Introduction

Acute kidney injury (AKI) is a widespread complication that occurs in 7.6–24% of hospitalized patients with traumatic brain injury (TBI).1–5 AKI can increase the risk of mortality and prolong the length of hospital stay among patients with TBI]1,5–7 The high prevalence of AKI in TBI may be caused by pathophysiological processes after the initial brain injury, including systemic inflammation, excessive sympathetic activation, hypoperfusion, and the common use of nephrotoxic medications such as hypertonic saline, blood product transfusion, and nephrotoxic antibiotics.8–10 The lack of drugs to alleviate and prevent renal injury emphasizes the importance of identifying patients with TBI who are at a high risk of AKI in the early stages. For TBI patients with a high likelihood of developing renal dysfunction, physicians should cautiously use or even withdraw nephrotoxic drugs and operations and try to maintain stable hemodynamics.

Several studies have identified various risk factors for AKI and developed predictive models for AKI in patients with TBI using conventional logistic or Cox regression.3,11–13 Compared to this conventional statistical method, machine learning has attracted much attention from researchers because of its advantages, such as handling complex datasets effectively and performing well in analyzing nonlinear data.14 Many studies have used various machine learning algorithms to predict outcomes, including the mortality and functional status of patients with TBI.15–19 However, few studies have explored the value of machine learning algorithms in predicting perioperative complications, especially AKI, in patients with TBI. As a classic machine learning algorithm, the decision tree algorithm has several advantages, including a fast calculation speed, convenient visualization, and applicability for high-dimensional data. It has been widely used to evaluate the prognosis and complications of other diseases such as pneumonia, cancer, stroke.20–23 Therefore, we designed this study to construct a predictive model for AKI in patients with TBI using the decision tree method and compared its predictive accuracy with that of a conventional logistic regression-based predictive model.

Materials and Methods Patients

Patients admitted to the emergency department of West China Hospital for TBI between January 2015 and June 2019 were eligible for this study. However, patients were excluded if they met the following criteria: (1) transferred from other hospitals after brain injury, (2) admitted to our hospital 6 h after brain injury, (3) lack of records of included variables, and (4) AKI occurred during the first day after admission. Finally, 376 patients were included in this study. This study was approved by the ethics committee of West China Hospital and conducted in accordance with the Declaration of Helsinki. Informed consent forms for joining the observational study were regularly signed by the patients themselves or their legally authorized representatives once the patients were admitted to our hospital.

Data Collection

Demographic information, injury mechanisms, and vital signs on admission, including systolic blood pressure, diastolic blood pressure, heart rate, body temperature, and respiratory rate, were recorded. Trauma scores including GCS; Abbreviated Injury Score (AIS) of the head, chest, abdomen, and limbs; and Injury Severity Score (ISS) were collected to reflect the injury severity of the brain and other body regions. Laboratory tests included white blood cells, neutrophils, lymphocytes, monocytes, platelets, hemoglobin, albumin, prothrombin time, red cell distribution width, glucose, lactate dehydrogenase, alkaline phosphatase, blood urea nitrogen, serum creatinine, serum uric acid, serum cystatin C, chloride, potassium, phosphorus, total cholesterol, triglyceride, high-density lipoprotein, and low-density lipoprotein. Laboratory tests were performed by analyzing the first blood sample collected on admission. Radiological signs, including epidural hematoma, subdural hematoma, subarachnoid hemorrhage, intraventricular hemorrhage, and diffuse axonal injury, and surgical options, including decompressive craniectomy and hematoma evacuation, were selected as variables. Additionally, medications during the first 24 h after admission, including platelet transfusion, fresh frozen plasma transfusion, cryoprecipitation transfusion, and furosemide use, were collected. The outcome of this study was the occurrence of AKI since the second day after admission. AKI was confirmed using the Kidney Disease Improving KDIGO serum creatinine criteria.

