Prognostic Significance of Plasma Insulin Level for Deep Venous Thrombosis in Patients with Severe Traumatic Brain Injury in Critical Care

Whether insulin resistance underlies deep venous thrombosis (DVT) development in patients with severe traumatic brain injury (TBI) is unclear. In this study, the association between plasma insulin levels and DVT was analyzed in patients with severe TBI. A prospective observational study of 73 patients measured insulin, glucose, glucagon-like peptide 1 (GLP-1), inflammatory factors, and hematological profiles within four preset times during the first 14 days after TBI. Ultrasonic surveillance of DVT was tracked. Two-way analysis of variance was used to determine the factors that discriminated between patients with and without DVT or with and without insulin therapy. Partial correlations of insulin level with all the variables were conducted separately in patients with DVT or patients without DVT. Factors associated with DVT were analyzed by multivariable logistic regression. Neurological outcomes 6 months after TBI were assessed. Among patients with a mean (± standard deviation) age of 53 (± 16 years), DVT developed in 20 patients (27%) on median 10.4 days (range 4–22), with higher Acute Physiology and Chronic Health Evaluation II scores but similar Sequential Organ Failure Assessment scores and TBI severity. Patients with DVT were more likely to receive insulin therapy than patients without DVT (60% vs. 28%; P = 0.012); hence, they had higher 14-day insulin levels. However, insulin levels were comparable between patients with DVT and patients without DVT in the subgroups of patients with insulin therapy (n = 27) and patients without insulin therapy (n = 46). The platelet profile significantly discriminated between patients with and without DVT. Surprisingly, none of the coagulation profiles, blood cell counts, or inflammatory mediators differed between the two groups. Patients with insulin therapy had significantly higher insulin (P = 0.006), glucose (P < 0.001), and GLP-1 (P = 0.01) levels and were more likely to develop DVT (60% vs. 15%; P < 0.001) along with concomitant platelet depletion. Insulin levels correlated with glucose, GLP-1 levels, and platelet count exclusively in patients without DVT. Conversely, in patients with DVT, insulin correlated negatively with GLP-1 levels (P = 0.016). Age (P = 0.01) and elevated insulin levels at days 4–7 (P = 0.04) were independently associated with DVT. Patients with insulin therapy also showed worse Glasgow Outcome Scale scores (P = 0.001). Elevated insulin levels in the first 14 days after TBI may indicate insulin resistance, which is associated with platelet hyperactivity, and thus increasing the risk of DVT.


