A Machine Learning Model Based on CT Imaging Metrics and Clinical Features to Predict the Risk of Hospital-Acquired Pneumonia After Traumatic Brain Injury

Introduction

With the development of society and the rise of road traffic, the incidence of accidental injuries, especially traumatic brain injury (TBI), caused by frequent traffic accidents has been increasing year by year, thus substantially increasing the risk of death.1 According to the US authorities, 64 to 74 million people worldwide suffer from TBI every year, of which about 55.9 million people are affected by mild TBI, while severe TBI affects about 5.48 million people.2 For every 100,000 people, there are 262 cases of death or disability due to TBI-related causes. Especially in patients with severe TBI, further serious complications such as cerebral contusions, axonal injury, and delayed hemorrhage may develop, all of which can severely affect the cognitive and motor functions of patients.3 It has been demonstrated that there is a complex and fine-grained interaction between the brain and other body systems, such as the central nervous system and the respiratory system, in a manner known as the “lung-brain axis”.4 On the one hand, the brain sends signals through the autonomic nervous system to directly regulate lung function, and the state of gas exchange in the lungs is markedly affected when in pathological conditions,5 on the other hand, dysregulation of the lung microbiota significantly affects the CNS immune response through the blood-brain barrier.6 One of the primary causes of death for TBI patients is severe pneumonia, which not only has a detrimental impact on the patients’ prognosis but also raises the financial burden on families.7 Therefore, therapeutic strategies for TBI should not only focus on cephalic conditions but also include interventions for complications such as pneumonia.

Earlier, hospital-acquired pneumonia (HAP) was a broad term covering ventilator-associated pneumonia (VAP) occurring after the establishment of an artificial airway and initiation of mechanical ventilation.8 However, the recently updated HAP/VAP guidelines in the United States differentiate between HAP and VAP, pointing out that they differ significantly in terms of pathogenic microorganisms, treatment experience, and preventive strategies, and treating them as two separate groups.9 Therefore, this study focused on patients with pneumonia unrelated to intubation and mechanical ventilation. For patients at high risk of HAP, HAP/VAP guidelines recommend immediate initiation of empirical antibiotic therapy.10 However, the use of prophylactic antibiotics does not appear to be effective in preventing pneumonia in patients who have suffered TBI and stroke.11–13 Nonetheless, certain studies have noted that prophylactic antibiotic use in specific groups can prevent pneumonia.14,15 This suggests that preventive and therapeutic approaches for TBI-associated pneumonia require individualized strategies. Therefore, the development of early prediction models for TBI-associated pneumonia is particularly important to help healthcare professionals identify high-risk patients in time to implement the necessary preventive measures, rapid diagnosis, and precise treatment.

Materials and Methods Data Source and Extraction

This was a retrospective study. The study protocol was approved by the Ethics Committee of the Second Affiliated Hospital of Fujian Medical University (2023–287). Since the study was retrospective in nature, informed permission was not required. Clinical data were collected electronically through the hospital’s electronic medical record system from 1 January 2018 to 1 June 2023 at the Second Affiliated Hospital of Fujian Medical University. The Ethics Committee was in charge of ensuring that the patients’ private information was properly protected. The inclusion criteria for patients with TBI in this study were as follows: (1) patients with definite TBI by cranial CT; (2) patients aged 18 years and above. The exclusion criteria were as follows: (1) Incomplete or missing key clinical data (CT was not perfected within 48h of admission); (2) Patients with tracheotomy and use of ventilator; (3) Patients discharged from the hospital within 72h (4) Patients with multiple injuries of the whole body or accompanied by serious life-threatening organ or system diseases. (5) Patients diagnosed with pneumonia before admission. After admission, each patient was examined and treated by a professional neurologist according to the TBI treatment guidelines.