Statistical Analysis

Categorical and continuous variables were presented as numbers (percentage) and mean ± standard deviation (normal distribution) or median (interquartile range) (non-normal distribution), respectively. The normality of the variables was verified using the Kolmogorov–Smirnov test. Differences in continuous variables between the two groups were compared using the Independent Student’s t-test (normal distribution) and the Mann–Whitney U-test (non-normal distribution). The χ2 test or Fisher’s exact test was used to compare the differences in categorical variables.

Minimizing the collinearity of the included variables and avoiding overfitting of these variables, least absolute shrinkage and selection operator (LASSO) regression was used to identify predictors of AKI with nonzero coefficients. Identifying the strongest predictors from a number of variables for outcomes with a small sample size is an advantage of LASSO regression, which fits the characteristics of our dataset, with 39 AKI occurrences and 49 independent variables. Predictors with non-zero coefficients were then combined to construct a predictive model for AKI in patients with TBI using multivariate logistic regression.

A decision tree predictive model for AKI in TBI was constructed using the classification and regression tree (CART) algorithm. The decision tree was set with a maximum depth of 10 layers, minimum of 40 cases for each parent node, and minimum of 20 cases for each child node. Branches were optimally split based on the Gini impurity index and pre-pruning was performed to avoid overfitting the CART model. A 10-fold cross-validation method was used to internally validate the CART model. This method randomly divides the original dataset into ten subsets of equal sizes, with nine subsets as the training set and the other as the validation set. This procedure was repeated 10 times, with each of the 10 subsets used once as the validation set. The optimal CART model was selected on the basis of its predictive accuracy.

Receiver operating characteristic (ROC) curves of the LASSO regression predictive model and the CART model were drawn, and the area under the ROC curves (AUC) was calculated to compare their predictive accuracy (the Z test was used to compare the AUC).

Statistical p value was defined as a two-sided p-value of < 0.05. SPSS software (version 22.0; IBM Corp., Armonk, NY, USA) and R software (version 3.6.1; R Foundation) (packages including glmnet, caret, rms, rpart, and rpart.plot) were used for all statistical analyses and figure drawings.

Results Baseline Characteristics of Included TBI Patients

A total of 376 patients had an AKI incidence of 10.4% (Table 1). The AKI group was older (p = 0.013) and had a lower GCS score (p < 0.001) than the non-AKI group. The AIS head (p = 0.001), AIS thoracic (p = 0.026), and ISS (p = 0.002) scores were higher in the AKI group. Laboratory tests showed that the AKI group had lower levels of lymphocytes (p = 0.001), hemoglobin (p = 0.012), albumin (p = 0.003), and total cholesterol (p = 0.005), and higher levels of prothrombin time (p = 0.003), red cell distribution width (p = 0.002), glucose (p <0.001), blood urea nitrogen (p = 0.005), serum creatinine (p <0.001), serum uric acid (p <0.001), cystatin C (p <0.001), and chloride (p <0.001). Records of medical drugs showed that transfusion rates of platelets (p = 0.002) and fresh frozen plasma (p <0.001) were higher in the AKI group. The length of ICU stay (p = 0.533) did not differ between the AKI and non-AKI groups, whereas the total length of hospital stay was shorter in the AKI group (p = 0.006). The overall patients mortality was 44.7%. The AKI group had a significantly higher in-hospital mortality rate than the non-AKI group (p <0.001).

Table 1 Baseline Characteristics of TBI Patients Grouped by AKI

Value of CART and Lasso Regression Models for Predicting AKI in TBI Patients

Lasso regression revealed five potent predictive factors for AKI: glucose, serum creatinine, cystatin C, serum uric acid, and fresh frozen plasma transfusions (Figure 1A and B). A logistic regression-based predictive model was constructed using these five factors. A decision tree model for predicting AKI was constructed using CART analysis, as shown in Figure 2. At the first node from the root, The SUA <314 was the most significant categorical discriminator to identify AKI risk in TBI patients. Cystatin C <1.00 and glucose <15.00 were at the second node. 40% of TBI patients who were SUA <314 and glucose <15.00, had an AKI risk of 22.0%. The feature importance of each variable analyzed using CART is shown in Figure 3. The AUC values of the decision tree and logistic regression models were 0.892 and 0.854, respectively (Table 2) (Figure 4). The AUC value of the decision tree model was higher than that of the logistic regression model, although the difference was not statistically significant (Z = 1.0209, p >0.05). The decision tree model had a higher specificity (0.821), whereas the logistic regression model had a higher sensitivity (1.000) for predicting AKI in patients with TBI.