Introduction
Patients with traumatic brain injury (TBI), especially those in the neurological intensive care unit (neuro-ICU), are at high risk of developing venous thromboembolism (VTE), including deep venous thrombosis (DVT) and pulmonary embolism (PE) [1][2][3]. The incidence of VTE was significantly higher than that of bleeding or recurrent TBI in hospitalized elderly patients with TBI [4]. The reported incidence of DVT following TBI ranged from 20 to 32% independent of anticoagulation therapy [2,3,[5][6][7]. We previously reported that the incidence of isolated distal DVT was 62% in patients after acute brain injury [8]. DVT is associated with increased morbidity and mortality, including the risk of fatal PE [9].
There is evidence that early initiation of mechanical and pharmacologic VTE prophylaxis within 24-72 h after admission is effective at reducing VTE in patients with TBI [10,11]. However, given the risk of hypocoagulopathy with prolonged intracranial bleeding and progression of hemorrhagic lesions, early (< 72 h) pharmacological prophylaxis of VTE is not generally recommended in patients who undergo neurosurgical interventions for TBI [9,12,13]. If chemoprophylaxis is withheld for more than 7 days, later initiation may increase the risk of VTE [1]. Therefore, chemoprophylaxis for VTE in patients with TBI has been a critical dilemma in the neuro-ICU.
Taking advantage of the existing pathophysiological knowledge and exploring innovative ideas may overcome this challenge. Patients with TBI lack key causal factors for coagulopathy, e.g., substantial blood loss, extensive fluid resuscitation, or a hypothermic state during the acute phase, suggesting that TBI-induced coagulopathy follows a distinct pathogenic pathway [14]. Several studies have demonstrated that insulin resistance, characterized by hyperglycemia with hyperinsulinemia, may affect thrombosis development in cardiovascular disease [15]. Insulin resistance is associated with a prothrombotic state in patients with type 2 diabetes, primarily due to impaired fibrinolysis [15]. Platelets obtained from patients with diabetes were hyperreactive, i.e., increased adhesiveness, exaggerated aggregation and thrombus generation [16]. In a community-based observational study, an increased risk of VTE was found in participants with increasing insulin resistance [17]. Insulin resistance assessed by homeostasis model assessment for insulin resistance (HOMA-IR) and fasting plasma insulin were the only metabolic variables with significant relationships to adenosine phosphate (ADP)-induced platelet aggregation [18]. Insulin resistance is also associated with endothelial damage [19,20], increased levels of prothrombotic factors (e.g., plasminogen activator inhibitor-1 [21,22]), fibrinogen [23] and von Willebrand factor antigen [17,24], and circulating tissue factor in platelets and monocytes in healthy participants [25].
Previous studies have demonstrated that insulin can exert direct antiplatelet effects [16,26]. Insulin stimulates AMP-activated protein kinase and Akt in a phosphatidylinositol 3-kinase-dependent manner [27] to decrease platelet Ca 2+ influx and attenuate agonist-induced platelet activation [28]. The insulin receptor substrate-1 gene was associated with a hyperreactive platelet phenotype in patients with coronary artery disease with type 2 diabetes [29]. In addition, knocking out the insulin receptor in murine megakaryocytes/platelets causes thrombocytosis and increases the platelet count and volume but does not cause platelet hyperactivity [30]. Insulin resistance is a major determinant of platelet activation, as shown by aggregation in obese adolescents [31]. Nitric oxide is a short-lived gaseous messenger for platelet adhesion and aggregation inhibition. The transport of L-arginine, a substrate of nitric oxide synthase, was diminished in platelets from obese patients and patients with metabolic syndrome and correlated negatively with insulin resistance [32].
TBI, as a severe stress, could provoke central insulin resistance in humans [33] and mice [34], possibly due to central glutamate excitotoxicity [35]. TBI may also induce critical hyperglycemia [36,37] and glucose intolerance, possibly via endoplasmic reticulum stress in the brain that elevates sympathetic nerve tone [38] or through adipose-derived monocyte chemoattractant protein-1 to impair peripheral insulin sensitivity [39]. Hence, peripheral insulin resistance following TBI in humans [40] or in mice [41,42] is an independent predictor of mortality [40]. Stress, as in TBI, could also increase coagulation factors through adipose-derived monocyte chemoattractant protein-1 but not give rise to thrombus formation [39]. In a murine model of TBI, acute platelet dysfunction occurred followed by rebound platelet hyperaggregation [43]. There was a dose-response relationship between TBI severity and the degree of platelet dysfunction [44]. Coagulopathy, defined as a platelet count ≤ 150,000/ μL, was found to be an early predictor of mortality [45]. Therefore, antithrombotic [46] and antiplatelet therapies [44,[47][48][49] are commonly used by patients with TBI.
Considering that insulin resistance was associated with platelet dysfunction and thrombosis development in metabolic disease, we wondered whether insulin resistance induced by TBI could also be associated with DVT development in patients with TBI. We performed a prospective observational study in 73 consecutive patients with TBI admitted to the intensive care unit (ICU), with the following objectives: (1) to investigate the kinetic associations between insulin level, coagulation and hemostatic profile, and metabolic index with DVT development; (2) to provide insight into the association of insulin levels with key host responses implicated in the pathogenesis of DVT; (3) to determine whether the metabolic index, coagulation and hemostatic profile, and key host responses were changed in patients with insulin therapy; and (4) to establish a predictive model for DVT, including insulin levels, in TBI.

Ethics Statement
This prospective study was conducted in a 28-bed neuro-ICU at the First Affiliated Hospital of University of Science and Technology of China between January 2019 and July 2020. All procedures involving patients were performed in compliance with the Declaration of Helsinki and National Institutes of Health guidelines and were approved by the hospital ethics committee (reference number: 2021-RE-017). All patients' relatives gave informed consent before being included in the study.

Patient inclusion and demographic data
Immobilized patients with TBI (n = 73) fulfilling the following inclusion criteria were consecutively included: (1) TBI confirmed on head computed tomography (CT) scan; (2) Glasgow Coma Score (GCS) ≤ 8; (3) presenting within 24 h of TBI; (4) age > 18 years old; (5) without previous neurological and diabetic disease; and (6) provided informed consent. A total of four patients died in the hospital. One patient died after completing the first sampling, two patients died after the second sampling, and one patient died after the third sampling. Eleven patients missed the fourth sampling due to being discharged from the hospital. Thus, there were 73, 72, 70, and 58 patients available at the four time points used for analysis, respectively.
TBI severity was determined on the basis of the lowest recorded GCS, length of coma, and/or posttraumatic amnesia, and injury features when admitted [50]. With approval from the institutional ethics committee, trained physicians collected the following data: demographic data, comorbidity, medication use, daily physiologic measurements, and vital signs. Thus, patient characteristics, such as the initial Acute Physiology and Chronic Health Evaluation (APACHE) II score, Sequential Organ Failure Assessment (SOFA) score, severity of disease, presence of sepsis, and therapeutic interventions in the neuro-ICU, were prospectively recorded.

