1Division of Spine Surgery, Department of Orthopedic Surgery, Faculty of Medicine Siriraj Hospital, Mahidol University, Bangkok, Thailand, 2Siriraj
Informatics and Data Innovation Center, Faculty of Medicine Siriraj Hospital, Mahidol University, Bangkok, Thailand.
ABSTRACT
0.68 postoperatively.
INTRODUCTION
Patients with cancer have a 4-6.5 times increased risk of developing venous thromboembolism (VTE), including deep venous thrombosis (DVT) and pulmonary
embolism (PE).1-3 The incidence of VTE in spine surgery was reported to range from 0.15-29.38%4, and the incidence of VTE among patients undergoing surgery for spinal metastasis was 9.48%-11%.5,6 This data demonstrates that
Corresponding author: Panya Luksanapruksa E-mail: panya.luk@mahidol.ac.th
Received 27 December 2023 Revised 16 April 2023 Accepted 26 April 2024 ORCID ID:http://orcid.org/0000-0002-9554-4259 https://doi.org/10.33192/smj.v76i6.266959
All material is licensed under terms of the Creative Commons Attribution 4.0 International (CC-BY-NC-ND 4.0) license unless otherwise stated.
patients undergoing surgery for spinal metastasis are at far greater risk for developing VTE. The development of VTE in cancer patients is associated with both increased morbidity and mortality compared to non-cancer patients3, and the cost of treatment is high.7
The two commonly used clinical assessments for VTE risk stratification among surgical patients are the Rogers score8 and the Caprini risk assessment model.9 The current models’ lack of specificity and limited predictive accuracy for this subgroup necessitate a more focused and refined approach to risk assessment.10,11
Enter the realm of machine learning (ML), a subdivision of artificial intelligence that thrives on its capacity to decipher complex patterns within large datasets, a challenge often faced in medical research. ML’s inherent capability to integrate and analyze multifaceted clinical data presents a promising avenue to enhance predictive models’ accuracy significantly. Previous research has validated ML’s potency in predicting adverse outcomes in spine surgery, setting a precedent for its application in VTE risk prediction.12
The urgency for advanced predictive analytics is clear: the current gap in precise VTE risk assessment for spinal metastasis surgery patients could be narrowed by employing machine learning models. By developing various MLM algorithms, this study aims to address this gap, offering potentially more accurate predictions for preoperative and postoperative VTE. Not only could this lead to better patient outcomes, but it also stands to optimize the use of healthcare resources and inform preventive strategies, thereby ameliorating the burden of VTE in oncological surgery. In doing so, we align our efforts with the evolving landscape of precision medicine, where patient care is increasingly informed by data-driven insights.
MATERIALS AND METHODS
Guidelines
We followed the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) guidelines, and the Guidelines for Developing and Reporting Machine Learning Models in Biomedical Research.13,14
Source of data
The data were obtained from the medical database of the Department of Orthopedic Surgery. We identified consecutive patients retrospectively who underwent spinal surgery between January 2009 and December 2020.
Participants
Participants were selected based on the following criteria: (1) a confirmed diagnosis of spinal metastases classified under the International Classification of Diseases, 10th Revision, Thai Modification (ICD 10-TM) with the specific codes C79.5 and C79.8, (2) being 18 years of age or older, and (3) a documented history of undergoing surgical intervention for spinal metastases located in the cervical, thoracic, lumbar, or sacral regions, which was categorized under the International Classification of Diseases, 9th Revision, Clinical Modification (ICD 9-CM) with the procedural codes 03.0, 03.4, 03.09, 81.0, and the range of 81.00 to 81.08.
We excluded patients with a history of VTE, who had major surgery or major trauma within one month before surgery, who were pregnant, or who had a genetic abnormality, such as anticoagulation proteins C and S, antithrombin III deficiency, or factor V Leiden mutation.
Outcome and predictive variables
The primary outcome of our study was the incidence of postoperative venous thromboembolism in patients with spinal metastasis at 30- and 90-days following surgery.
