Prognostic value of coagulation tests for in-hospital mortality in patients with traumatic brain injury

Traumatic brain injury (TBI) is the leading cause of morbidity and disability in trauma patients and is responsible for a significant proportion of traumatic deaths in young adults [1, 2]. Coagulopathy is commonly observed after TBI [3‐5]. Although the incidence of coagulopathy is strongly associated with the severity of the injury, coagulopathy itself exerts an independent effect on mortality [4, 6, 7].

The reported incidence of TBI-associated coagulopathy ranges from 10 to 87.5% [8‐10]. The wide range of values reflects the lack of a standard definition for coagulopathy. Other factors contributing to the variability in incidence rates include differences in the patient populations evaluated, blood sampled at different time points, and the use of various coagulation assays. Moreover, studies investigating the same coagulation marker may use different cutoff values or sensitivity levels, thereby limiting generalizability. A recent meta-analysis of 22 studies found the overall incidence of TBI-associated coagulopathy to be 35.2% [11]. A previous study found that the presence of coagulopathy was associated with a nine-fold increase in the odds for mortality and increased the likelihood of a poor outcome by a factor of 36 [9]. Thus, it is clear that the development of coagulopathy after TBI is significantly associated with increased mortality and poor outcomes [12, 13].

Standard laboratory tests used to measure hemostasis and bleeding risk in patients with TBI include the international normalized ratio (INR), prothrombin time (PT), activated partial thromboplastin time (APTT), and platelet counts (PLT). D-dimer and fibrinogen (FIB) levels may provide additional useful data; however, their use is not routine. PT and APTT were originally developed to measure the in vitro activity of specific coagulation factors; however, they are currently used to predict the bleeding risk in perioperative neurosurgical patients [14]. The coagulation panel and PLT may also be used to predict the bleeding risk.

Some admission predictors such as age, absence of pupillary reactivity, the Glasgow Coma Scale (GCS) score, and CT characteristics have been routinely used to predict outcome in patients with TBI [15]. Although coagulation abnormalities may be a better predictor of mortality than the absence of the bilateral pupillary light reflex in some patients [16], prognosis is rarely predicted by coagulation status alone in the clinical setting. However, it is not known whether using the standard independent predictors in conjunction with coagulation tests would improve their prognostic value.

The aims of our study were two-fold: first, to determine the incidence of TBI-associated coagulopathy in patients with isolated TBI (iTBI) who attended an adult neurotrauma center; second, to evaluate the prognostic value of coagulation tests with respect to in-hospital mortality and to test their predictive power in prediction models for in-hospital mortality in patients with iTBI. Furthermore, we performed validation tests to assess the internal and external validity.

Overall, 18.6% (n = 432) of the patients in our study developed coagulopathy after iTBI. Coagulopathy developed in 30.4% of patients with severe iTBI and in 11.4% (n = 126) of patients with mild iTBI. The prevalence of acute traumatic coagulopathy is associated with the severity of the brain injury. We observed an INR >1.25 in 5.8% of patients, PT >14 s in 8.1%, APTT >36 s in 5.6%, PLT <100 × 10⁹/L in 10.7%, FIB level < 1.5 g/L in 15.3%, and D-dimer level > 5 mg/L in 22.1% of patients. These percentages were closely associated with the severity of brain injury, with significance detected among the three groups. Patients with severe TBI had a significantly higher median INR, PT, APTT, D-dimer level and lower PLT and FIB level than those with milder injuries (Table 1).

The patient characteristics and outcomes for the coagulopathy and non-coagulopathy groups are shown in Table 2. The proportions of patients with none pupillary reactivity, IPH, ICP monitoring and craniectomy were comparatively high in the coagulopathy group and low in non-coagulopathy group. The glucose and LOS were higher in the coagulopathy group, whereas the GCS at admission and Hb levels were lower in the coagulopathy group. The in-hospital mortality rate was significantly higher in the coagulopathy compared with the non-coagulopathy group.

The patient characteristics and outcomes for the model-development and validation groups are shown in Table 3. We found several significant between-group differences: the validation patients were older than those in the development group (mean age, 48.07 vs. 47.84 years, respectively), and the proportions of patients with bilateral pupillary reactivity, IPH, SAH, and a fractured skull were comparatively low in the development group and high in the validation patients. The proportions of those with diffuse axonal injury, PT >14 s and PLT <100 × 10⁹/L were high in the development compared with the validation group. The median glucose, HCT, and D-dimer levels were higher in the development patients, whereas the median INR, PT, and Hb levels were higher in the validation patients. The in-hospital mortality rate was not significantly different between groups.

The univariate analysis revealed that all predictors were statistically significant with respect to in-hospital mortality. A nonlinear relationship was observed between PLT and the coagulation tests; thus, each variable was rescaled (Fig. 1).

The results of the multivariable logistic regression analysis are shown in Table 4. Age, pupillary reactivity, GCS score, EDH, and glucose levels were independent prognostic factors for in-hospital mortality in model A. In model B, age, pupillary reactivity, GCS score, EDH, glucose levels, INR >1.25, and APTT >36 s were strong prognostic indicators of in-hospital mortality. Epidural hemorrhage detected by CT was a relatively favorable sign, whereas INR >1.25 and APTT >36 s were associated with higher in-hospital mortality.

We developed two prediction models for in-hospital mortality. The performance of each model is shown in Table 5. The discrimination for in-hospital mortality in the development group was good in both models (model A, c = 0.882 and model B, c = 0.893), and the H-L test revealed good calibration in both models (p > 0.05).