Clinical Monitoring and Treatment
We followed our standardized diagnostic and treatment regimen (< 24 h of admission) [51,52]. Our treatment protocol in the ICU included hourly neurologic monitoring, continuous invasive blood pressure and body temperature measurements, daily optical nerve monitoring, cerebral hemodynamic monitoring by transcranial Doppler (DWL; Doppler Box, Germany), and the application of nimodipine for 14-21 consecutive days [52]. All patients were mechanically ventilated and treated with continuous sedation and analgesia (midazolam and remifentanil). The treatment protocol aimed to keep the optic nerve sheath diameter, which reflects intracranial pressure, between 0.55 and 0.57 cm as detected by ultrasound (Vivid iq; GE Healthcare, Chicago, IL) [53].
Nutritional support was delivered according to the protocols of the American Society for Parenteral and Enteral Nutrition [54]. Enteral nutritional support (Peptison Enteral Nutritional Suspension/Enteral Nutritional Suspension; Nutricia Co., Ltd. WuXi, JiangSu, China) was initiated within 24-48 h after admission once the patient was hemodynamically stable. The nutrition goal was established to the standard with 20-25 nonprotein kilocalories per kilogram (kg) of body weight per 24 h over the first week. The protein supply was in the range of 1.5-2.5 g/kg/d. Parenteral support was initiated on days 7-10 if 70% of goal nutrition could not be attained enterally.

Ultrasonic Surveillance of DVT
Deep venous thrombosis prophylaxis was conducted according to the guidelines [9]. Beginning at the time of ICU admission, patients received physical prophylaxis of the lower extremities with an adjunctive intermittent venous pneumatic compression device (Lifotronic Airpro 800; Pumen Science and Technology Lt.D., Shenzhen, China) [55]. Although our patients in the neuro-ICU were severely injured, intracranial lesions were usually stable after 1 week of TBI, and the patients started chemical DVT prophylaxis, e.g., low molecular weight heparin (LMWH) (3000-6000 IU/day) subcutaneously approximately 2 weeks after TBI [9]. The LMWH dose was often held for procedures leading to therapy interruption.
As part of routine practice, ultrasonic surveillance of DVT in upper and lower extremities was performed immediately after admission using ultrasound (Vivid iq, GE Healthcare) and once weekly thereafter [56][57][58][59]. Ultrasound surveillance was used more frequently if patients displayed DVT symptoms such as unexplained fever, unilateral extremity edema, desaturation, or increased work of breathing [60]. If the patients presented with signs of hypoxemia and dyspnea with complications of hypotension, they were subjected to pulmonary CT angiography to diagnose PE.

Intensive Insulin Therapy
Whole blood glucose levels were measured by finger sticks using Accu-Chek Performa (Roche, Beijing, China). If patients showed consistent high glucose levels (> 10.0 mmol/L) even after cessation of glucose-containing resuscitation fluid ≥ 1 h, they were immediately started on insulin therapy, as previously reported [61]. Human insulin (Wan-Bang Biochemical Medicine Ltd.) was infused intravenously at 1 IU/ml at a rate of 2-20 IU/h within 2-4 h via a microinfusing pump (MP-30, Medcaptain, Shenzhen, China). The target for conventional glycemic control was maintained at 8.0-10.0 mmol/L [36,62]. During insulin infusion, glucose levels were measured repeatedly at fixed time points between 1 and 2 h after the last glucose measurement [61].

Blood Sample Collection and Analysis
During the first 14 days after TBI, patients were in a semirecumbent position for at least 1 h in the morning at the predefined time periods: postinjury day (PID) 1-3, PID 4-7, PID 8-10, and PID 11-14. Blood (4 mL) was collected from an arterial catheter into a glass vacutainer containing ethylenediaminetetraacetic acid dipotassium salt dehydrate (RongYe Science and Technology Ltd., JiangSu, China). Blood samples were centrifuged at 12,000 rpm × 15 min at 4 °C, and the plasma was stored at − 80 °C. Because the patients were continually infused with enteral nutrition at a rate of 100-200 ml/h often overnight, we could not guarantee that they were always in the fasted state when blood was sampled.
During the same 24-h period, blood was also drawn from the antebrachial vein in the morning (6:30-7:30 a.m.) at the predefined time periods. The patients were subjected to assessments of hematological parameters, including total white cell count, coagulation profile, glucose, creatinine, bilirubin, and total albumin, in a clinical laboratory using an automatic biochemical analyzer (Cobas C8000, Roche, Switzerland).