The following risk factor variables for VTE were identified from previous studies: gender15,16, age15-19, diabetes mellitus (DM)20, hypertension (HT)16, chronic kidney disease (CKD)21,22, American Society of Anesthesiologists (ASA) classification23, body mass index (BMI)15,24,25, preoperative ambulation status15,19, preoperative albumin level23, preoperative hemoglobin level5, preoperative partial thromboplastin time (PPT)5, spinal surgery location15, operative time6,23, and postoperative blood transfusion.26-28 We divided the variables into two groups: the first
one using only pre-operative data of patients, and the latter including intra-operative and post-operative data such as blood loss, operative time, etc. Patients with symptomatic venous VTE, as documented in the medical database, were used as the dependent variable in this investigation.
Preprocessing
Multiple imputations with chained equations imputed missing preoperative laboratory characteristics with less than 25% missing data. A class weighting strategy was also used to ensure that the trained model would take each class into account equally despite class imbalance. To reduce the influence of different variable units and quantity levels, we scale numerical variables to a standard
deviation of one and a mean of zero, and we employ dummy encoding for categorical variables. We also get rid of outliers whose lab values are three standard deviations from the average lab value at our hospital.
Algorithm training and validation
For the development and evaluation of our models, we opted for a suite of algorithms including random forest, neural network, logistic regression, gradient boosted tree, support vector machine, XGBoost, decision tree, and stochastic gradient descent. These were implemented using the Python 3.9 programming language and the Scikit-learn 1.0.1 library, which is distributed under a permissive open-source BSD license.
In the process of preparing the dataset for analysis, we performed both manual and automated tuning of hyperparameters via grid and random search methods to ensure the highest predictive accuracy during a fivefold internal cross-validation for each algorithm. The data were split into an 80:20 ratio for training and testing purposes, with the training set being utilized to fit the models, which were then validated through a fivefold cross-validation approach. To counteract any imbalance across the classes, a class weighting technique was employed during model training
The efficacy of the algorithms was assessed using the test set, scrutinizing metrics such as the area under the receiver operating characteristic (ROC) curve (AUC), the F1-score, overall accuracy, and calibration loss. The ROC curve is critical for illustrating the trade-off between the true positive and false positive rates that arise from various thresholds in the predictive model. An AUC score between 0.7 and 0.8 denotes fair model performance, while a score above 0.8 is indicative of good performance. The F1-score—a harmonic mean of precision and recall—reaches its optimal value at 1.0, representing impeccable precision and sensitivity. To evaluate a model’s precision, we also analyzed the confusion matrix, which juxtaposes actual versus predicted outcomes. Given that different metrics may present certain trade-offs, such as the balance between precision and recall, the AUC was primarily used for selecting the most suitable model for practical application.
Statistical analyses
The baseline demographic and clinical characteristics of study patients were summarized using descriptive statistics. Continuous data, such as partial thromboplastin time and operative time, are presented as mean plus/
minus standard deviation. All other evaluated factors are reported as number and percentage. Study data were recorded and calculated using Microsoft Excel software (Microsoft Corporation, Redmond, WA, USA).
RESULTS
A total of 334 patients were included in this study. The mean age of study patients was 57.6 years, and 191 (57.2%) of them were men. The postoperative prevalence of VTE within 30 and 90 days of surgery was 30 (8.98%)
and 45 (13.47%), respectively (PE 20%, and DVT 80%). Patients who developed VTE were younger, had lower BMI, were not able to ambulate preoperatively, had a higher albumin level, and had a higher hemoglobin level. Other baseline demographic/clinical characteristics and patient outcomes are shown in Table 1. The most common primary tumor sites were lung (27.5%), breast (17.4%), and prostate gland (12.9%) (Table 2).