Definition of HAP

HAP was defined as pneumonia occurring in the hospital 48 hours after admission.9 HAP was diagnosed by the presence of typical chest X-ray or computed tomography (CT) support, plus two or more of the following three clinical signs: (1) fever with a temperature of >38°C; (2) new onset of purulent sputum, change in sputum nature, increase in respiratory secretions, or increase in the need for suctioning; and (3) a peripheral blood leukocyte count of >10 × 10 9/L or < 4 × 109/L.16 In this study, HAP was used as an outcome variable.

Covariates

Data on the patient’s general status upon admission, neuroimaging features, and laboratory tests were gathered from the hospital’s electronic medical record system at the time of initial admission. The general conditions on admission included the patient’s sex, age, temperature on admission, heart rate, respiration, systolic blood pressure, diastolic blood pressure, cause of injury, time of onset, pupillary reflex, GCS, history of hypertension, diabetes mellitus, and coronary artery disease. Laboratory investigations on admission included leukocytes, centrocytes, lymphocytes, monocytes, hemoglobin, platelets, blood glucose, uric acid, creatinine, urea nitrogen, potassium, sodium, calcium, osmolality, APTT, PT, INR, Fib, TT, and D-dimer, with the addition of novel combined inflammatory markers SII, SIRI, PLR, NLR, and LMR.In addition, a record of the Neuroimaging features of TBI, including frontal lobe injury, parietal lobe injury, temporal lobe injury, occipital lobe injury, brainstem injury, cerebellar injury, epidural hemorrhage, subdural hemorrhage, craniovertebral fracture, skull base fracture, type of brain herniation, and Rotterdam scores (including cerebral pool compression or disappearance, subarachnoid hemorrhage, ventricular hemorrhage, midline shift of >5 mm, and extradural space-occupying lesions).

Statistical Analysis

In this study, the data collected from TBI patients were analyzed using R version 4.3.1, Python 3.12.0 and SPSS version 25.0 for subsequent statistical analysis. For continuous variables, data were expressed as mean ± standard deviation (X ± SD). For categorical variables, frequencies or percentages (%) were used. Different data metrics were compared between the two groups (training and validation datasets). An independent samples t-test was utilized to compare groups for continuously distributed variables that were normally distributed, whereas non-parametric tests were employed for variables that were not normally distributed. For comparing categorical variables between groups, the chi-square test was employed. The significance level was set at α=0.05, and P<0.05 was considered statistically significant. The LASSO was used to select significant features associated with pneumonia. Finally, seven features (glucose, type of brain hernia, age, GCS, Rotterdam, SIRI, D-dimer) were included in the established model. To obtain the highest predictive performance, five models were built, including a LR, a RF, a XGB classifier, a SVM classifier, and a NB classifier. The optimal model was evaluated based on the key performance metrics (Accuracy, Matthews correlation coefficient (MCC), recall, and F1 score) calculated from the confusion matrix. To further explain the optimal model, this study used multifactorial logistic regression to identify independent risk factors for HAP. In this model, the variance inflation factor (VIF) was used to estimate multicollinearity. A clinical prediction model was constructed based on the relative weights of the risk factors, and a dynamic nomogram was plotted. The discriminative power of the model was assessed using subject work characteristics (ROC) curves and AUC was calculated. Furthermore, calibration curves were constructed and the clinical prediction model was internally validated using the Bootstrap resampling approach with a repetition number of B=1000. Lastly, the Decision Clinical Curve (DCA) analysis was used to evaluate the clinical validity of the predictive models.

Results Baseline Characteristics

This study comprised 998 participants who satisfied the primary diagnostic. After reviewing the medical records, we excluded 135 patients who were younger than 18 years of age and 5 patients with missing data. Finally, 858 eligible participants were included in this study for analysis. The flowchart of patient recruitment is shown below (Figure 1). The 858 included TBI patients were randomly split into a training group (n = 686) and a validation dataset (n = 172) in an 8:2 ratio to avoid overfitting in the impact factor analysis. Comparing the data between the two datasets (Table 1), we found no statistically significant differences in all covariates included (P≥0.05), suggesting that the partitioning of our dataset was reasonable and comparable. Patients were categorized into HAP and Non-HAP groups based on whether pneumonia occurred after 2 days of admission (Table 1). Modelling was performed from the data collected in the validation dataset and the predictive models were validated in the validation dataset. Using 10-fold cross-validation, the optimal parameter (lambda) for the LASSO model was determined (Figure 2A). Vertical dashed lines were plotted at the optimal values using the minimum criterion and 1 SE of the minimum criterion.