Clinical Outcomes
We used DVT as our primary end point, mortality in the ICU or at 6 months, length of stay in the ICU, and Glasgow Outcome Scale (GOS) at 6 months after TBI as the secondary outcomes. Neurofunctional outcomes were evaluated prospectively 6 months after injury with a 5-point version of the GOS rated from death to symptom-free full recovery by telephone interview. GOS is defined as follows: 1 = death, 2 = persistent vegetative state, 3 = severe disability, 4 = moderate disability, and 5 = good recovery [50]. All clinical and outcome end points were classified according to a priori criteria and adjudicated at a weekly research meeting, as previously described.

Statistical Analysis
Continuous variables were expressed as the mean and standard deviation (SD) or medians and interquartile range. Unpaired t and one-way analysis of variance (ANOVA) tests were used to analyze normally distributed continuous variables, whereas the Mann-Whitney U-test and Kruskal-Wallis test were used to analyze nonnormally distributed continuous variables. Categorical data were reported as frequency distributions and analyzed using the χ 2 test. Partial correlation of insulin level with other parameters in either thrombotic or nonthrombotic patients was performed by controlling the factor of "time. " Associations between blood test values and the development of DVT over time were tested by using repeated measures ANOVA.
We used multiple linear regression models for the continuous outcomes and multiple logistic regression models for the binary outcomes. A regression coefficient (β) and P value are reported for significant variables in this analysis. To assess the independent association of insulin level with DVT, a multivariable logistic regression model was developed to estimate odds ratios (ORs) and 95% confidence intervals (CIs). Candidate variables with a P < 0.05 on unadjusted analyses were included. These included the following potential confounders: APACHE II, age, ICU stay, ventilation days, insulin levels on PID 1-10, maximum (max) insulin level, and platelet count on PID 4-10. Receiver operating characteristic curve analysis was used to estimate the area under the curve for DVT. Then, the significance of each insulin level on the respective PID, with both positive predictive value and negative predictive value, was assessed. The clinical prediction model including the above potential confounders was finally established for DVT. All tests of significance used a two-sided P < 0.05. Statistics were performed by using Prism 9.0 (GraphPad Software, San Diego, CA) software and SPSS 19.0 (IBM, NJ).

Patient Characteristics
Arterial blood samples (n = 272) were drawn from 73 patients during the 14 days immediately after admission. Nineteen women (26%) and 54 men (74%) were included with a mean (± SD) age of 53 (± 16) years. Patients were enrolled at a median (interquartile range) of 1.6 days (0.6-2.0 d) after injury. Two patients (3%) had moderately severe injuries, 33 patients (45%) had severe injuries, and 38 patients (52%) had extremely severe injuries. The mortality rate was 5% (n = 4) in the hospital and 18% (n = 13) at 6 months after injury. None of the patients had a history of diabetes in the annual physical examination, and 13 had hypertension (18%).
The incidence of DVT was 27% (n = 20) in the neuro-ICU, where two patients experienced PE simultaneously. The median time from neuro-ICU admission to DVT diagnosis was 10.4 days (range 4-22 days). Patients with and without DVT had similar TBI severity (P = 0.53), admission GCS (P = 0.95), and SOFA scores (P = 0.84). However, the SOFA score during infection was much higher in patients with DVT than in patients without DVT (P = 0.010). The incidence of cerebral hernia (P = 0.46) and a history of hypertension (P = 0.76) were also comparable between the two groups. However, patients with DVT were much older and had higher initial APACHE II scores (P = 0.036). Even if the two groups had similar chances for neurosurgery, patients with DVT were much more likely to receive lumbar cerebrospinal fluid drainage (P < 0.0001) and more likely to receive a tracheotomy (P = 0.003) and consequent mechanical ventilation (P = 0.049) than patients without DVT. Patients with DVT were more likely to receive LMWH therapy than those without DVT (60% vs. 17%, P < 0.001). Intriguingly, patients with DVT were more likely to receive insulin therapy than patients without DVT (P = 0.012). No patient ever received renal replacement therapy.
Regarding the secondary outcomes, mortality was similar between patients with and without DVT in either hospital (P = 0.21) or at 6 months (P = 0.70), as was the 6-month GOS score (P = 0.10). Although the incidence of pneumonia (P = 0.67) and sepsis (P = 0.30) did not differ between the two groups, patients with DVT stayed longer in both the ICU and hospital (Table 1).