Regarding VTE prediction within 30 days before spinal surgery, the gradient boosted tree algorithm showed age, preoperative ambulatory status, BMI, albumin level, hemoglobin level, and PTT to be predictive factors (Fig 1). This algorithm had the best AUC (0.77) with accuracy (proportion of correct predictions within the test set) of 0.88, precision (proportion of positive predictions that were indeed positive) of 1, recall (proportion of actual positive values found by the classifier) of 0.11, and F1 score (harmonic mean between precision and recall) of
0.2 (Table 3).
Concerning VTE prediction within 90 days before spinal surgery, the support vector machine algorithm had the best AUC (0.72) with accuracy, precision, recall, and F1 scores of 0.58, 0.25, 0.82, and 0.38, respectively. Regarding VTE prediction within 30 days after spinal surgery, the gradient boosted tree algorithm showed age, BMI, albumin level, hemoglobin level, PTT, and operative time to be predictive factors (Fig 2). This algorithm had the best AUC (0.71) with accuracy, precision, recall, and
F1 scores of 0.88, 0.1, 0.11, and 0.2, respectively.
Concerning VTE within 90 days after spinal surgery, the support vector machine algorithm has the best AUC (0.68) with accuracy, precision, recall, and F1 scores of 0.86, 1.00, 0.09, and 0.17, respectively.
DISCUSSION
This study including cancer patients with spinal metastasis who underwent spinal surgery. The incidence of postoperative VTE events was 13.5%. Zacharia, et al.5
TABLE 1. Baseline Characteristics of Patients Undergoing Surgery for Spinal Metastatic Disease, n=334
Sex Male 191 (57.2%) 178 (58.6%) 13 (43.3%) 171 (59.2%) 20 (44.4%)
Age (Years)
<65
65-79
>79
241 (72.2%) 218 (71.7%) 23 (76.7%)
86 (25.7%) 79 (26%) 7(23.3)
7 (2.1%) 7 (2.3%) 0
206 (71.3%) 35 (77.8%)
76(26.3%) 10 (22.2%)
7(2.4%) 0
Female 143 (42.8%) 126 (41.4%) 17 (56.7%) 118 (40.8%) 25 (55.6%)
Body mass index
<25 258 (77.2%) 233 (76.6%) 25 (83.3%) 223 (77.2%) 35 (77.8%)
(kg/m2) 25-30 62 (18.6%) 59 (19.4%) 3 (10%) 55 (19%) 7 (15.6%)
American Society of Anesthesiologists
Classification
1-2
195 (58.4%) 181 (59.5%) 14 (46.7%)
171 (59.2%) 24 (53.3%)
3-5
139 (41.6%)
123 (40.5%)
16 (53.3%)
118 (40.8%)
21 (46.7%)
>30 14 (4.2%) 12 (4%) 2 (6.7%) 11 (3.8%) 3 (6.6%)
Diabetes 51 (15.3%) 46 (15.1%) 5 (16.7%) 44 (15.2%) 7 (15.6%)
Comorbidities Hypertension 119 (35.6%) 106 (34.9%) 13 (43.3%) 99 (83.2%) 20 (16.8%)
Chronic kidney 15 (4.5%) 14 (4.6%) 1 (3.3%) 14 (4.8%) 1 (2.2%)
Preoperative ambulatory
activity status
No
Yes
187 (56%)
147 (44%)
165 (54.3%) 22 (73.3%)
139 (45.7%) 8 (26.7%)
154 (53.3%) 33 (73.3%)
135 (46.7%) 12 (26.7%)
disease
Albumin (g/dL)
<3.5 80 (24%) 74 (24.3%) 6 (20%) 71 (24.6%) 9 (20%)
Hemoglobin
(g/dL)
<10
≥10
40 (12%)
294 (88%)
35 (11.5%) 5 (16.7%)
269 (88.5%) 25 (83.3%)
34 (11.8%) 6 (13.3%)
255 (88.2%) 39 (86.7%)
≥3.5 249 (76%) 230 (75.7%) 24 (80%) 218 (75.4%) 36 (80%)
Index spinal
location
Cervical Thoracic Lumbar
Sacral
33 (9.9%) 26 (8.6%) 7 (23.3%)
186 (55.7%) 175 (57.6%) 11 (36.7%)
112 (33.5%) 100 (32.8%) 13 (40%)
3 (0.9%) 3 (1.0%) 0 (0%)
25 (8.7%) 8 (17.8%)
167 (57.8%) 19 (42.2%)
94 (32.5%) 18 (40%)
3 (1.0%) 0 (0%)
Partial thromboplastin Time (Sec.)
