression and analysing it on the training set. Following a thorough analysis of the regression coefficients, the risk factors with significant effects on multisegmental OVCF were identified(Fig.4B).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.3 Multivariate Logistic Regression\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe variables identified by LASSO regression were included in a multivariate logistic regression analysis. This analysis showed that T-P1NT levels and the rate of fat infiltration in the paravertebral muscles significantly affected the recurrence of multisegmental osteoporotic vertebral compression fractures (Table.3).\u0026nbsp;Consequently, total type I collagen amino-terminal prolongation peptide (T-PINT) and paravertebral muscle fat infiltration rate (PMFIR) have been identified as independent risk factors for the development of multisegmental OVCF.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.4\u003c/strong\u003e\u003cstrong\u003eMachine Learning Predictive Model Construction and Evaluation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn this study, total type I collagen amino-terminal extender peptide (T-PINT) and paravertebral muscle fat infiltration rate (PMFIR) were selected for the construction of a model and its application to 13 machine learning algorithms. The selection of algorithms included decision trees, random forests, AdaBoost, GBDT, CatBoost, KNN, Extra Trees, ANN, SVM, XGBoost, LightGBM, Simple Bayes, and Logistic Regression. The calculation of each algorithm's accuracy, recall, precision, F1 score and AUROC is conducted independently.\u0026nbsp;Following a thorough evaluation, it was determined that the XGBoost model demonstrated the most effective performance and was thus identified as the optimal machine learning algorithm for predicting multi-segment OVCF(Fig.5).\u0026nbsp;The XGBoost model performed the best with an accuracy of 0.944, precision of 0.945 and AUROC of 0.958.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.5\u003c/strong\u003e\u003cstrong\u003eSHAP Analysis of Predictive Models\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn this study, SHAP analysis was employed to facilitate a more intuitive interpretation of the XGBoost prediction model at both the aggregate and individual levels, respectively. At the aggregate level, the SHAP values for each risk factor were calculated to quantify the degree of dependence of the model on different risk factors for prediction. The most significant features were identified as total type I collagen amino-terminal extender peptide and paraspinal muscle fat infiltration rate(Fig.6 A.B). At the same time, we label the two samples as 'A' and 'B'. We then interpret the results separately using waterfall and force diagrams(Fig.6 C-F).\u003c/p\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eAs the population ages and the average life expectancy increases, the incidence of osteoporosis is on the rise. Among the complications caused by osteoporosis, OVCF is particularly prevalent and a major health concern worldwide\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. The occurrence of osteoporotic vertebral compression fracture (OVCF) causes long-term back pain in the elderly\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e. This not only seriously affects their mobility, but also leads to limitations in their daily activities, greatly reducing their quality of life. To treat the disease and relieve the pain, patients must incur significant medical expenses, which undoubtedly places a heavy financial burden on their families. Complications such as re-fracture and spinal cord compression are more common with multi-segment OVCF than with single-segment OVCF\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. Surgical treatment requires intervention on multiple vertebrae and is therefore more difficult. It also carries a higher risk of complications and refracture. Current clinical studies on multisegmental osteoporotic vertebral compression fracture (OVCF) mostly focus on treatment and prognosis, with little analysis of risk factors. In this study, we employed statistical methods and multifactorial logistic regression to identify total amino-terminal-propeptide of type I collagen and paravertebral muscle fat infiltration rate as risk factors for the development of multisegmental OVCF. These factors were then incorporated into the construction of a prediction model.