Insulin Levels, Unlike Other Metabolic Parameters, Were Differentially Elevated in Patients with DVT
To analyze the kinetics of insulin levels and their association with thrombosis onset, we measured insulin levels within 14 days after TBI. The mixed-effects model revealed that insulin levels did not show significant time dependency (P = 0.18). Neither sex (P = 0.16, Suppl. Two-way ANOVA of the patients with and without DVT showed that insulin levels differed significantly between the two groups (F(1,71) = 8.01, P = 0.01), but neither the effect of time (P = 0.25) nor the group × time interaction (F(3, 213) = 0.490, P = 0.69) was found. Post hoc analysis revealed that insulin levels were significantly increased on PID 1-3 in patients with DVT compared with patients without DVT (Fig. 1). The insulin level in patients with DVT remained significantly higher until PID 8-10 before it declined slightly and reached a level comparable to that in patients without DVT at PID 11-14. The cutoff values for the insulin levels used to predict DVT are shown. This result showed a significant association between insulin levels and the development of thrombosis starting from PID 1-3 to PID 8-10 (Fig. 1a). To our knowledge, this is the first report showing elevated insulin levels in thrombotic versus Table 1 Clinical characteristics of patients with and without deep vein thrombosis during the first 14 days after TBI APACHE, Acute Physiology and Chronic Health Evaluation; DVT, deep vein thrombosis; GCS, Glasgow Coma Score; GOS, Glasgow Outcome Scale; ICU, intensive care unit; IQR, interquartile range; LMWH, low molecular weight heparin; SOFA, Sequential Organ Failure Assessment; TBI, traumatic brain injury *P < 0.05, **P < 0.01, ***P < 0.001  nonthrombotic patients within 14 days after TBI, which covers the peak interval of thrombosis development.

Parameters
Mixed-effects analysis of glucose revealed neither a significant difference between the DVT and non-DVT groups (P = 0.23) nor the effect of time on its level (P = 0.18) (Fig. 1b). GLP-1 is an incretin that can enhance insulin secretion from pancreatic β cells to lower blood glucose in a glucose-dependent manner [63]. Two-way ANOVA of GLP-1 levels revealed no difference between the two groups throughout the 14 days (F(1, 71) = 0.057, P = 0.81) (Fig. 1c).

Platelet Profile is Differentially Changed in Patients with DVT
Although the platelet numbers were progressively increased in both the patients with DVT and patients without DVT, they were consistently lower in patients with DVT than in patients without DVT. Mixed-effects analysis revealed a significant effect of DVT (P = 0.008), time (P < 0.001), and their interaction (P = 0.026) on platelet number (Fig. 2a). Moreover, platelet volume, an indication of platelet activation, was decreased in patients without DVT. In contrast, it remained at a   (Fig. 2b). However, platelet distribution, which represented its volume variation, tended to be decreased in patients with DVT relative to patients without DVT (P = 0.05; Fig. 2c).
Unexpectedly, all the coagulation parameters were indistinguishable between the two groups over the 2 weeks after injury. D-dimer, the fibrinolytic product, was significantly decreased in patients with DVT relative to patients without DVT during the progressive period of the injury (days 4-10). However, it was indistinguishable between the two groups at the beginning (days 1-3) and the recovery period (days 10-14) (Fig. 3). Fibrinogen, an important mediator of platelet aggregation, did not discriminate significantly between patients with and without DVT. The kinetics of the soluble P-selectin level were not affected by either DVT or insulin therapy (Suppl. Fig. 3).
The number of patients receiving insulin therapy was 12 at PID 1-3, 19 at PID 4-7, 23 at PID 8-10, and 21 at PID 11-14. The insulin therapy led to consistently higher insulin levels relative to patients without insulin therapy (P = 0.006; Fig. 4a). Even so, the blood glucose level in patients with insulin therapy remained higher than that in those without insulin therapy (P < 0.001; Fig. 4b), demonstrating resistance to insulin therapy. Thus, the insulin level, which represents the insulin dose used to maintain euglycemia, could be used as an insulin resistance marker. Interestingly, patients with insulin therapy had higher GLP-1 levels (P = 0.010; Fig. 4c) and were more likely to develop DVT (60% vs. 15%, P < 0.001) than patients without insulin therapy.