26±3.6 26.12±3.63 25.03±3.88 26.14±3.65 25.21±3.72
Postoperative
blood transfusion
243 (72.8%) 221 (72.7%) 22 (73.3%)
209 (72.3%) 34 (75.6%)
Operative Time (min.) 192.04±65.93 192.49±65.66 187.50±69.60 193.32±65.58 183.89±68.29
TABLE 2. Origin of the primary tumor (n=334)
Characteristic | N (%) |
Lung | 92 (27.5) |
Breast | 58 (17.4) |
Prostate | 43 (12.9) |
Liver | 21 (6.3) |
Kidney | 17 (5.1) |
Colorectal | 14 (4.2) |
Thyroid | 14 (4.2) |
Hematology | 12 (3.6) |
CholangioCA | 11 (3.3) |
Nasopharyngeal cancer | 9 (2.7) |
Cervix | 6 (1.8) |
Unknown | 8 (2.4) |
Others* | 29 (8.7) |
*pancreatic cancer 4 (1.2%), soft tissue sarcoma 4 (1.2%), Endometrial
cancer 3 (0.9%), tongue cancer 3 (0.9%), bladder cancer 2 (0.6%),
esophagus cancer 2 (0.3%), neuroendocrine tumors 1 (0.3%) and
hard palate cancer 1 (0.3%).
conducted preoperative DVT screening in patients who were to undergo surgical treatment for spinal metastasis, and they found a 9.48% preoperative incidence of DVT. Groot, et al.6 reported that 11% of patients developed symptomatic VTE within 90 days of surgical treatment for spinal metastasis. Based on the results of the present study, the most accurate model for predicting VTE within 30 days before or after surgery is the gradient boosted tree algorithm (AUC for preoperative prediction: 0.77, and AUC for postoperative prediction: 0.71). We also found the most accurate model for predicting VTE within 90 days before or after surgery to be the support vector machine algorithm (AUC for preoperative prediction: 0.72, and AUC for postoperative prediction: 0.68). The most important variables in these algorithms are age, BMI, preoperative ambulatory status, albumin level, hemoglobin level, PTT, and operative time. These factors are consistent with previously published studies that reported risk factors for VTE after spine surgery.5,15,18,19,23,24,26 Other studies have investigated models that predict complications after spine surgery. Bekelis, et al.29 conducted a retrospective cohort study that included 13,660 patients. The 30-day incidence of DVT was only 0.6%. Increasing age, alcohol consumption, preoperative neurologic deficit, and corpectomy were found to be significantly associated with a higher likelihood of postoperative DVT. Their model to predict postoperative risk of DVT had an AUC of 0.74. Han, et al.12 developed machine learning models for predicting adverse events following spine surgery. The incidence of DVT in their study was a low 1.8%; however, the AUC for an overall AE was 0.7 among the six individual prediction models that they had developed.
Fig 1. Variable importance for the preoperative prediction model.