Table 1 Baseline Data Comparison Between Training and Validation Datasets

Figure 1 Flow chart of the study.

Figure 2 Lasso regression results for admission clinical characteristics and imaging characteristics variables. (A) LASSO regression path diagram. (B) LASSO Cross validation diagram.

A vertical line was plotted at the selected values using 10-fold cross-validation. where the optimal lambda yields 7 feature variables with non-zero coefficients (Figure 2B). We selected 7 non-zero feature variables in the LASSO regression results (Table 2), including age, type of brain herniation, admission GCS, Rotterdam score (Figure 3A–F), glucose, D-dimer, and SIRI.

Table 2 Table of Regression Coefficients for Important Variables Related to HAP (Training Dataset)

Figure 3 Rotterdam CT score corresponding to CT signs. (A) Compression of the cerebral pool. (B) Loss of the cerebral pool. (C) Epidural hemorrhage. (D) Subarachnoid hemorrhage. (E) Ventricular hemorrhage. (F) midline shift ≥5 mm.

Development of Predictive Machine Learning Models and Model Evaluation

We used five algorithms to construct a prediction model for HAP and to visualize the strengths and weaknesses of the models, a confusion matrix was built in this study (Figure 4A–E), and several key performance metrics were calculated for each model: Accuracy, MCC, Recall, and F1-Score (Table 3). According to the results, the LR on Recall, Accuracy, and F1-Score all perform well and also have a high MCC value, indicating that it has a better balance and high-quality prediction performance in all of the pneumonia predictions. Although RF is close to LR in MCC, it has a significant shortfall in Recall. In addition, the analysis of the ROC curve reveals that the LR model exhibits significant performance on the training dataset, as evidenced by a higher area under the curve (AUC value of 0.818) (Figure 5A). While on the validation dataset, the LR model exhibited the highest area under the curve (AUC value of 0.819) compared to the other four models (Figure 5B), highlighting its superior generalization ability. The results indicate that LR is the optimal model, which performs more balanced and effective in identifying HAP.

Table 3 Performance Metrics of the Five Machine Learning Models

Figure 4 (A–E) represents the confusion matrix for each of the five models in the validation dataset. Every classification model’s performance is detailed in the confusion matrix. For example, the LR model has a balanced prediction of 27 false negatives and 14 false positives.

Figure 5 (A and B) represents the ROC curves of the five models in the training and validation datasets, respectively.

Developing and Validating Nomogram Using Logistic Regression Models

Since the prediction of LR performed well, this study continued to use logistic regression to develop and test predictive models. In the training dataset, seven lasso regression-screened variables such as glucose, type of brain hernia, age, GCS at admission, Rotterdam, SIRI, and D-dimer were included in the multivariate logistic regression analysis. The results showed that age, GCS at admission, Rotterdam, SIRI, and D-dimer were the independent risk factors, while glucose and type of brain hernia were not statistically significant in the multivariate model (Figure 6). Meanwhile, the variance inflation factors of the independent risk factors were all covariance diagnostic <2, indicating no multicollinearity between risk factors.

Figure 6 Multiple logistic regression forest plot in the training dataset.