Table 2 Comparison of Predictive Accuracy Between CART Model and Lasso Based Predictive Model

Figure 1 (A) The predictive factors selection using LASSO binary logistic regression. Two dotted vertical lines mark the optimal values by minimum criteria and 1-s.e. criteria. Five variables were selected by LASSO binary logistic regression including glucose, serum creatinine, cystatin C, serum uric acid and fresh frozen plasma transfusion. (B) LASSO coefficient profiles of 49 variables.

Abbreviation: LASSO, least absolute shrinkage and selection operator.

Figure 2 Decision tree model for the prediction of AKI in included TBI patients using CART analysis.

Abbreviations: Cl, serum chloride; SUA, serum uric acid; AKI, acute kidney injury; TBI, traumatic brain injury; CART, Classification And Regression Tree.

Figure 3 Feature importance of variables recognized by CART analysis.

Abbreviations: CART, Classification And Regression Tree; SUA, serum uric acid; Scr, serum creatinine; Cl, serum chloride; PT, prothrombin time; GCS, Glasgow Coma Scale; ISS, Injury Severity Score; RR, respiratory rate; WBC, white blood cell; BUN, blood urea nitrogen; RDW, red cell distribution width; SAH, subarachnoid hemorrhage; LDL, low density cholesterol; SBP, systolic blood pressure; DBP, diastolic blood pressure; HDL, high density cholesterol.

Figure 4 Receiver operating characteristic curve of CART and the LASSO logistic regression models for predicting AKI in included TBI patients. The AUC of decision tree model and Lasso regression model were 0.892 (0.838–0.947) and 0.854 (0.806–0.903), respectively.

Abbreviations: AUC, area under the receiver operating characteristic curve; CART, Classification And Regression Tree; LASSO, least absolute shrinkage and selection operator; AKI, acute kidney injury; TBI, traumatic brain injury.

Discussion

The AKI incidence in TBI patients included in this study was 10.4%, which was similar to the 7.6% to 24% reported in previous studies.1–5 The relatively low incidence of AKI in this study is attributable to the exclusion of patients who developed AKI on the first day after admission. These patients usually show a rapid increase in serum creatinine level after admission and were therefore excluded due to obvious signs of renal dysfunction without the need for prediction, and immediate AKI after admission was difficult to avoid in the short-term. The CART algorithm in our study discovered four variables to split the three branches: glucose, serum uric acid, serum cystatin, and serum chloride levels.

Glucose level was used to split branches with higher glucose levels, indicating a higher likelihood of AKI in our constructed decision tree. Many previous studies have confirmed that hyperglycemia is a risk factor for AKI in various patients, such as those with acute coronary syndrome, myocardial infarction, and sepsis.24–28 Stress-induced hyperglycemia is prevalent in patients with TBI, with reported incidences ranging from 7.8% to 29.4%.29–33 Acute hyperglycemia after trauma causes loss of the glycocalyx layer, endothelial cell inflammation, and coagulation activation, which may also aggravate renal injury.34

Two other indices indicating renal dysfunction, serum uric acid and serum cystatin C levels, were also incorporated into the decision tree. As the end product of purine metabolism in the body, uric acid is mainly excreted from the kidneys. Renal dysfunction with a reduced glomerular filtration rate (GFR) may lead to uric acid accumulation and increased blood concentration. Cystatin C is also filtered through the glomerulus and completely reabsorbed by the proximal renal tubule. Increased serum cystatin C level may indicate impaired renal function. Previous studies have verified the predictive value of serum uric acid and cystatin C in patients with AKI, such as those with cirrhosis, myocardial infarction, and those treated with cardiac surgery or cystectomy.35–43