The internal validation test showed no over-optimism bias in the predictive c statistic of either model. The external validation test showed good discrimination for mortality in both predictive models (model A, c = 0.868 and model B, c = 0.875). However, calibration was better in model B than in model A (H-L test, p = 0.152 vs. p = 0.046, respectively). Thus, model B was generalizable and predicted in-hospital mortality in new patients more accurately compared with model A. Calibration curves for the outcomes are shown in Fig. 2.

We examined the prognostic value of admission coagulation tests with regard to in-hospital mortality after iTBI and developed a series of prognostic models to predict the probability of in-hospital mortality.

Multivariate logistic regression analysis revealed that age, pupillary reactivity, GCS, EDH, glucose levels, INR >1.25, and APTT >36 s were independently associated with in-hospital mortality. These variables can be readily obtained on admission to a neurosurgical unit and are consistent with prior studies of prognostic predictors [15]. Both of our prediction models, which were based on admission predictors, had excellent discrimination and calibration in the development group. Good generalizability is essential for predicting outcomes in new patients; thus, we assessed the external validity of our prognostic models to assess their generalizability. External validation confirmed that the prediction model using a combination of standard predictors and coagulation tests had better and more accurate calibration than that of the model based on standard predictors alone and had good generalizability. Thus, the most important and novel finding of our study is that the addition of coagulation test results to a multivariate logistic regression analysis can improve the predictive power of the standard prognostic model for in-hospital mortality. To the best of our knowledge, our study is the first to demonstrate the feasibility of this combined approach to predict outcomes in patients with TBI.

The ability to predict outcomes is crucial for effective care of patients with TBI [18, 19]. Information provided to relatives should be based on solid clinical and scientific evidence, which will help them prepare for the future and facilitate their understanding of the risky and potentially painful interventions that TBI patients may be required to undergo. Predictive systems promote quality assurance by providing a means for assessing patient care that can be used to make comparisons across or within institutions [20, 21]. The clinical value of predictors in an outcome prediction model is determined by their reliability on assessment, the prevalence of abnormalities, and the strength of the prognostic effect (odds ratios). The coagulation tests we investigated are standardized among laboratories and, thus, are objective and reliable. The prevalence of abnormal values was substantial for the coagulation tests investigated. The strongest predictive effects were observed for INR and APTT. Multiple associations were observed among coagulation tests and between coagulation tests and clinical parameters; however, the prognostic effects remained substantial following adjusted analysis, suggesting that the coagulation tests are of considerable prognostic relevance in TBI.

We found that 18.6% of the study population developed coagulopathy after iTBI, and 30.4% of the patients with severe iTBI experienced coagulopathy. These findings are consistent with previous reports [9, 10]. A meta-analysis of 22 studies found an overall incidence of TBI-associated coagulopathy of 35.2%; however, the definition of coagulopathy and the patient populations varied among the included studies [11].

Previous studies have shown that the most consistent coagulation abnormality is PT [7, 22]. PT reflects the activation time of the extrinsic, or tissue factor, pathway based on the cascade model of hemostasis. Most previous investigations of TBI-associated coagulopathy focused on PT or INR abnormalities [23, 24]. The International Mission on Prognosis and Analysis of Clinical Trials in TBI (IMPACT) study found that PT prolongation on admission was present in 221 of 850 patients (26%) and was associated with a 64% increase in mortality risk [12]. APTT reflects the activation time of the intrinsic, or contact activation, pathway and is particularly sensitive to deficiencies in coagulation factors IX, XI, and VIII. Although affected less often than the PT, APTT is more highly correlated with poor outcome and mortality than are other markers of coagulation [25, 26]. Thrombocytopenia on admission is a complication of TBI in fewer than 10% of cases [12, 27, 28]. In our study, 10.7% of patients had a PLT <100 × 10⁹/L. Thus, coagulation tests may provide more useful information on mortality after TBI than do the standard admission variables.

Recognition of the importance of coagulopathy in TBI is increasing. The mechanisms underlying TBI-associated coagulopathy are not well understood, although massive release of tissue factor, altered protein C homeostasis, microparticle upregulation, and platelet hyperactivity have been shown to play prominent roles [5, 29]. Hypocoagulable and hypercoagulable phenotypes have been identified in patients after TBI; however, their clinical significance, pathophysiological mechanisms, and temporal relationships are not well understood. Routine coagulation tests, such as PT, APTT, and PLT, demonstrate poor sensitivity to the disturbances associated with TBI-related coagulopathy and do not explain the observed hypercoagulability.

Although our results clearly indicate that coagulation tests may play a significant role in prognostic models and calculators for patients with TBI, caution should be exercised in interpreting our data. First, although our sample size was relatively large, the time course of our study was relatively long and different levels of emergency may exist. Furthermore, the low rate of mortality among our patients may have exaggerated the predictive power of our models. A second limitation of our study is that although we demonstrated the potential prognostic power of coagulation tests used in combination with parameters obtained at admission, the technology and methodology we used to assess coagulation tests cannot be readily obtained at admission.

We believe our findings highlight the importance of including coagulation test results in state-of-the-art outcome prediction models and set the stage for using this approach in future large-scale clinical trials. Moreover, we believe our results pave the way for the development of tools that connect basic science and clinical research with clinical evidence-based decision making that will ultimately improve the care of patients with TBI.

Coagulopathy is commonly observed after TBI and is associated with the severity of brain injury. Coagulation tests can improve the predictive power of the standard model for in-hospital mortality after TBI.