The validation dataset was used to evaluate the performance of the predictive model. The AUCs in the training (Figure 7A) and validation datasets were 0.818 and 0.819, respectively, and the optimal cut-off values were 0.41 and 0.402, respectively. The calibration curve results showed that the deviation of the ideal curve from the actual curve was small. The actual prediction curves were consistent with the ideal curves (B=1000 repetitions), indicating that the prediction model had good accuracy (mean absolute error (train) = 0.005 and mean absolute error (validation) = 0.031). The predictive model was internally validated to have some predictive value (Figure 7). In addition, both the training dataset and validation dataset calibration curves showed good performance according to the Hosmer-Lemeshow test results (p (train) = 0.4307; p(validation) = 0.3387) (Figure 8). The DCA for the training dataset showed that patients could achieve satisfactory net benefit from the predictive model, with a wide range of high-risk thresholds for the DCA (10–85%), which also performed well in the validation dataset (10–80%) (Figure 9).

Figure 7 (A and B) represents the ROC curves in the training and validation datasets, respectively.

Figure 8 (A and B) represents the calibration curves for the training and validation datasets, respectively.

Figure 9 (A and B) represent the DCA curves in the training dataset and validation dataset, respectively.

All independent predictors were considered in the construction of the dynamic nomogram. Compared to the complex logistic regression formulae, the dynamic nomogram are simple straightforward, and more clinically useful (Figure 10). As an example, the first participant in this database (age 83 years, GCS score of 15, Rotterdam score of 3, SIRI of 42.058, and D-dimer of 11.97) had a risk of 49.8% (95% CI=0.4–0.597) of developing HAP. Thus, the dynamic nomogram allows rapid screening of patients at high risk of HAP based on individual patient admissions.

Figure 10 Web-based dynamic nomogram used to predict HAP after TBI. Predicted probabilities and 95% confidence intervals can be found on the right side of the page by entering the precise values of the respective variables on the left side.

Discussion

This is a study using machine algorithm models to predict the incidence of HAP in patients with TBI. In this retrospective study, we used five machine-learning algorithms for the assessment of outcome events, and LR performed well on Recall, Accuracy, and F1-Score, along with high MCC values. In addition, LR showed a higher area under the curve in both the training and validation datasets (train-AUC: 0.818; validation-AUC: 0.819). The results indicate that LR has a good balance of high-quality predictive performance coupled with superior generalization ability. To further explain the model, based on the optimal model, we identified age, GCS on admission, Rotterdam, SIRI, and D-dimer as independent risk factors for HAP in TBI patients. In the validation of the model, the calibration curve and DCA of LR performed well. Finally, a simple and accurate web-based dynamic nomogram was developed in this study (https://shaojieli.shinyapps.io/DynNomapp/).

HAP after TBI is a complication following brain injury due to brain-pulmonary axis interactions.17,18 Pneumonia not only increases the economic burden of patients with traumatic brain injury but is also strongly associated with poor disease prognosis. Predictive models for predicting the prevalence of pneumonia in stroke patients have been explored in several previous studies.19–21 In a population-based study of 1333 patients with spontaneous cerebral hemorrhage, older age, multilobar involvement, extension of ICH to the ventricles, dysphagia, impaired consciousness, and poor muscle strength in male patients were included as independent risk factors for pneumonia after ICH, and the predictive accuracy was shown to be 0.8116 by a ROC curve.19 In a prospective study conducted by Wang et al cerebral hemorrhage, NLR, SII and SIRI at admission had high predictive accuracy for pneumonia and were strongly associated with poorer neurological outcomes at discharge.20 Similarly, in patients with ischaemic stroke (IS), the prevalence of pneumonia is of interest. A retrospective study combining five large medical centers demonstrated that the severity of malnutrition was associated with the prevalence of pneumonia in patients with IS and that the malnutrition index was an independent predictor of pneumonia and improved the predictive efficacy of pneumonia after stroke after adjusting for confounders.22 However, to the best of our knowledge, no studies are reporting early predictive tools that can identify the occurrence of HAP in patients with TBI. Rotterdam CT score is an objective representation of the imaging characteristics of patients with traumatic brain injury and is commonly used in the assessment of the severity of TBI.23 Previous studies have mainly used it for intracranial hypertension monitoring24 and mortality studies.25 In this study, it was used for the first time to predict pneumonia in patients with TBI. The results showed that, in addition to the above data on basic clinical features, combining some of the cranial imaging examinations better improves the diagnostic efficacy of pneumonia.