Finally, serum chloride level was included in the constructed decision tree to split the parent node. As shown in Figure 2, patients with chloride ≥115.0 mmol/L had a higher incidence of AKI than those with chloride <115.0 mmol/L. Hyperchloremia has been confirmed to be independently associated with AKI in some patients, including subarachnoid hemorrhage, sepsis, intracerebral hemorrhage, and brain tumor resection.44–49 Furthermore, one study found that prolonged hyperchloremia duration, but not hyperchloremia occurrence, was independently correlated with AKI in patients.13 A potential mechanism underlying this association has been proposed in previous studies, which revealed that excessive chloride can induce renal vasoconstriction and reduce renal cortical perfusion.50,51

Many studies have explored the value of machine learning algorithms in predicting outcomes in patients.15–17,19,52,53 However, few studies have explored the effectiveness of machine-learning algorithms in predicting subsequent complications in hospitalized patients with TBI. TBI is a complex disease that involves pathophysiological processes in systemic organs. Non-neurological complications are prevalent in patients with TBI and are associated with poor outcomes. Evaluating the risk of complications after admission may be helpful for physicians and surgeons to develop personalized medical strategies to improve the prognosis of patients with TBI.

In this study, we developed a decision tree to evaluate the risk of AKI in hospitalized TBI patients using the CART algorithm. Decision trees are popular in medical decision-making because of their advantages, including fast calculation speed, high accuracy, collective processing of continuous and categorical variables, applicability to high-dimensional data unaffected by data scaling, and convenient visualization as a form of flowchart. The CART is a learning method that outputs the conditional probability distribution of the random variable Y under the given input random variable X conditions. The advantages of CART include strong interpretability, good stability, excellent visualization effect, high computational efficiency. The CART can present results in an intuitive and easily understandable way, with each node representing a feature and each branch representing a decision, thus clearly explaining the relationships between data and the decision-making process in a graphical way. Also, the CART has relatively low computational complexity and can train models in a short period of time. The CART has wide applications in sociology, economics, and medicine. In medicine, CART can be used to diagnose diseases, predict disease progression and prognosis, and develop treatment plans. One study developed a CART model to assist clinical prediction for tracheostomy in patients with traumatic cervical spinal cord injury using three simple indexes including American Spinal Injury Association classification, neurological level of impairment, Injury Severity Score.54 Another study constructed a classification and regression tree which could identify subgroups of childhood type 1 diabetes using three parameters including age, hemoglobin A1c and body mass index.55

The predictive value of the decision-tree model in our study was not inferior to that of the Lasso regression based predictive model. The difference in AUC between these two models may be statistically significant in future validation cohorts with larger sample sizes. Regardless, compared with the LASSO regression model requiring a particular calculation of AKI probability, the decision tree model is more comprehensive and easier to evaluate the risk of AKI. This could prompt physicians to reduce the use of nephrotoxic medications to prevent the occurrence or progression of AKI in clinical practice.

This study had several limitations. First, the decision tree model was developed using patients from a single medical center and internally validated using 10-fold cross validation. Selection bias is inevitable, and the stability and generalizability of this model should be externally verified in other medical centers with larger sample sizes. Second, although many factors were selected for this study, there are still some potential factors, such as nephrotoxic antibiotics, that have not been included in the analysis. Third, the dynamic fluctuation of filtration markers, including cystatin C and serum uric acid, was not recorded, so we could not analyze the value of this fluctuation, which may better reflect deteriorating renal function.

Conclusion

The decision tree predictive model, composed of glucose, cystatin C, serum uric acid, and chloride, is valuable for predicting AKI among patients with TBI. This tree-based flowchart is visual and convenient for physicians to identify patients with TBI with a high risk of AKI and consequently prompts physicians to develop suitable therapeutic strategies.

Data Sharing Statement

The datasets are available from the corresponding author upon reasonable request.