Platelet profile was differentially changed in patients with insulin therapy
The platelet counts in patients with insulin therapy decreased progressively compared with those without insulin therapy (P < 0.0001, Fig. 5a). Conversely, the mean platelet volume (MPV), which represents platelet activation, remained at a consistent level in patients subjected to insulin therapy, in contrast with a progressive decline in patients without insulin therapy (P = 0.017, Fig. 5b). Platelet distribution width (PDW), which represents the variation in platelet volume, did not differ between the two groups (P = 0.76, Fig. 5c).

Interaction between insulin resistance/insulin therapy and DVT
Partial correlation analysis demonstrated that insulin resistance occurred exclusively in patients with DVT. In patients without DVT, plasma insulin levels correlated significantly with GLP-1, glucose, creatinine levels and white blood cell, neutrophil, and platelet counts. However, in patients with DVT, insulin levels correlated with none of the above parameters except GLP-1 levels (negative correlation, r = − 0.297, P = 0.016), indicating that the patients with DVT were resistant to the insulin-stimulating effect of GLP-1 ( Table 2).
Patients with insulin therapy were more likely to develop DVT (12 of 27) than those without insulin therapy (8 of 46) (44% vs. 17%, P = 0.01). To avoid the confounding factor of insulin infusion, we compared the insulin levels between the patients with and without DVT in the subgroups of patients with or without insulin therapy. In both subgroups (Fig. 6), insulin levels were comparable between the patients with and without DVT, indicating that only insulin therapy determined DVT development. Taken together, it can be concluded that the patients receiving insulin therapy appeared to be more resistant to insulin and were more likely to develop DVT.

Distinct effect of DVT and insulin therapy on intestinal barrier permeability
The 14-day profile of the three parameters tested for intestinal barrier permeability, which originates from the intestinal tract, was severely suppressed in patients with DVT relative to patients without DVT (Fig. 8a). In contrast, it was indistinguishable between the patients with and without insulin therapy (Fig. 8b). It has been reported that, in the first few weeks after TBI, patients have reduced intestinal contractile activity and absorption as a result of the disruption of the brain-gut axis [64]. Consistent with the higher infection SOFA score in patients with DVT versus patients without DVT, more severe intestinal dysfunction in patients with DVT could result in a decrease in the transfer of metabolic products from the intestinal tract into the blood than patients without DVT, resulting in lower plasma levels of the three parameters.

Insulin level was associated with DVT onset
Multiple logistic regression including all the parameters differed significantly between the groups with and without DVT (APACHE II, age, ICU stay, ventilation days, insulin PID 1-3, insulin PID 4-7, insulin PID 8-10, platelet PID 4-7, and platelet PID 8-10), as well as 14-day max insulin level which revealed that only age (P = 0.01) and insulin PID 4-7 (P = 0.04) were independently associated with the increased risk of thrombotic events during the ICU stay (Table 3). This model could also significantly predict DVT development (receiver operating characteristic 0.923, 95% CI 0.858-0.989, P < 0.0001), with negative predictive value = 89.36% and positive predictive value = 64.71% (Fig. 9).

Table 2 Partial correlation of plasma insulin levels with plasma metabolic and biochemical parameters and blood leucocytes at the four defined phases in patients with or without thrombosis during the first 14 days after TBI
GLP-1, glucagon-like peptide-1; IL, interleukin; TBI, traumatic brain injury; WBC, white blood cell *P < 0.05, **P < 0.01, ***P < 0.001 significantly to ICU stay. Patients with insulin therapy also showed worse GOS than patients without insulin therapy 6 months after TBI (P = 0.001).

Discussion
Traumatic brain injury is characterized by a delayed prothrombotic state emerging partly due to hyperactive   platelets and fibrinolysis shutdown [65]. VTEs are major causes of hospital-related morbidity and mortality in patients with traumatic injury [66]. Recent studies demonstrated that von Willebrand factor mediated platelet adhesion [67], the interaction of platelet-neutrophil extracellular traps [68] and thrombin generation on the activated platelet surface [69], likely play important roles in DVT development. A critical question is how a localized brain injury rapidly disseminates to alter homeostasis systemically, when it is not directly affected as in patients without extracranial trauma and hemorrhagic shock [70]. In this prospective study, we provide evidence that insulin resistance induced by TBI was strongly associated with platelet activity and subsequent DVT occurrence. The plasma insulin level of patients between days 4-7 after TBI could independently predict DVT during their stay in the neuro-ICU.