\u003c/p\u003e\u003cp\u003eBone turnover markers are important indicators that reflect the state of bone metabolism\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. They assist in the diagnosis, differential diagnosis, treatment and evaluation of the efficacy of osteoporosis\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e. Under normal physiological conditions, the human skeleton maintains a dynamic balance of bone turnover. A number of bone metabolites reflecting the activity of osteoblasts and osteoclasts are produced in this process, and these are also known as bone turnover markers. Previous studies have reported that re-fracture is influenced by various risk factors. In the present study, the focus was on type I collagen amino-terminal extender peptide and paravertebral muscle fat infiltration rate as the main influences on re-fracture in osteoporotic vertebral compression fracture (OVCF)\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. In diagnosing and treating osteoporosis, bone turnover markers play a very important role\u003csup\u003e\u003cspan additionalcitationids=\"CR23\" citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e. Changes in their levels in the patient's serum can be used to assess treatment effectiveness and predict osteoporotic fractures\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e. Currently, type I collagen amino-terminal extender peptide is generally recommended internationally as the preferred bone formation marker for evaluating the bone formation process, as well as for diagnosing and studying related diseases\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e. Some studies have shown that type I collagen amino-terminal extender peptide is an independent risk factor for re-fracture after PKP in patients with OVCF\u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e. It has also been demonstrated that, in patients with osteoporotic vertebral compression fractures who experience delayed fracture healing following vertebroplasty, the expression level of serum type I collagen amino-terminal elongation peptide is increased in the early postoperative period\u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e. There is also a certain correlation with prognostic indexes, which can be used as an additional indicator to predict the occurrence of delayed fracture healing in OVCF patients following surgery.\u003c/p\u003e\u003cp\u003eThe paravertebral muscles are primarily made up of the multifidus and the erector spinae muscles. These muscles play an important role in maintaining spinal stability and movement. Past studies have found a strong association between muscle mass and osteoporosis, and muscle mass can be used to predict osteoporosis to some extent\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e,\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e. When these relationships were considered in the context of osteoporosis and fracture risk, it was found that muscle and fat were associated with each other with regard to fracture risk in patients with osteoporosis. In a study by Cheng et al., it was demonstrated that low paraspinal muscle lean mass at the L4 level was an independent predictor of adjacent vertebral fracture in patients treated with PKP for OVCF\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e. The severity of OVCF was found to be correlated with the degree of fat infiltration in the paraspinal muscles; higher levels of infiltration were associated with greater vertebral compression\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e. These studies further demonstrate the strong association between paravertebral muscles and OVCF. Based on this, this study collected CT images of patients with OVCF and processed them using software to identify and output the paraspinal muscle fat infiltration rate and density, and the lumbar major muscle fat infiltration rate and density. Ultimately, it was determined that an elevated paraspinal muscle fat infiltration rate is a risk factor for multisegmental OVCF.\u003c/p\u003e\u003cp\u003eIn recent years, machine learning has become widely used in various medical disciplines. It can help us to predict disease trends, screen high-risk groups, and guide prevention strategies, all of which play an important role in accelerating the development of biomedicine. Machine learning algorithms can produce the best model predictions with the fewest risk factor models. They can also find optimal subsets of features, significantly improving model utility\u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e,\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e. With machine learning algorithms, we can accurately map various features to specific classifications. This mapping operation allows us to make predictions that are useful in practice. In our study, the total type I collagen amino-terminal extension peptide and the paravertebral muscle fat infiltration rate were selected for inclusion in the prediction model using decision trees, random forests, AdaBoost, GBDT, CatBoost, KNN, Extra Trees, ANN, SVM, XGBoost, LightGBM, the plain Bayesian algorithm and logistic regression. Thirteen commonly used machine learning algorithms were employed to construct the prediction model. By comparing the predictive efficacy of each algorithm, XGBoost was selected to build the model for multisegmental OVCF due to its high probability of correct prediction and good differentiation of patients with the condition. Machine learning algorithms can capture nonlinear, high-dimensional relationships between predictors and provide an accurate, personalised approach to predicting the re-fracture of osteoporotic vertebral fractures. In our study, we constructed a machine learning model and a logistic regression (LR) model, finding that the machine learning model had higher area under the curve (AUC) values compared to the LR model.