Ethical Approval and Informed Consent

This study was approved by the ethics committee of West China Hospital and conducted in accordance with the Declaration of Helsinki. Informed consent forms for joining the observational study were regularly signed by the patients themselves or their legally authorized representatives once the patients were admitted to our hospital.

Author Contributions

All authors made a significant contribution to the work reported, whether in the conception, study design, execution, acquisition of data, analysis, and interpretation, or in all these areas, took part in drafting, revising, or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Funding

This study was funded by the 1·3·5 project for disciplines of excellence: Clinical Research Incubation Project, West China Hospital, Sichuan University (2020HXFH036); Knowledge Innovation Program of the Chinese Academy of Sciences (JH2022007); National Natural Science Foundation of China (82173175); Sichuan Science and Technology Program (2021YFS0082); and Post-Doctor Research Project, Sichuan University (2021SCU12027).

Disclosure

Ruoran Wang and Jing Zhang are co-first authors for this study. The authors declare that they have no conflicts of interest for this work.

References

1. Corral L, Javierre CF, Ventura JL, Marcos P, Herrero JI, Manez R. Impact of non-neurological complications in severe traumatic brain injury outcome. Critical Care. 2012;16(2):R44. doi:10.1186/cc11243

2. Ahmed M, Sriganesh K, Vinay B, Umamaheswara Rao GS. Umamaheswara Rao, Acute kidney injury in survivors of surgery for severe traumatic brain injury: incidence, risk factors, and outcome from a tertiary neuroscience center in India. Br J Neurosurg. 2015;29(4):544–548. doi:10.3109/02688697.2015.1016892

3. Robba C, Banzato E, Rebora P, et al. Acute kidney injury in traumatic brain injury patients: results from the collaborative European neurotrauma effectiveness research in traumatic brain injury study. Crit Care Med. 2021;49(1):112–126. doi:10.1097/CCM.0000000000004673

4. Li N, Zhao WG, Xu FL, Zhang WF, Gu WT. Neutrophil gelatinase-associated lipocalin as an early marker of acute kidney injury in patients with traumatic brain injury. J Nephrol. 2013;26(6):1083–1088. doi:10.5301/jn.5000282

5. Moore EM, Bellomo R, Nichol A, Harley N, Macisaac C, Cooper DJ. The incidence of acute kidney injury in patients with traumatic brain injury. Renal Failure. 2010;32(9):1060–1065. doi:10.3109/0886022X.2010.510234

6. Li N, Zhao WG, Zhang WF. Acute kidney injury in patients with severe traumatic brain injury: implementation of the acute kidney injury network stage system. Neurocritical Care. 2011;14(3):377–381. doi:10.1007/s12028-011-9511-1

7. Skrifvars MB, Moore E, Martensson J, et al. Erythropoietin in traumatic brain injury associated acute kidney injury: a randomized controlled trial. Acta Anaesthesiol Scandinavica. 2019;63(2):200–207. doi:10.1111/aas.13244

8. Büttner S, Stadler A, Mayer C, et al. Incidence, risk factors, and outcome of acute kidney injury in neurocritical care. J Int Care Med. 2020;35(4):338–346. doi:10.1177/0885066617748596

9. An S, Luo H, Wang J, et al. An acute kidney injury prediction nomogram based on neurosurgical intensive care unit profiles. Ann Translat Med. 2020;8(5):194. doi:10.21037/atm.2020.01.60

10. Deng Y, Yuan J, Chi R, et al. The incidence, risk factors and outcomes of postoperative acute kidney injury in neurosurgical critically ill patients. Sci Rep. 2017;7(1):4245. doi:10.1038/s41598-017-04627-3

11. Wang RR, He M, Gui X, Kang Y. A nomogram based on serum cystatin C for predicting acute kidney injury in patients with traumatic brain injury. Renal Failure. 2021;43(1):206–215. doi:10.1080/0886022X.2021.1871919

12. Wang RR, He M, Ou XF, Xie XQ, Kang Y. The predictive value of serum uric acid on acute kidney injury following traumatic brain injury. Biomed Res Int. 2020;2020:2874369. doi:10.1155/2020/2874369

13. Yamane DP, Maghami S, Graham A, Vaziri K, Davison D. Association of hyperchloremia and acute kidney injury in patients with traumatic brain injury. J Int Care Med. 2020;3:885066620978735.