Insulin Resistance in Patients with TBI
Insulin resistance is defined as the reduced capacity of target cells to take up glucose in response to insulin, leading to elevated blood glucose and compensatory increases in insulin [71]. Measures of insulin resistance in the clinical setting include fasting insulin and HOMA-IR. However, it was impossible to critically measure insulin resistance in our patients who received insulin therapy as soon as they were diagnosed with hyperglycemia. Plasma insulin, measured in patients receiving insulin therapy, included both endogenous and exogenous sources. As shown in Fig. 8, there was no difference in insulin levels between the patients with and without thrombosis regardless of insulin therapy. This result suggested that receiving insulin therapy accounted for the insulin difference between the patients with and without thrombosis, as shown in Fig. 1a. Therefore, plasma insulin levels could reflect the insulin dose required to maintain euglycemia and could also be regarded as a measure of insulin resistance.
We found that glucose levels were higher in patients receiving insulin therapy than in those without insulin therapy, as were insulin and GLP-1 levels. Although in Fig. 1, the thrombotic patients had higher insulin levels than the nonthrombotic patients, both the glucose and GLP-1 levels were comparable between the two groups. However, in Table 2, plasma insulin correlated closely with glucose, GLP-1, and platelet count exclusively in patients without DVT instead of in patients with DVT, demonstrating that insulin resistance was only present in patients with DVT. Furthermore, insulin levels correlated negatively with GLP-1 levels in patients with DVT, indicating that patients with DVT even had GLP-1 resistance. Incretin-based therapies represent a novel and promising intervention to treat hyperglycemia in hospital settings [63]. Our novel finding warrants application in critical care with certain precautions. Our findings that TBI could induce acute insulin resistance following a single injury were consistent with the literature [33,40].

Platelets are Involved in Thrombosis Development
Platelets have a pivotal role in homeostasis via two key pathways: facilitation of procoagulant events and regulation of endothelial integrity [72]. Platelets become hyperactive following their initial dysfunction, increasing the risk of thrombosis seen in patients with TBI. Thrombin generated from extrinsic coagulation [73] can activate platelets in the acute phase of TBI. Similar to the insulin level, there was a distinct platelet profile between thrombotic and nonthrombotic patients (Fig. 2), suggesting that platelets, an essential hemostatic ingredient, may contribute to thrombosis. It was recently demonstrated that critical TBI correlates with a significant worsening of traumatic coagulopathy in comparison with moderate/severe TBI [74]. Surprisingly, even though our patients were severely injured, the coagulation profile between the patients with and without DVT was indistinguishable (Fig. 3). Recently, it was reported that platelets from patients with transient ischemic attack were activated and aggregated, forming thicker fibrin fibers; however, the function of the coagulation cascade was normal, as examined by thromboelastography [75]. In contrast, restraint stress-induced adipose inflammation increased plasminogen activator inhibitor-1 and tissue factor in white adipose tissue but did not give rise to thrombus formation in mice [39]. In parallel with the above results, our findings indicated that pathological clot formation in patients with TBI was not mainly caused by alterations in the coagulation cascade but rather by the premature activation of platelets that in turn caused altered fibrin formation.
In our study, in comparison with patients without DVT, patients with DVT demonstrated decreased platelet numbers. This was in line with the postinjury thrombocytopenia in patients with TBI [76]. However, the MPV level, an index that reflects the platelet production rate and stimulation [77], dropped rapidly in patients without DVT but remained at the same level in patients with DVT throughout the period. In a critical care setting, MPV was associated with a longer length of stay, need for vasopressin support, and assisted mechanical ventilation in patients with sepsis [78]. This result indicated that following TBI, platelets underwent depletion as well as activation simultaneously in patients with DVT.

Effect of Insulin on Platelet Activity
Insulin has been reported to attenuate agonist-induced platelet activation [28]. A metabolic disturbance with glucose intolerance and/or a high level of insulin resistance was a prerequisite for platelet reactivity in patients with minor ischemic stroke or transient ischemic attack [18]. Obese adolescents with insulin resistance are associated with increased platelet activation in China [31]. In line with these reports, compared with patients without insulin therapy, we found that patients with insulin therapy also had progressively reduced platelet numbers (Fig. 6). The decrease in platelet count could be attributed to the consumption of platelets, as in fibrin clot formation and late consumptive depletion and exhaustion following platelet activation [65]. In contrast, MPV declined progressively in patients without insulin therapy, while it remained at a consistent level in patients with insulin therapy throughout the study period (Fig. 6). A positive correlation between MPV and HOMA-IR was found in patients with gestational diabetes mellitus [79] or in nondiabetic patients with slow coronary flow [80], both of whom had increased MPV [79,80]. Thus, in our patients with insulin therapy, the consistent MPV level, in contrast with the progressively increased glucose level, indicated that platelets were consistently activated in patients with insulin resistance. PDW describes the size variation of platelets. PDW levels were associated with injury severity and mortality in patients with severe TBI [81]. However, we did not find any difference in PDW between the patients with and without DVT.