In this study, we analyzed in depth the risk factors associated with the development of HAP in patients with TBI. The findings revealed that elderly patients and TBI patients with lower GCS were more likely to develop pneumonia, a finding that is consistent with previous studies.19,26,27 In addition, we paid special attention to the use of the Rotterdam CT score in the management of TBI. The Rotterdam CT score is based on several metrics on cranial CT, including midline shift of more than 5 mm, intraventricular hemorrhage, subarachnoid hemorrhage, epidural space-occupying lesions, and compression or disappearance of the basaltic pools,23 and provides an objective, comprehensive and convenient method.27 Previous studies have focused on the use of this score to predict intracranial hypertension24 and mortality,25 and the present study is the first to investigate its application in predicting the risk of HAP in patients with TBI. By combining data on basic clinical characteristics with cranial imaging findings, we found that the Rotterdam CT score is not only an independent predictor of poor outcomes in patients with TBI but also improves the efficacy of pneumonia diagnosis. Our analyses showed that the Rotterdam CT score presented a significant correlation with the occurrence of HAP. We hypothesize that the mechanism is large as follows: elevated Rotterdam CT scores usually reflect severe brain damage in patients, such as compression or loss of brain pools and cerebral hemorrhage. These brain injuries may affect the respiratory control center, leading to impaired regulation of respiratory function, which in turn causes inadequate lung ventilation and retention of respiratory secretions,28 increasing the risk of pulmonary infection in patients. In addition, more severe neurological impairment may be accompanied by decreased cough and swallowing reflexes, making it difficult for patients to effectively clear respiratory secretions,29 thus increasing the risk of HAP. Within the current field of research, the association between SIRI and the development of HAP after TBI and its underlying mechanisms are not fully understood. However, the association of SIRI with stroke-associated pneumonia (SAP) after hemorrhagic stroke has been empirically supported.30 It has been suggested that a sustained inflammatory response may lead to decreased immune system function and increased patient susceptibility to SAP. In addition, SIRI was found to predict the risk of SAP and adverse outcomes at discharge in patients with hemorrhagic stroke (ICH). Further, Yan et al revealed by constructing a response curve (RCS) model that in ischaemic stroke patients, the risk of pneumonia was significantly increased when the SIRI value reached or exceeded 2.74 (OR: 5.82, 95% CI: 4.54–7.49), implying that higher SIRI was a significant risk factor for the development of SAP in acute ischaemic stroke patients.31 This coincides with our findings. Consequently, early identification of patients at high risk of pneumonia and the use of suitable therapies may be made possible by monitoring SIRI. Chen et al found that D-dimer levels were elevated early in the development of TBI and may predict short- and long-term mortality in patients with TBI.32 Elevated D-dimer levels have been linked to the degree of tissue damage following traumatic brain injury, according to recent research.32,33 Tissue factor (TF) is released from the compromised blood-brain barrier throughout the body during the acute phase of traumatic brain injury (TBI), which causes a sharp rise in plasma D-dimer levels.34,35 Elevated D-dimer enhances local and systemic inflammatory responses and immunosuppression by modulating local immune responses and releasing pro-inflammatory factors (eg, IL-6).36 A 2021 correlation study indicated that D-dimer, a traditional marker of fibrinolysis, fluctuates faster than serum CRP and ESR during inflammation and may be effective for early diagnosis of infections,37 which supports our conjecture.

This study combines some of the imaging indices to explore, for the first time, the risk factors associated with the development of HAP in patients with TBI, and produces a simple dynamic column chart to guide the clinic. This study does, however, still have certain shortcomings. First, this is a single-center retrospective study, and although we screened the optimal model from several different machine-learning algorithms to try to ensure the accuracy of the prediction, there is inevitably a selection bias. We hope that clinical data from multiple centers will be used as external validation of the model in the future. Second, this study collected as many clinically available variables as possible, but some other clinical information may have had confounding effects and influenced the risk of pneumonia. For example, influencing patients’ hormone and immunosuppressant use may increase the risk of pneumonia.