14. Liu NT, Salinas J. Machine learning for predicting outcomes in trauma. Shock. 2017;48(5):504–510. doi:10.1097/SHK.0000000000000898

15. Hanko M, Snopko P, Opšenák R, et al. Random forest-based prediction of outcome and mortality in patients with traumatic brain injury undergoing primary decompressive craniectomy. World Neurosurg. 2021;148:e450–e458. doi:10.1016/j.wneu.2021.01.002

16. Matsuo K, Aihara H, Nakai T, Morishita A, Tohma Y, Kohmura E. Machine Learning to Predict In-Hospital Morbidity and Mortality after Traumatic Brain Injury. J Neurotrauma. 2020;37(1):202–210. doi:10.1089/neu.2018.6276

17. Abujaber A, Gammoh A, Abdelrahman H, Mollazehi M, El-Menyar A. Prediction of in-hospital mortality in patients with post traumatic brain injury using National trauma registry and machine learning approach. Scandina J Trauma Resuscit Emer Med. 2020;28(1):44. doi:10.1186/s13049-020-00738-5

18. Abujaber A, Gammoh D, Abdelrahman H, Mollazehi M, El-Menyar A. Prediction of in-hospital mortality in patients on mechanical ventilation post traumatic brain injury: machine learning approach. BMC Med Inf Decis Making. 2020;20(1):336. doi:10.1186/s12911-020-01363-z

19. Rau CS, Kuo PJ, Chien PC, Huang CY, Hsieh HY. Mortality prediction in patients with isolated moderate and severe traumatic brain injury using machine learning models. PLoS One. 2018;13(11):e0207192. doi:10.1371/journal.pone.0207192

20. Hostettler IC, Muroi C, Richter JK, et al. Decision tree analysis in subarachnoid hemorrhage: prediction of outcome parameters during the course of aneurysmal subarachnoid hemorrhage using decision tree analysis. J Neurosurg. 2018;129(6):1499–1510. doi:10.3171/2017.7.JNS17677

21. Phan TG, Chen J, Singhal S, et al. Exploratory use of decision tree analysis in classification of outcome in hypoxic-ischemic brain injury. Front Neurol. 2018;9:126. doi:10.3389/fneur.2018.00126

22. Yang Q, Li J, Zhang Z, et al. Clinical characteristics and a decision tree model to predict death outcome in severe COVID-19 patients. BMC Infect Dis. 2021;21(1):783. doi:10.1186/s12879-021-06478-w

23. Brims FJ, Meniawy TM, Duffus I, et al. A novel clinical prediction model for prognosis in malignant pleural mesothelioma using decision tree analysis. J Thorac Oncol. 2016;11(4):573–582. doi:10.1016/j.jtho.2015.12.108

24. Lin KY, Shang XL, Guo YS, et al. Association of preprocedural hyperglycemia with contrast-induced acute kidney injury and poor outcomes after emergency percutaneous coronary intervention. Angiology. 2018;69(9):770–778. doi:10.1177/0003319718758140

25. Moriyama N, Ishihara M, Noguchi T, et al. Admission hyperglycemia is an independent predictor of acute kidney injury in patients with acute myocardial infarction. Circ J. 2014;78(6):1475–1480. doi:10.1253/circj.CJ-14-0117

26. Shacham Y, Gal-Oz A, Leshem-Rubinow E, et al. Admission glucose levels and the risk of acute kidney injury in nondiabetic ST segment elevation myocardial infarction patients undergoing primary percutaneous coronary intervention. Cardio Med. 2015;5(3):191–198. doi:10.1159/000430472

27. Naruse H, Ishii J, Hashimoto T, et al. Pre-procedural glucose levels and the risk for contrast-induced acute kidney injury in patients undergoing emergency coronary intervention. Circ J. 2012;76(8):1848–1855. doi:10.1253/circj.CJ-11-1248

28. Yang S, Su T, Huang L, et al. A novel risk-predicted nomogram for sepsis associated-acute kidney injury among critically ill patients. BMC Nephrol. 2021;22(1):173. doi:10.1186/s12882-021-02379-x

29. Rau CS, Wu SC, Chen YC, et al. Higher mortality in trauma patients is associated with stress-induced hyperglycemia, but not diabetic hyperglycemia: a cross-sectional analysis based on a propensity-score matching approach. Int J Environ Res Public Health. 2017;14:10.