TABLE 3. Machine Learning Model Performance for 30 and 90 days VTE Prediction in Patients Undergoing Surgery for Spinal Metastatic
AUC | 0.62 | 0.64 | 0.59 | 0.65 | 0.77 | 0.59 | 0.63 | 0.69 |
Accuracy | 0.88 | 0.87 | 0.88 | 0.87 | 0.88 | 0.87 | 0.87 | 0.87 |
Precision | 1 | 0.5 | 1 | 1 | 1 | 1 | 0.5 | 1 |
Recall | 0.11 | 0.11 | 0.11 | 0 | 0.11 | 0 | 0.11 | 0 |
F1 score | 0.2 | 0.18 | 0.2 | 0 | 0.2 | 0 | 0.18 | 0 |
Pre op 90 days Prediction | ||||||||
AUC | 0.69 | 0.63 | 0.72 | 0.66 | 0.61 | 0.56 | 0.63 | 0.56 |
Accuracy | 0.59 | 0.59 | 0.58 | 0.7 | 0.57 | 0.16 | 0.52 | 0.57 |
Precision | 0.24 | 0.23 | 0.25 | 0.27 | 0.21 | 0.16 | 0.21 | 0.19 |
Recall | 0.73 | 0.64 | 0.82 | 0.55 | 0.64 | 1 | 0.73 | 0.55 |
F1 score | 0.36 | 0.33 | 0.38 | 0.36 | 0.32 | 0.27 | 0.33 | 0.29 |
Post op 30 days Prediction | ||||||||
AUC | 0.62 | 0.65 | 0.66 | 0.63 | 0.71 | 0.55 | 0.61 | 0.6 |
Accuracy | 0.88 | 87 | 0.87 | 0.87 | 0.88 | 0.87 | 0.87 | 0.87 |
Precision | 1 | 1 | 0.5 | 1 | 1 | 0.5 | 1 | 0.5 |
Recall | 0.11 | 0 | 0.11 | 0 | 0.11 | 0.11 | 0 | 0.22 |
F1 score | 0.2 | 0 | 0.18 | 0 | 0.2 | 0.18 | 0 | 0.31 |
Post-op 90 days Prediction | ||||||||
AUC | 0.67 | 0.59 | 0.68 | 0.66 | 0.63 | 0.55 | 0.63 | 0.65 |
Accuracy | 0.86 | 0.84 | 0.86 | 0.86 | 0.84 | 0.86 | 0.86 | 0.84 |
Precision | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
Recall | 0 | 0 | 0.09 | 0.09 | 0 | 0.09 | 0.09 | 0.09 |
F1 score | 0.17 | 0 | 0.17 | 0.17 | 0 | 0.17 | 0.17 | 0 |
Abbreviations: AUROC = area under the receiver operating characteristic, SVM = Support Vector Machine, SGD = Stochastic Gradient Descent
Fig 2. Variable importance for the postoperative prediction model.
Rogers score8 and Caprini score9 are two commonly used VTE risk assessment models in surgical patients. However, neither of these tools is in any way specific to spinal metastasis surgery, neither includes operative time or blood chemistry investigations, which may influence postoperative VTE development.
Previously developed prediction models included fewer risk factors due to analytic limitations. However, with the use of machine learning modeling, we were able to include 14 important risk factors in the development of our prediction algorithms. To our knowledge, this is the first study in machine learning prediction algorithm development to include both preoperative and postoperative factors in the development of algorithms that predict preoperative and postoperative VTE in patients scheduled to undergo surgery for spinal metastasis.
This study’s findings are subject to several limitations, including the inability to differentiate between symptomatic and asymptomatic VTE, which may affect the clinical application of our predictive models. The external validity of our results is constrained by the use of data from a single institution, which may not be representative of broader patient populations. Furthermore, the potential for algorithmic overfitting and the inherent complexity of the machine learning models used could limit the interpretability and generalizability of our findings. Future research should focus on external validation with diverse cohorts, incorporate a broader array of predictive variables, and explore a range of performance metrics beyond AUC to ensure the clinical relevance of the predictive models.
CONCLUSION
Our study affirms the utility of machine learning models in predicting VTE for spinal metastasis surgery patients, with the gradient boosted trees algorithm showing strong predictive performance for the 30-day timeframe and the support vector machine algorithm standing out for the 90-day predictions. These models have demonstrated good to fair predictive capabilities, as indicated by AUC values aligning with established thresholds. Key factors like age, BMI, and preoperative status, consistent with prior studies, have been instrumental in model accuracy, supporting their integration into clinical risk assessments to potentially enhance patient care.