Plasma Insulin Level Predicts DVT Onset
In patients with insulin resistance induced by TBI, the attenuation of agonist-induced platelet activation by insulin could be lost, resulting in the persistent activation of platelets. The subsequent platelet aggregation might accelerate DVT development. We did show that insulin levels on days 4-7 could independently predict DVT occurrence. We thus established an independent corroboration for the previous finding that the insulin infusion rate may be more closely associated with outcomes in patients with TBI [40]. Another possibility could be that insulin resistance is also associated with vascular stiffness [82][83][84]. It is reasonable to speculate that the accelerated vascular stiffness makes it easy for the deep vein to be injured in TBI, which is also attributed to DVT development.

Other Factors with DVT Onset
P-selectin is an integral membrane glycoprotein of platelets and endothelial cells [85], and the soluble form of P-selectin is hypothesized to play a role in the initiation of atherosclerosis and acute myocardial infarction [86]. However, we did not detect any difference in P-selectin levels between patients with DVT and patients without DVT. Moreover, it was intriguing to find that none of the blood count, inflammatory mediators, incidence of pneumonia or sepsis was distinguishable between patients with and without DVT or between the patients with and without insulin therapy. A previous study found that the premature activation of platelets could be the result of chronic inflammation in TBI [87]. TNF-ɑ is the key aging-associated proinflammatory cytokine responsible for platelet hyperreactivity and thrombosis [88].
Nonetheless, compared with patients without DVT, patients with DVT were more likely to receive lumbar cerebrospinal fluid drainage, which was performed in cases of cerebral infection, and had higher infection SOFA scores. It is possible that the patients with DVT had higher neuroinflammation than patients without DVT, and the disseminated intravascular coagulation in patients with DVT may lead to more severe multiple organ dysfunction.
Our finding (Table 3) was in keeping with a retrospective study showing that older age strongly predicted developing DVT in patients with moderate to severe TBI [2,6] and that older age also increased mortality in patients with TBI [89]. However, in line with a previous report [2], the mortality in both the neuro-ICU and 6 months were comparable between the patients with and without DVT, suggesting that developing DVT was not associated with excess mortality.

Limitation
There are limitations in the present study. First, we examined the insulin level in both the patients with and without insulin therapy. Exogenous insulin may confound the endogenous insulin level, which theoretically reflects insulin resistance. To avoid this, another project focusing on fasting insulin at admission is underway, in which insulin therapy has not been initiated. Second, we initiated chemoprophylaxis approximately 2 weeks after TBI, which may be too conservative. It was recently reported that early chemoprophylaxis was not associated with the progression of hemorrhage or the need for neurosurgical intervention in patients with stable brain CT scan findings at 7 h following TBI [90]. Third, we only recruited severely injured patients in the ICU. There was a doseresponse relationship between TBI severity and the degree of platelet dysfunction [44]. TBI with different severities and longer periods of monitoring of DVT is recommended.

Clinical significance
Despite intensive research, specific therapeutic interventions are still lacking for VTE prevention in patients with TBI [12]. Novel therapies can be more beneficial if directed against specific platelet responses, populations, interactions or priming conditions. We have shown for the first time that insulin resistance is closely associated with DVT occurrence. This indicates that increasing insulin sensitivity may prevent thrombus formation and may have a beneficial effect on outcomes after TBI. Furthermore, both the traditional laboratory hemostatic assays (e.g., INR and platelet counts) and the increasing use of viscoelastic assays (e.g., thromboelastogram) to evaluate homeostasis in patients with acute TBI [70] are time-consuming and costly. Our results support the use of low-cost biomarkers, e.g., HOMA-IR, in the prognosis of thrombosis in patients with TBI.

Conclusions
The plasma insulin level was progressively increased in patients with DVT relative to patients without DVT, accompanied by consistent activation and depletion of platelets. The insulin level on days 4-7 after TBI independently predicted DVT in the ICU.