30. Matovu P, Kirya M, Kiryabwire J, et al. Hyperglycemia in severe traumatic brain injury patients and its association with thirty-day mortality: a prospective observational cohort study in Uganda. PeerJ. 2021;9:e10589. doi:10.7717/peerj.10589

31. Tsai YC, Wu SC, Hsieh TM, et al. Association of stress-induced hyperglycemia and diabetic hyperglycemia with mortality in patients with traumatic brain injury: analysis of a propensity score-matched population. Int J Environ Res Public Health. 2020;17(12):4266. doi:10.3390/ijerph17124266

32. Rau CS, Wu SC, Chen YC, et al. Stress-induced hyperglycemia, but not diabetic hyperglycemia, is associated with higher mortality in patients with isolated moderate and severe traumatic brain injury: analysis of a propensity score-matched population. Int J Environ Res Public Health. 2017;14:11.

33. Bosarge PL, Shoultz TH, Griffin RL, Kerby JD. Stress-induced hyperglycemia is associated with higher mortality in severe traumatic brain injury. J Trauma Acute Care Surg. 2015;79(2):289–294. doi:10.1097/TA.0000000000000716

34. Diebel LN, Martin JV, Liberati DM, Liberati DM. Acute hyperglycemia exacerbates trauma-induced endothelial and glycocalyx injury: an in vitro model. J Trauma Acute Care Surg. 2018;85(5):960–967. doi:10.1097/TA.0000000000001993

35. Joung KW, Choi SS, Kong YG, Yu J, Lim J, Kim YK. Incidence and risk factors of acute kidney injury after radical cystectomy: importance of preoperative serum uric acid level. Int J Med Sci. 2015;12(7):599–604. doi:10.7150/ijms.12106

36. Ejaz AA, Beaver TM, Shimada M, et al. Uric acid: a novel risk factor for acute kidney injury in high-risk cardiac surgery patients? Am j Nephrol. 2009;30(5):425–429. doi:10.1159/000238824

37. Wang X, Xie B, Huang R, et al. Early serum cystatin C-enhanced risk prediction for acute kidney injury post cardiac surgery: a prospective, observational, cohort study. Biomarkers. 2020;25(1):20–26. doi:10.1080/1354750X.2019.1688865

38. El-Sadek AE, El-Gamasy MA, Behiry EG, Torky AA, Fathy MA. Plasma cystatin C versus renal resistive index as early predictors of acute kidney injury in critically ill neonates. J Pediatr Urol. 2020;16(2):206.e1–206.e8. doi:10.1016/j.jpurol.2019.12.001

39. Maiwall R, Kumar A, Kumar G, Bhadoria AS, Sarin SK, Sarin SK. Cystatin C predicts acute kidney injury and mortality in cirrhotics: a prospective cohort study. Liver Int. 2018;38(4):654–664. doi:10.1111/liv.13600

40. Zappitelli M, Krawczeski CD, Devarajan P, et al. Early postoperative serum cystatin C predicts severe acute kidney injury following pediatric cardiac surgery. Kidney Int. 2011;80(6):655–662. doi:10.1038/ki.2011.123

41. Krawczeski CD, Vandevoorde RG, Kathman T, et al. Serum cystatin C is an early predictive biomarker of acute kidney injury after pediatric cardiopulmonary bypass. Clin J Am Soc Nephrol. 2010;5(9):1552–1557. doi:10.2215/CJN.02040310