ACKNOWLEDGEMENTS
The authors gratefully acknowledge Miss Sarunya Poolpol in Research Unit, Department of orthopedics
Faculty of Medicine Siriraj Hospital, Mahidol University for assistance with statistical analysis, manuscript preparation and journal submission process.
This research project received support from the Siriraj Research Fund, Grant number (O)R016438001, Faculty of Medicine, Siriraj Hospital, Mahidol University.
All authors declare no personal or professional conflicts of interest relating to any aspect of this study.
The protocol for this study was approved by the Siriraj Institutional Review Board (SIRB) (COA no. Si 1065/2020), and written informed consent was not obtained due to the retrospective nature of this study.
REFERENCES
Walker AJ, Card TR, West J, Crooks C, Grainge MJ. Incidence of venous thromboembolism in patients with cancer–a cohort study using linked United Kingdom databases. Eur J Cancer. 2013; 49(6):1404-13.
Heit JA, Silverstein MD, Mohr DN Petterson TM, O’Fallon WM, Melton LJ 3rd. Risk factors for deep vein thrombosis and pulmonary embolism: a population-based case-control study. Arch Intern Med. 2000;160(6):809-15.
Gussoni G, Frasson S, La Regina M, Di Micco P, Monreal M, RIETE Investigators. Three-month mortality rate and clinical predictors in patients with venous thromboembolism and cancer. Findings from the RIETE registry.Thromb Res. 2013;131(1): 24-30.
Zhang L, Cao H, Chen Y, Jiao G. Risk factors for venous thromboembolism following spinal surgery: A meta-analysis. Medicine (Baltimore). 2020;99(29):e20954.
Zacharia BE, Kahn S, Bander ED, Cederquist GY, Cope WP, McLaughlin L, et al. Incidence and risk factors for preoperative deep venous thrombosis in 314 consecutive patients undergoing surgery for spinal metastasis. J Neurosurg Spine. 2017;27(2): 189-97.
Groot OQ, Ogink PT, Pereira NRP, Ferrone ML, Harris MB, Lozano-Calderon SA, et al. High risk of symptomatic venous thromboembolism after surgery for spine metastatic bone lesions: a retrospective study. Clin Orthop Relat Res. 2019;477(7):1674- 86.
Cohoon KP, Ransom JE, Leibson CL, Ashrani AA, Petterson TM, Long KH, et al. Direct medical costs attributable to cancer- associated venous thromboembolism: a population-based longitudinal study. Am J Med. 2016;129(9):1000.e15-25.
Rogers SO Jr, Kilaru RK, Hosokawa P, Henderson WG, Zinner MJ, Khuri SF. Multivariable predictors of postoperative venous thromboembolic events after general and vascular surgery: results from the patient safety in surgery study. J Am Coll Surg. 2007;204(6):1211-21.
Caprini JA, Arcelus JI, Reyna JJ. Effective risk stratification
of surgical and nonsurgical patients for venous thromboembolic disease. Semin Hematol. 2001;38(2 Suppl 5):12-9.
Barber EL, Clarke-Pearson DL. The limited utility of currently available venous thromboembolism risk assessment tools in gynecological oncology patients. Am J Obstet Gynecol. 2016; 215(4):445.e1-9.
Khorana AA, Kuderer NM, Culakova E, Lyman GH, Francis CW. Development and validation of a predictive model for chemotherapy- associated thrombosis. Blood. 2008;111(10):4902-7.
Han SS, Azad TD, Suarez PA, Ratliff JK. A machine learning approach for predictive models of adverse events following spine surgery. Spine J. 2019;19(11):1772-81.
Collins GS, Reitsma JB, Altman DG, Moons KG.Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) the TRIPOD statement. Circulation. 2015;131(2):211-9.
Luo W, Phung D, Tran T, Gupta S, Rana S, Karmakar C, et al. Guidelines for developing and reporting machine learning predictive models in biomedical research: a multidisciplinary view. J Med Internet Res. 2016;18(12):e323.