\u003c/p\u003e\u003cp\u003eThe strength of this multicentre study is that it included OVCF patients from three tertiary hospitals, enhancing the applicability of the results. This study innovatively proposed that the fat infiltration rate of the paravertebral muscle is closely related to the occurrence of multisegmental OVCF. Thirteen machine learning models were used in the modelling process. After comparing the predictive performance of each model, XGBoost was chosen to construct a prediction model for multisegmental OVCF, demonstrating the value of studying XGBoost in this context. SHAP analysis improved the interpretability of the prediction model, enabling clinicians to make more informed choices about whether to trust and adopt the model's predictions. However, there are limitations to this study. As a retrospective study, selective bias may occur. The sample size is relatively small and the evidence is insufficient. A large number of samples are needed for further study. Although we evaluated a large number of variables, the influence of confounding factors that were not included due to poorly documented patient histories cannot be ignored. Furthermore, the model's performance has not been validated by an external dataset. To address these limitations, subsequent prospective multicentre cohort studies are required to incorporate a more comprehensive set of variables in order to build an optimisation model and thus verify the generalisation efficacy of the current findings.\u003c/p\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eThis study identified multiple factors as potential risk indicators for multisegmental osteoporotic vertebral compression fracture (OVCF). Through LASSO regression and multifactorial logistic regression, the fat infiltration rate of the paravertebral muscle and the total type I collagen amino-terminal extension peptide were selected as predictive indexes to be included in the machine learning model. The XGBoost model was ultimately chosen as it demonstrated excellent predictive performance. Meanwhile, SHAP analysis visualised and interpreted the model, indicating that the fat infiltration rate of the paravertebral muscle was the most important feature for prediction. This provides clinicians with sufficient information to understand and apply the model.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics Approval and Consent to Participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was approved by the ethics committee of Qilu Hospital. Informed consent was obtained from all participants.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported partly by the National Natural Science Foundation of China (82402758).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor\u0026rsquo;s contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eXuetao Zhu designed the experiments., Dejian Liu and Ling Zhang acquired the data. Yuanqiang Zhang performed the statistical analyses. Aibo Song and Weibing Si edited the manuscript. All authors drafted and reviewed the manuscript..\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets generated and/or analyzed during the current study are available from the corresponding authors on reasonable request.\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eEnsrud KE, Blackwell TL, Fink HA, et al. What Proportion of Incident Radiographic Vertebral Fractures in Older Men Is Clinically Diagnosed and Vice Versa: A Prospective Study. \u003cem\u003eJ Bone Miner Res\u003c/em\u003e 2016; \u003cstrong\u003e31\u003c/strong\u003e(8): 1500\u0026ndash;3.\u003c/li\u003e\n\u003cli\u003eFink HA, Milavetz DL, Palermo L, et al. What proportion of incident radiographic vertebral deformities is clinically diagnosed and vice versa? \u003cem\u003eJ Bone Miner Res\u003c/em\u003e 2005; \u003cstrong\u003e20\u003c/strong\u003e(7): 1216\u0026ndash;22.\u003c/li\u003e\n\u003cli\u003eMiddleton ET, Steel SA. Routine versus targeted vertebral fracture assessment for the detection of vertebral fractures. \u003cem\u003eOsteoporos Int\u003c/em\u003e 2008; \u003cstrong\u003e19\u003c/strong\u003e(8): 1167\u0026ndash;73.\u003c/li\u003e\n\u003cli\u003eWang O, Hu Y, Gong S, et al. A survey of outcomes and management of patients post fragility fractures in China. \u003cem\u003eOsteoporos Int\u003c/em\u003e 2015; \u003cstrong\u003e26\u003c/strong\u003e(11): 2631\u0026ndash;40.\u003c/li\u003e\n\u003cli\u003eZvekic-Svorcan J, Aleksic J, Jankovic T, et al. Capture the vertebral fracture: Risk factors as a prediction. \u003cem\u003eJ Back Musculoskelet Rehabil\u003c/em\u003e 2019; \u003cstrong\u003e32\u003c/strong\u003e(2): 269\u0026ndash;76.