Conclusions

In this study, we have implemented five distinct machine learning algorithms to develop predictive models for HAP in patients with TBI. Among these, the LR model has shown superior performance, effectively highlighting its robustness in handling clinical data. This approach has enabled us, for the first time, to pinpoint key independent risk factors for HAP in TBI patients, namely age, Rotterdam, GCS, SIRI, and D-dimer levels. Our predictive model significantly enhances clinical utility by employing a dynamic nomogram format, providing healthcare professionals with an effective tool to swiftly identify patients at risk of pneumonia. This allows for the implementation of individualized treatment strategies for high-risk patients. Specific interventions include targeted nebulization, physical expectoration techniques, and intensive cough training. These are supplemented by the proactive use of prophylactic antibiotics and ongoing monitoring of relevant clinical indicators. Such personalized interventions are designed to optimize respiratory function and reduce the complications associated with hospital-acquired pneumonia (HAP), potentially decreasing its incidence and significantly improving patient outcomes. However, future enhancements to the prediction model can be realized by conducting a larger, well-structured prospective study, which should include external validation to confirm its effectiveness.

Data Sharing Statement

Research data are not available at this time.

Ethical Approval and Consent to Participate

The study protocol was approved by the Ethics Committee of the Second Affiliated Hospital of Fujian Medical University (2023-287). This study was conducted in accordance with the Declaration of Helsinki. The need for written informed consent was waived by the Ethics Committee of the Second Affiliated Hospital of Fujian Medical University due to the non-interventional design of the study. Additionally, this research exclusively utilized previously collected medical record information from which all personally identifiable information had been removed, ensuring no risk to the subjects and no adverse effects on their rights and health.

Author Contributions

All authors made a significant contribution to the work reported, whether that is 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

The work was supported by Quanzhou Science and Technology Program (2022NS067).

Disclosure

The authors declare that they have no competing interests.

References

1. Lefevre-Dognin C, Cogné M, Perdrieau V, Granger A, Heslot C, Azouvi P. Definition and epidemiology of mild traumatic brain injury. Neurochirurgie. 2021;67:218–221. doi:10.1016/j.neuchi.2020.02.002

2. Mukherjee S, Arisi GM, Mims K, Hollingsworth G, O’Neil K, Shapiro LA. Neuroinflammatory mechanisms of post-traumatic epilepsy. J Neuroinflammation. 2020;17:193. doi:10.1186/s12974-020-01854-w

3. Pearn ML, Niesman IR, Egawa J, et al. Pathophysiology associated with traumatic brain injury: current treatments and potential novel therapeutics. Cell Mol Neurobiol. 2017;37:571–585. doi:10.1007/s10571-016-0400-1

4. Bajinka O, Simbilyabo L, Tan Y, Jabang J, Saleem SA. Lung-brain axis. Crit Rev Microbiol. 2022;48:257–269. doi:10.1080/1040841X.2021.1960483

5. Heuer JF, Pelosi P, Hermann P, et al. Acute effects of intracranial hypertension and ARDS on pulmonary and neuronal damage: a randomized experimental study in pigs. Intensive Care Med. 2011;37:1182–1191. doi:10.1007/s00134-011-2232-2

6. Parker A, Fonseca S, Carding SR. Gut microbes and metabolites as modulators of blood-brain barrier integrity and brain health. Gut Microbes. 2020;11:135–157. doi:10.1080/19490976.2019.1638722

7. Li Y, Liu C, Xiao W, Song T, Wang S. Incidence, risk factors, and outcomes of ventilator-associated pneumonia in traumatic brain injury: a meta-analysis. Neurocrit Care. 2020;32:272–285. doi:10.1007/s12028-019-00773-w