Ikeda T, Miyamoto H, Hashimoto K, Akagi M. Predictable factors of deep venous thrombosis in patients undergoing spine surgery. J Orthop Sci. 2017;22(2):197-200.
Yoshioka K, Murakami H, Demura S, Kato S, Tsuchiya H. Prevalence and risk factors for development of venous thromboembolism after degenerative spinal surgery. Spine (Phila Pa 1976). 2015; 40(5):E301-6.
Wei J, Li W, Pei Y, Shen Y, Li J. Clinical analysis of preoperative risk factors for the incidence of deep venous thromboembolism in patients undergoing posterior lumbar interbody fusion. J Orthop Surg Res. 2016;11(1):68.
Yang SD, Liu H, Sun YP, Yang DL, Shen Y, Feng SQ, et al. Prevalence and risk factors of deep vein thrombosis in patients after spine surgery: a retrospective case-cohort study. Sci Rep. 2015; 5:11834.
Tominaga H, Setoguchi T, Tanabe F, Kawamura I, Tsuneyoshi Y, Kawabata N, et al. Risk factors for venous thromboembolism after spine surgery. Medicine (Baltimore). 2015;94(5):e466.
Guzman JZ, Latridis JC, Skovrlj B, Cutler HS, Hecht AC, Qureshi SA, et al. Outcomes and complications of diabetes mellitus on patients undergoing degenerative lumbar spine surgery. Spine (Phila Pa 1976). 2014 Sep 1;39(19):1596-604.
Adogwa O, Elsamadicy AA, Sergesketter A, Oyeyemi D, Galan D, Vuong VD, et al. The impact of chronic kidney disease on postoperative outcomes in patients undergoing lumbar decompression and fusion. World Neurosurg. 2018;110:e266-e70.
De la Garza Ramos R, Jain A, Nakhla J, Nasser R, Puvanesarajah V, Hassanzadeh H, et al. Postoperative morbidity and mortality after elective anterior cervical fusion in patients with chronic and end-stage renal disease. World Neurosurg. 2016;95:480-5.
Sebastian AS, Currier BL, Kakar S, Nguyen EC, Wagie AE, Habermann ES, et al. Risk factors for venous thromboembolism following thoracolumbar surgery: analysis of 43,777 patients from the American College of Surgeons National Surgical Quality Improvement Program 2005 to 2012. Global Spine J. 2016;6(8):
738-43.
Phan K, Kothari P, Lee NJ, Virk S, Kim JS, Cho SK. Impact of obesity on outcomes in adults undergoing elective posterior cervical fusion. Spine (Phila Pa 1976). 2017;42(4):261-6.
Flippin M, Harris J, Paxton EW, Prentice HA, Fithian DC, Ward SR, et al. Effect of body mass index on patient outcomes of surgical intervention for the lumbar spine. J Spine Surg. 2017; 3(3):349-57.
Johnson DJ, Johnson CC, Cohen DB, Wetler JA, Kbaish KM, Frank SM. Thrombotic and infectious morbidity are associated with transfusion in posterior spine fusion. HSS J. 2017;13(2): 152-8.
Aoude A, Aldebeyan S, Fortin M, Nooh A, Jarzem P, Ouellet JA, et al. Prevalence and complications of postoperative transfusion for cervical fusion procedures in spine surgery: an analysis of 11,588 patients from the American College of Surgeons National Surgical Quality Improvement Program Database. Asian Spine J. 2017;11(6):880-91.
Aoude A, Nooh A, Fortin M, Aldebeyan S, Jarzem P, Ouellet J, et al. Incidence, predictors, and postoperative complications of blood transfusion in thoracic and lumbar fusion surgery: an analysis of 13,695 patients from the American College of Surgeons National Surgical Quality Improvement Program Database. Global Spine J. 2016;6(8):756-64.
Bekelis K, Desai A, Bakhoum SF, Missios S. A predictive model of complications after spine surgery: the National Surgical Quality Improvement Program (NSQIP) 2005–2010. Spine J. 2014; 14(7):1247-55.