\u003c/li\u003e\n\u003cli\u003eJi C, Rong Y, Wang J, et al. Risk Factors for Refracture following Primary Osteoporotic Vertebral Compression Fractures. \u003cem\u003ePain Physician\u003c/em\u003e 2021; \u003cstrong\u003e24\u003c/strong\u003e(3): E335\u0026ndash;e40.\u003c/li\u003e\n\u003cli\u003eWang WF, Lin CW, Xie CN, et al. The association between sarcopenia and osteoporotic vertebral compression refractures. \u003cem\u003eOsteoporos Int\u003c/em\u003e 2019; \u003cstrong\u003e30\u003c/strong\u003e(12): 2459\u0026ndash;67.\u003c/li\u003e\n\u003cli\u003eLidar S, Salame K, Chua M, et al. Sarcopenia Is an Independent Risk Factor for Subsequent Osteoporotic Vertebral Fractures Following Percutaneous Cement Augmentation in Elderly Patients. \u003cem\u003eJ Clin Med\u003c/em\u003e 2022; \u003cstrong\u003e11\u003c/strong\u003e(19).\u003c/li\u003e\n\u003cli\u003eTakahashi S, Hoshino M, Ohyama S, et al. Relationship of back muscle and knee extensors with the compensatory mechanism of sagittal alignment in a community-dwelling elderly population. \u003cem\u003eSci Rep\u003c/em\u003e 2021; \u003cstrong\u003e11\u003c/strong\u003e(1): 2179.\u003c/li\u003e\n\u003cli\u003eBanno T, Arima H, Hasegawa T, et al. The Effect of Paravertebral Muscle on the Maintenance of Upright Posture in Patients With Adult Spinal Deformity. \u003cem\u003eSpine Deform\u003c/em\u003e 2019; \u003cstrong\u003e7\u003c/strong\u003e(1): 125\u0026ndash;31.\u003c/li\u003e\n\u003cli\u003eLee DG, Bae JH. Fatty infiltration of the multifidus muscle independently increases osteoporotic vertebral compression fracture risk. \u003cem\u003eBMC Musculoskelet Disord\u003c/em\u003e 2023; \u003cstrong\u003e24\u003c/strong\u003e(1): 508.\u003c/li\u003e\n\u003cli\u003eWang M, Liu J, Chen X, Wang X, Chen X. Muscle quality and spine fractures. \u003cem\u003eJ Cachexia Sarcopenia Muscle\u003c/em\u003e 2022; \u003cstrong\u003e13\u003c/strong\u003e(2): 1426\u0026ndash;8.\u003c/li\u003e\n\u003cli\u003eZhu X, Liu D, Liu L, et al. Fully Automatic Deep Learning Model for Spine Refracture in Patients with OVCF: A Multi-Center Study. \u003cem\u003eOrthop Surg\u003c/em\u003e 2024; \u003cstrong\u003e16\u003c/strong\u003e(8): 2052\u0026ndash;65.\u003c/li\u003e\n\u003cli\u003eSmets J, Shevroja E, H\u0026uuml;gle T, Leslie WD, Hans D. Machine Learning Solutions for Osteoporosis-A Review. \u003cem\u003eJ Bone Miner Res\u003c/em\u003e 2021; \u003cstrong\u003e36\u003c/strong\u003e(5): 833\u0026ndash;51.\u003c/li\u003e\n\u003cli\u003eSebro R, Elmahdy M. Machine Learning for Opportunistic Screening for Osteoporosis and Osteopenia Using Knee CT Scans. \u003cem\u003eCan Assoc Radiol J\u003c/em\u003e 2023; \u003cstrong\u003e74\u003c/strong\u003e(4): 676\u0026ndash;87.\u003c/li\u003e\n\u003cli\u003eFerizi U, Honig S, Chang G. Artificial intelligence, osteoporosis and fragility fractures. \u003cem\u003eCurr Opin Rheumatol\u003c/em\u003e 2019; \u003cstrong\u003e31\u003c/strong\u003e(4): 368\u0026ndash;75.\u003c/li\u003e\n\u003cli\u003ePatil S, Rawall S, Singh D, et al. Surgical patterns in osteoporotic vertebral compression fractures. \u003cem\u003eEur Spine J\u003c/em\u003e 2013; \u003cstrong\u003e22\u003c/strong\u003e(4): 883\u0026ndash;91.\u003c/li\u003e\n\u003cli\u003eTseng YY, Yang TC, Tu PH, Lo YL, Yang ST. Repeated and multiple new vertebral compression fractures after percutaneous transpedicular vertebroplasty. \u003cem\u003eSpine (Phila Pa 1976)\u003c/em\u003e 2009; \u003cstrong\u003e34\u003c/strong\u003e(18): 1917\u0026ndash;22.\u003c/li\u003e\n\u003cli\u003eSchini M, Vilaca T, Gossiel F, Salam S, Eastell R. Bone Turnover Markers: Basic Biology to Clinical Applications. \u003cem\u003eEndocr Rev\u003c/em\u003e 2023; \u003cstrong\u003e44\u003c/strong\u003e(3): 417\u0026ndash;73.\u003c/li\u003e\n\u003cli\u003eBrown JP, Don-Wauchope A, Douville P, Albert C, Vasikaran SD. Current use of bone turnover markers in the management of osteoporosis. \u003cem\u003eClin Biochem\u003c/em\u003e 2022; \u003cstrong\u003e109-110\u003c/strong\u003e: 1\u0026ndash;10.\u003c/li\u003e\n\u003cli\u003eZhang J, Hu Y, Cai W. Bone metabolism factors in predicting the risk of osteoporosis fracture in the elderly. \u003cem\u003eBMC Musculoskelet Disord\u003c/em\u003e 2024; \u003cstrong\u003e25\u003c/strong\u003e(1): 442.\u003c/li\u003e\n\u003cli\u003eBhattoa HP, Vasikaran S, Trifonidi I, et al. Update on the role of bone turnover markers in the diagnosis and management of osteoporosis: a consensus paper from The European Society for Clinical and Economic Aspects of Osteoporosis, Osteoarthritis and Musculoskeletal Diseases (ESCEO), International Osteoporosis Foundation (IOF), and International Federation of Clinical Chemistry and Laboratory Medicine (IFCC). \u003cem\u003eOsteoporos Int\u003c/em\u003e 2025; \u003cstrong\u003e36\u003c/strong\u003e(4): 579\u0026ndash;608.\u003c/li\u003e\n\u003cli\u003eTan RZ, Loh TP, Vasikaran S. Bone turnover marker monitoring in osteoporosis treatment response. \u003cem\u003eEur J Endocrinol\u003c/em\u003e 2020; \u003cstrong\u003e183\u003c/strong\u003e(1): C5\u0026ndash;c7.