8. Martin-Loeches I, Rodriguez AH, Torres A. New guidelines for hospital-acquired pneumonia/ventilator-associated pneumonia: USA vs. Eur Curr Opin Crit Care. 2018;24:347–352. doi:10.1097/MCC.0000000000000535

9. Kalil AC, Metersky ML, Klompas M, et al. Management of adults with hospital-acquired and ventilator-associated pneumonia: 2016 clinical practice guidelines by the infectious diseases society of America and the American thoracic society. Clin Infect Dis. 2016;63:e61–111. doi:10.1093/cid/ciw353

10. Klompas M, Branson R, Cawcutt K, et al. Strategies to prevent ventilator-associated pneumonia, ventilator-associated events, and nonventilator hospital-acquired pneumonia in acute-care hospitals: 2022 Update. Infect Control Hosp Epidemiol. 2022;43:687–713. doi:10.1017/ice.2022.88

11. Schwarz S. Prophylactic antibiotic therapy for preventing poststroke infection. Neurotherapeutics. 2016;13:783–790. doi:10.1007/s13311-016-0466-y

12. Meisel A. Preventive antibiotic therapy in stroke: pASSed away? Lancet. 2015;385:1486–1487. doi:10.1016/S0140-6736(15)60076-9

13. Harms H, Prass K, Meisel C, et al. Preventive antibacterial therapy in acute ischemic stroke: a randomized controlled trial. PLoS One. 2008;3:e2158. doi:10.1371/journal.pone.0002158

14. Dragan V, Wei Y, Elligsen M, Kiss A, Walker SAN, Leis JA. Prophylactic antimicrobial therapy for acute aspiration pneumonitis. Clin Infect Dis. 2018;67:513–518. doi:10.1093/cid/ciy120

15. Reizine F, Asehnoune K, Roquilly A, et al. Effects of antibiotic prophylaxis on ventilator-associated pneumonia in severe traumatic brain injury. A post hoc analysis of two trials. J Crit Care. 2019;50:221–226. doi:10.1016/j.jcrc.2018.12.010

16. Pei Y, Huang Y, Pan X, et al. Nomogram for predicting 90-day mortality in patients with Acinetobacter baumannii-caused hospital-acquired and ventilator-associated pneumonia in the respiratory intensive care unit. J Int Med Res. 2023;51:3000605231161481. doi:10.1177/03000605231161481

17. Weber DJ, Allette YM, Wilkes DS, White FA. The HMGB1-RAGE inflammatory pathway: implications for brain injury-induced pulmonary dysfunction. Antioxid Redox Signal. 2015;23:1316–1328. doi:10.1089/ars.2015.6299

18. Kerr N, de Rivero Vaccari JP, Dietrich WD, Keane RW. Neural-respiratory inflammasome axis in traumatic brain injury. Exp Neurol. 2020;323:113080. doi:10.1016/j.expneurol.2019.113080

19. Wang Y, Chen Y, Chen R, et al. Development and validation of a nomogram model for prediction of stroke-associated pneumonia associated with intracerebral hemorrhage. BMC Geriatr. 2023;23:633. doi:10.1186/s12877-023-04310-5

20. Wang R-H, Wen W-X, Jiang Z-P, et al. The clinical value of neutrophil-to-lymphocyte ratio (NLR), systemic immune-inflammation index (SII), platelet-to-lymphocyte ratio (PLR) and systemic inflammation response index (SIRI) for predicting the occurrence and severity of pneumonia in patients with intracerebral hemorrhage. Front Immunol. 2023;14:1115031. doi:10.3389/fimmu.2023.1115031

21. Li J, Luo H, Chen Y, et al. Comparison of the predictive value of inflammatory biomarkers for the risk of stroke-associated pneumonia in patients with acute ischemic stroke. Clin Interv Aging. 2023;18:1477–1490. doi:10.2147/CIA.S425393

22. Li D, Liu Y, Jia Y, et al. Association between malnutrition and stroke-associated pneumonia in patients with ischemic stroke. BMC Neurol. 2023;23:290. doi:10.1186/s12883-023-03340-1