42. Kaufeld T, Foerster KA, Schilling T, et al. Preoperative serum uric acid predicts incident acute kidney injury following cardiac surgery. BMC Nephrol. 2018;19(1):161. doi:10.1186/s12882-018-0970-x

43. Mandurino-Mirizzi A, Kajana V, Cornara S, et al. Elevated serum uric acid is a predictor of contrast associated acute kidney injury in patient with ST-segment elevation myocardial infarction undergoing primary percutaneous coronary intervention, Nutrition, metabolism, and cardiovascular diseases. NMCD. 2021;31(7):2140–2143. doi:10.1016/j.numecd.2021.04.002

44. Suetrong B, Pisitsak C, Boyd JH, Russell JA, Walley KR. Hyperchloremia and moderate increase in serum chloride are associated with acute kidney injury in severe sepsis and septic shock patients. Critical Care. 2016;20(1):315. doi:10.1186/s13054-016-1499-7

45. Sadan O, Singbartl K, Kandiah PA, Martin KS, Samuels OB. Hyperchloremia is associated with acute kidney injury in patients with subarachnoid hemorrhage. Crit Care Med. 2017;45(8):1382–1388. doi:10.1097/CCM.0000000000002497

46. Oh TK, Jeon YT, Sohn H, et al. Association of perioperative hyperchloremia and hyperchloremic metabolic acidosis with acute kidney injury after craniotomy for intracranial hemorrhage. World Neurosurg. 2019;125:e1226–e1240. doi:10.1016/j.wneu.2019.02.015

47. Oh TK, Kim CY, Jeon YT, Hwang JW, Do SH. Perioperative hyperchloremia and its association with postoperative acute kidney injury after craniotomy for primary brain tumor resection: a retrospective observational study. J Neurosurg Anesthesiol. 2019;31(3):311–317. doi:10.1097/ANA.0000000000000512

48. Lombardi G, Ferraro PM, Bargagli M, Naticchia A, D’Alonzo S, Gambaro G. Hyperchloremia and acute kidney injury: a retrospective observational cohort study on a general mixed medical-surgical not ICU-hospitalized population. Int Emerg Med. 2020;15(2):273–280. doi:10.1007/s11739-019-02165-6

49. Stenson EK, Cvijanovich NZ, Allen GL, et al. Hyperchloremia is associated with acute kidney injury in pediatric patients with septic shock. Intensive Care Med. 2018;44(11):2004–2005. doi:10.1007/s00134-018-5368-5

50. Wilcox CS. Regulation of renal blood flow by plasma chloride. J Clin Invest. 1983;71(3):726–735. doi:10.1172/JCI110820

51. Chowdhury AH, Cox EF, Francis ST, Lobo DN. A randomized, controlled, double-blind crossover study on the effects of 2-L infusions of 0.9% saline and plasma-lyte® 148 on renal blood flow velocity and renal cortical tissue perfusion in healthy volunteers. Ann Surg. 2012;256(1):18–24. doi:10.1097/SLA.0b013e318256be72

52. Raj R, Luostarinen T, Pursiainen E, et al. Machine learning-based dynamic mortality prediction after traumatic brain injury. Sci Rep. 2019;9(1):17672. doi:10.1038/s41598-019-53889-6

53. Hsu SD, Chao E, Chen SJ, Hueng DY, Lan HY, Chiang HH. Machine learning algorithms to predict in-hospital mortality in patients with traumatic brain injury. J Personal Med. 2021;11(11):1144. doi:10.3390/jpm11111144

54. Sun D, Zhao H, Zhang Z. Classification and regression tree (CART) model to assist clinical prediction for tracheostomy in patients with traumatic cervical spinal cord injury: a 7-year study of 340 patients. Eur Spine J. 2022;31(5):1283–1290. doi:10.1007/s00586-022-07154-6

55. Achenbach P, Hippich M, Zapardiel-Gonzalo J, et al. A classification and regression tree analysis identifies subgroups of childhood type 1 diabetes. EBioMedicine. 2022;82:104118. doi:10.1016/j.ebiom.2022.104118