\u003c/li\u003e\n\u003cli\u003eEastell R, Szulc P. Use of bone turnover markers in postmenopausal osteoporosis. \u003cem\u003eLancet Diabetes Endocrinol\u003c/em\u003e 2017; \u003cstrong\u003e5\u003c/strong\u003e(11): 908\u0026ndash;23.\u003c/li\u003e\n\u003cli\u003eDi Medio L, Brandi ML. Advances in bone turnover markers. \u003cem\u003eAdv Clin Chem\u003c/em\u003e 2021; \u003cstrong\u003e105\u003c/strong\u003e: 101\u0026ndash;40.\u003c/li\u003e\n\u003cli\u003eLi YX, Guo DQ, Zhang SC, et al. Risk factor analysis for re-collapse of cemented vertebrae after percutaneous vertebroplasty (PVP) or percutaneous kyphoplasty (PKP). \u003cem\u003eInt Orthop\u003c/em\u003e 2018; \u003cstrong\u003e42\u003c/strong\u003e(9): 2131\u0026ndash;9.\u003c/li\u003e\n\u003cli\u003eCauley JA, Chalhoub D, Kassem AM, Fuleihan Gel H. Geographic and ethnic disparities in osteoporotic fractures. \u003cem\u003eNat Rev Endocrinol\u003c/em\u003e 2014; \u003cstrong\u003e10\u003c/strong\u003e(6): 338\u0026ndash;51.\u003c/li\u003e\n\u003cli\u003eYang S, Center JR, Eisman JA, Nguyen TV. Association between fat mass, lean mass, and bone loss: the Dubbo Osteoporosis Epidemiology Study. \u003cem\u003eOsteoporos Int\u003c/em\u003e 2015; \u003cstrong\u003e26\u003c/strong\u003e(4): 1381\u0026ndash;6.\u003c/li\u003e\n\u003cli\u003eRosen CJ, Klibanski A. Bone, fat, and body composition: evolving concepts in the pathogenesis of osteoporosis. \u003cem\u003eAm J Med\u003c/em\u003e 2009; \u003cstrong\u003e122\u003c/strong\u003e(5): 409\u0026ndash;14.\u003c/li\u003e\n\u003cli\u003eCheng Y, Yang H, Hai Y, et al. Low paraspinal lean muscle mass is an independent predictor of adjacent vertebral compression fractures after percutaneous kyphoplasty: A propensity score-matched case-control study. \u003cem\u003eFront Surg\u003c/em\u003e 2022; \u003cstrong\u003e9\u003c/strong\u003e: 965332.\u003c/li\u003e\n\u003cli\u003eHuang W, Cai XH, Li YR, et al. The association between paraspinal muscle degeneration and osteoporotic vertebral compression fracture severity in postmenopausal women. \u003cem\u003eJ Back Musculoskelet Rehabil\u003c/em\u003e 2023; \u003cstrong\u003e36\u003c/strong\u003e(2): 323\u0026ndash;9.\u003c/li\u003e\n\u003cli\u003eWu EQ, Hu D, Deng PY, et al. Nonparametric Bayesian Prior Inducing Deep Network for Automatic Detection of Cognitive Status. \u003cem\u003eIEEE Trans Cybern\u003c/em\u003e 2021; \u003cstrong\u003e51\u003c/strong\u003e(11): 5483\u0026ndash;96.\u003c/li\u003e\n\u003cli\u003eWu EQ, Zhou M, Hu D, et al. Self-Paced Dynamic Infinite Mixture Model for Fatigue Evaluation of Pilots\u0026apos; Brains. \u003cem\u003eIEEE Trans Cybern\u003c/em\u003e 2022; \u003cstrong\u003e52\u003c/strong\u003e(7): 5623\u0026ndash;38.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTable1 General information of OVCF patients\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" align=\"\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003echaracterization\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 55px;\"\u003e\n \u003cp\u003esingle \u0026nbsp;fracture \u0026nbsp;group \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; Multiple fracture group\u003c/p\u003e\n \u003cp\u003e（n=181） (n=117)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003eP\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003eGender (n,%)\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eMale\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e34(18.78)\u003c/p\u003e\n \u003cp\u003e147(81.22)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e17(14.53)\u003c/p\u003e\n \u003cp\u003e100(85.47)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e0.341\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003ePre-existing OVCF\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;history (n,%)\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eYES\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eNO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e53(29.28)\u003c/p\u003e\n \u003cp\u003e128(70.72)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e48(41.03)\u003c/p\u003e\n \u003cp\u003e69(58.97)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e0.036\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003eDiseases of the cardiovascular system (%)\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eYES\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eNO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e68(37.57)\u003c/p\u003e\n \u003cp\u003e113(62.43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e40(34.19)\u003c/p\u003e\n \u003cp\u003e77(65.81)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e0.553\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003eDiabetes (n,%)\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eYES\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eNO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e19(10.50)\u003c/p\u003e\n \u003cp\u003e162(89.50)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e8(6.84)\u003c/p\u003e\n \u003cp\u003e109(93.16)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e0.282\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003eRespiratory diseases (n,%)\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eYES\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eNO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e8(4.42)\u003c/p\u003e\n \u003cp\u003e173(95.58)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e8(6.84)\u003c/p\u003e\n \u003cp\u003e109(93.16)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e0.366\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003eRheumatologic diseases (n,%)\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eYES\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eNO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e4(2.21)\u003c/p\u003e\n \u003cp\u003e177(97.79)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e5(4.27)\u003c/p\u003e\n \u003cp\u003e112(95.73)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e0.503\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003eAge (median,IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003e69(63.5,77.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23px;\"\u003e\n \u003cp\u003e69(64.0,76.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e0.957\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003eTable.2 Imaging and laboratory characteristics of OVCF patients\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" align=\"\" width=\"100%\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003echaracterization\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 55px;\"\u003e\n \u003cp\u003esingle fracture group\u003c/p\u003e\n \u003cp\u003e（n=181)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 55px;\"\u003e\n \u003cp\u003eMultiple fracture group\u003c/p\u003e\n \u003cp\u003e(n=117)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003eP\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003eErythrocyte sedimentation rate(mm/h) (median,IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003e26(14.0,41.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 21px;\"\u003e\n \u003cp\u003e23(13.0,36.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e0.432\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003eC-reactive protein (mg/L)( median,IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003e4.230(1.9,11.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 21px;\"\u003e\n \u003cp\u003e3.230(1.2,10.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.074\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003eparathyroid hormone (pg/ml)( median,IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003e35.810(26.4,47.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 21px;\"\u003e\n \u003cp\u003e32.950(24.3,43.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.086\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003e25-hydroxy vitamin D (ng/ml)(median,IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003e17.500(13.3,24.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 21px;\"\u003e\n \u003cp\u003e21.950(16.2,29.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.000\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003eAmino-terminal and midcourse osteocalcin (ng/ml) (median,IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e16.510(12.1,22.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 21px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e19.740(13.9,26.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e0.004\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 31px;\"\u003e\n \u003cp\u003e\u0026beta;-crosslaps (ng/ml) (median,IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 33px;\"\u003e\n \u003cp\u003e0.716(0.5,0.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 21px;\"\u003e\n \u003cp\u003e0.810(0.6,1.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.007\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003eT-PINT(ng/ml) (median,IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e58.480(41.2,77.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23px;\"\u003e\n \u003cp\u003e70.04(54.4,93.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.000\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003eCalcium (mmol/L)(median,IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e2.280(2.2,2.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23px;\"\u003e\n \u003cp\u003e2.300(2.2,2.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e0.323\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003ePhosphorus (mmol/L) (mean\u0026plusmn;SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e1.18\u0026plusmn;0.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23px;\"\u003e\n \u003cp\u003e1.22\u0026plusmn;0.