23. Amakhian AO, Obi-Egbedi-Ejakpovi EB, Morgan E, Adeyekun AA, Abubakar MM. Correlation between optic nerve sheath diameter at initial head CT and the Rotterdam CT score. Cureus. 2023;15:e41995. doi:10.7759/cureus.41995

24. Cnossen MC, Huijben JA, van der Jagt M, et al. Variation in monitoring and treatment policies for intracranial hypertension in traumatic brain injury: a survey in 66 neurotrauma centers participating in the CENTER-TBI study. Crit Care. 2017;21:233. doi:10.1186/s13054-017-1816-9

25. Voormolen DC, Zeldovich M, Haagsma JA, et al. Outcomes after complicated and uncomplicated mild traumatic brain injury at three-and six-months post-injury: results from the CENTER-TBI study. J Clin Med. 2020;9:1525. doi:10.3390/jcm9051525

26. Zawiah M, Khan AH, Abu Farha R, et al. Predictors of stroke-associated pneumonia and the predictive value of neutrophil percentage-to-albumin ratio. Postgrad Med. 2023;135:681–689. doi:10.1080/00325481.2023.2261354

27. Garza N, Toussi A, Wilson M, Shahlaie K, Martin R. The increasing age of TBI patients at a single level 1 trauma center and the discordance between GCS and CT Rotterdam scores in the elderly. Front Neurol. 2020;11:112. doi:10.3389/fneur.2020.00112

28. Meyfroidt G, Bouzat P, Casaer MP, et al. Management of moderate to severe traumatic brain injury: an update for the intensivist. Intensive Care Med. 2022;48:649–666. doi:10.1007/s00134-022-06702-4

29. Terré R, Mearin F. Prospective evaluation of oro-pharyngeal dysphagia after severe traumatic brain injury. Brain Inj. 2007;21:1411–1417. doi:10.1080/02699050701785096

30. Xu M, Wang J, Zhan C, et al. Association of follow-up neutrophil-to-lymphocyte ratio and systemic inflammation response index with stroke-associated pneumonia and functional outcomes in cerebral hemorrhage patients: a case controlled study. Int J Surg. 2024. doi:10.1097/JS9.0000000000001329

31. Yan D, Dai C, Xu R, Huang Q, Ren W. Predictive ability of systemic inflammation response index for the risk of pneumonia in patients with acute ischemic stroke. Gerontology. 2023;69:181–188. doi:10.1159/000524759

32. Chen X, Wang X, Liu Y, et al. Plasma D-dimer levels are a biomarker for in-hospital complications and long-term mortality in patients with traumatic brain injury. Front Mol Neurosci. 2023;16:1276726. doi:10.3389/fnmol.2023.1276726

33. Kowalski EA, Chen J, Hazy A, et al. Peripheral loss of EphA4 ameliorates TBI-induced neuroinflammation and tissue damage. J Neuroinflammation. 2019;16:210. doi:10.1186/s12974-019-1605-2

34. Hubbard WB, Sim MMS, Saatman KE, Sullivan PG, Wood JP. Tissue factor release following traumatic brain injury drives thrombin generation. Res Pract Thromb Haemost. 2022;6:e12734. doi:10.1002/rth2.12734

35. Kwaan HC. The central role of fibrinolytic response in trauma-induced coagulopathy: a hematologist’s perspective. Semin Thromb Hemost. 2020;46:116–124. doi:10.1055/s-0039-3402428

36. Medcalf RL, Keragala CB, Draxler DF. Fibrinolysis and the immune response in trauma. Semin Thromb Hemost. 2020;46:176–182. doi:10.1055/s-0040-1702170

37. Chen X, Li H, Zhu S, Wang Y, Qian W. Is D-dimer a reliable biomarker compared to ESR and CRP in the diagnosis of periprosthetic joint infection? Bone Joint Res. 2020;9:701–708. doi:10.1302/2046-3758.910.BJR-2020-0172.R2

Comments (0)

No login
gif