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.034\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003eMagnesium (mmol/L) (mean\u0026plusmn;SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e0.91\u0026plusmn;0.07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23px;\"\u003e\n \u003cp\u003e0.90\u0026plusmn;0.08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e0.411\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003eAlbumin (g/L)(median,IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e41.400(38.1,44.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23px;\"\u003e\n \u003cp\u003e40.500(37.6,43.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.057\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003eCreatinine (umol/L)(median,IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e58.000(52.0,70.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23px;\"\u003e\n \u003cp\u003e56.000(50.0,66.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.051\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003ePrealbumin (g/L) (median,IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e22.600(19.2,30.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23px;\"\u003e\n \u003cp\u003e22.300(18.3,32.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e0.511\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003eBile acid (umol/L)(median,IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e4.100(2.5,7.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23px;\"\u003e\n \u003cp\u003e4.100(2.7,7.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e0.839\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003eCystatin C (mg/L)(median,IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e0.910(0.8,1.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23px;\"\u003e\n \u003cp\u003e0.940(0.8,1.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e0.51\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003eBone mineral density (median,IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e-3.000(-3.8, -2.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23px;\"\u003e\n \u003cp\u003e-3.300(-4.3, -2.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.030\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003ePMD(HU)(median,IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e28.450(22.0,33.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 23px;\"\u003e\n \u003cp\u003e21.780(14.0,28.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.000\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003ePMMD(HU)(median,IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e41.840(37.8,45.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e38.940(34.3,43.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.001\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003ePMFIR(%) (median,IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e25.663(19.9,35.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e37.014(30.6,46.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.000\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003ePMMFIR (%)(median,IQR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29px;\"\u003e\n \u003cp\u003e2.7(1.5,5.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25px;\"\u003e\n \u003cp\u003e3.908(2.3,6.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 16px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.004\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable.3 Risk factors of multi-level OVCF determined by multivariable logistic regression\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"100%\" class=\"fr-table-selection-hover\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" style=\"width: 15.5521%;\"\u003e\n \u003cp\u003eVariable\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" rowspan=\"2\" style=\"width: 30.9487%;\"\u003e\n \u003cp\u003eRegression coefficient\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" rowspan=\"2\" style=\"width: 15.8534%;\"\u003e\n \u003cp\u003eP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" colspan=\"3\" style=\"width: 11.0809%;\"\u003e\n \u003cp\u003eOR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 26.5552%;\"\u003e\n \u003cp\u003e95%CI\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.3577%;\"\u003e\n \u003cp\u003elower limit\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.1505%;\"\u003e\n \u003cp\u003eUpper Limit\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.5521%;\"\u003e\n \u003cp\u003econstant\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 30.9487%;\"\u003e\n \u003cp\u003e-4.665\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 15.8534%;\"\u003e\n \u003cp\u003e0.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 11.0809%;\"\u003e\n \u003cp\u003e0.009\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15.3577%;\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.1505%;\"\u003e\n \u003cp\u003e0.042\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.5521%;\"\u003e\n \u003cp\u003e\u0026beta;-CROSS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 30.9487%;\"\u003e\n \u003cp\u003e0.41\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 15.8534%;\"\u003e\n \u003cp\u003e0.41\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 11.0809%;\"\u003e\n \u003cp\u003e1.508\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15.3577%;\"\u003e\n \u003cp\u003e0.565\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.1505%;\"\u003e\n \u003cp\u003e4.024\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.5521%;\"\u003e\n \u003cp\u003eT-P1NT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 30.9487%;\"\u003e\n \u003cp\u003e0.011\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 15.8534%;\"\u003e\n \u003cp\u003e0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 11.0809%;\"\u003e\n \u003cp\u003e1.011\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15.3577%;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.1505%;\"\u003e\n \u003cp\u003e1.023\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.5521%;\"\u003e\n \u003cp\u003eBMD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 30.9487%;\"\u003e\n \u003cp\u003e-0.226\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 15.8534%;\"\u003e\n \u003cp\u003e0.06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 11.0809%;\"\u003e\n \u003cp\u003e0.798\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15.3577%;\"\u003e\n \u003cp\u003e0.628\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.1505%;\"\u003e\n \u003cp\u003e1.013\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.5521%;\"\u003e\n \u003cp\u003ePMFIR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 30.9487%;\"\u003e\n \u003cp\u003e0.07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 15.8534%;\"\u003e\n \u003cp\u003e0.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 11.0809%;\"\u003e\n \u003cp\u003e1.073\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 15.3577%;\"\u003e\n \u003cp\u003e1.044\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 11.1505%;\"\u003e\n \u003cp\u003e1.103\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-7295247/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7295247/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e\u003cp\u003eThe aim is to investigate the risk factors for re-fracture of multisegmental osteoporotic vertebral compression fractures (OVCF) and to construct a clinical prediction model for their occurrence using machine learning algorithms.\u003c/p\u003e\u003ch2\u003eObjective\u003c/h2\u003e\u003cp\u003eThe aim is to investigate the risk factors for re-fracture of multisegmental osteoporotic vertebral compression fractures (OVCF) and to construct a clinical prediction model for their occurrence using machine learning algorithms.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e\u003cp\u003eThe CT images were imported into Slice-O-Matic software, and the fat infiltration rate of paraspinal and lumbar major muscles, paraspinal muscle mass, and lumbar major muscle mass were measured for each patient. The screening process was conducted through a multifaceted approach encompassing between-group comparison analysis, LASSO regression, and multivariate logistic regression. To ensure the robustness of the models, a total of 13 machine learning algorithms were employed in their construction. The prediction performance of each model was evaluated and the optimal model was selected through accuracy, recall, precision, F1 score and area under the receiver operating characteristic curve (AUROC). Finally, the models were interpreted using SHAP analysis to elucidate the importance of the features in the models and their impact on the direction of prediction.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e\u003cp\u003eIn this study, the fat infiltration rate of paravertebral muscle and total type I collagen amino-terminal extender peptide were identified as potential risk factors for the development of multisegmental OVCF. A prediction model for the occurrence of multisegmental OVCF was constructed by the XGBoost model, and the model was evaluated to have good predictive performance. SHAP analysis was utilised to enhance the interpretability of the model, thereby demonstrating the importance of paraspinal muscle fat infiltration rate and total amino-terminal-propeptide of type I collagen in the prediction of the model.\u003c/p\u003e","manuscriptTitle":"Modelling of multisegmental osteoporotic vertebral compression fracture using machine learning to analyse and predict risk factors.","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-03 08:45:27","doi":"10.21203/rs.3.rs-7295247/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"9d6d1eeb-c0af-44e8-b78a-3adbd948d707","owner":[],"postedDate":"September 3rd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-01-23T05:09:58+00:00","versionOfRecord":[],"versionCreatedAt":"2025-09-03 08:45:27","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7295247","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7295247","identity":"rs-7295247","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}