FormalPara Take-home message

Patients with traumatic brain injury admitted to intensive care units are older and often less severe than in previous studies. Substantial between-center variation exists in ICU admission and treatment policies across Europe.


Traumatic brain injury (TBI) causes a social and economic global burden with about 82,000 deaths in Europe every year [1]. Patients with severe TBI often receive a highly intensive and multidisciplinary approach to prevent or mitigate both secondary brain injury and systemic complications [2]. For less severe TBI cases (without severe extracranial injury), clinicians have to estimate whether they will benefit from ICU admission, since guidelines with high-level evidence on ICU admission criteria are lacking. ICU admission is costly, and might also potentially be inappropriate for the patient, with risk of overtreatment and ICU-related complications, such as infections from multi-resistant bacteria [3].

In previous studies, intensive care admission was described merely for the most severe TBI cases, typically young male victims of high-energy road traffic incidents. In high-income countries, however, the aging population and the reduction of road traffic incidents have led to important changes in TBI epidemiology, which now includes older patients, who are often victims of falls, and present with frequent co-morbidities but less severe brain injury. Recent data suggest that the landscape of TBI in Europe is changing and that, correspondingly, ICU admission policies may have been modified, including a larger proportion of milder TBI patients [4, 5].

The aims of this study were:

  1. 1.

    to provide a general description of ICU stay, selected management aspects and outcome in TBI patients across Europe and,

  2. 2.

    to quantify variation across centers.



The Collaborative European NeuroTrauma Effectiveness Research in Traumatic Brain Injury (CENTER-TBI study, registered at NCT02210221), entails a longitudinal prospective collection of TBI patient data across 63 centers in Europe and Israel between December 19, 2014 and December 17, 2017. Inclusion criteria were: (1) clinical diagnosis of TBI; (2) indication for a brain CT scan; and (3) presentation to the hospital within 24 h post-injury. The presence of a severe preexisting neurological disorder, potentially confounding outcome assessment, was the only exclusion criterion. The CENTER-TBI study was approved by the medical ethics committees of all participating centers and informed consent from the patient or legal representative was obtained according to local regulations [4, 6].

ICU population and data collection

All patients directly admitted from the Emergency Room or transferred within 24 h of injury from another hospital to the ICU were analyzed [4]. Patients who deteriorated at the trauma, neurological or neurosurgical ward and were (re)admitted to the ICU were not included. Clinical data were collected at ICU admission, during ICU stay and at ICU discharge. For the current study, we extracted data on demographics, injury, imaging, admission, monitoring, treatment, and outcome characteristics. Patients were stratified using baseline GCS scores as mild (GCS 13–15), moderate (GCS 9–12), or severe TBI (GCS < 9) [4].

ICP and ICP-lowering treatments

Intracranial pressure and cerebral perfusion pressure (CPP) values were collected every 2 h. Intracranial hypertension was defined as a value above 20 mmHg, while 60 mmHg was chosen as a threshold for low CPP. To quantify the intensity of ICP-targeted therapies, a recently updated and validated version of the therapy intensity level (TIL) scale was used [7]. This scale summarizes in a score the number and the intensity of treatments. In addition, we analyzed the use of aggressive treatments for raised ICP as hypothermia, intense hypocapnia, barbiturates and decompressive craniectomy.


Outcome was measured at 6 months after injury using the Glasgow Outcome Scale-Extended (GOSE), administered by interview or postal questionnaire. The categories ‘vegetative state (GOSE 2)’ and ‘lower severe disability (GOSE 3)’ were combined, resulting in a seven-point ordinal scale.

Statistical analysis

Patient characteristics are described as mean and standard deviation (SD) or as median and interquartile range (IQR). We defined three groups: early deaths (died within ≤ 72 h of ICU admission), short stay (≤ 72 h in the ICU) and long stay (> 72 h in the ICU). Patient characteristics, treatments and outcome were compared between these groups with χ2 tests for categorical variables, and ANOVA and t tests for continuous variables. We used the IMPACT Core model to calculate expected mortality and proportion with unfavorable outcome (GOSE < 5).

The variation between centers was quantified using random-effect logistic and ordinal regression models with a random intercept for center, and expressed as the median odds ratio [8] for:

  1. 1.

    The proportion of patients with a short stay (≤ 72 h in the ICU) versus long stay (> 72 h) and early deaths (≤ 72 h).

  2. 2.

    The proportion of cases having received ICP monitoring. Also, a sensitivity analysis of the proportion of cases having received ICP monitoring in a subset of patients with a GCS < 8 and CT abnormalities was performed.

  3. 3.

    The use of aggressive ICP-lowering treatments (any use of decompressive craniectomy, metabolic suppression, hypothermia therapy or intensive hypocapnia).

  4. 4.

    6-month GOSE outcome.

The MOR is a measure of variation in treatments or outcomes between hospitals that is not explained by factors in the model or attributable to chance. The MOR is related to τ2, which is the variance of the random effects:

$$ {\text{MOR}} = \exp \left[ {\sqrt {2 \times \tau^{2} \times 0.6745} } \right] \approx \exp \left( {0.95\tau } \right). $$

The MOR can be interpreted as the odds ratio for comparing two randomly selected centers. For example, a MOR equal to one indicates no differences between centers. If there is considerable between-center variation, the MOR will be large. For example, a MOR of 2 for a certain treatment indicates that if two TBI patients with the same injury severity and characteristics presented to two random centers in our sample, one patient will have an over twofold probability to receive that treatment. To adjust for differences in baseline risk, we included the variables from the International Mission for Prognosis and Analysis of Clinical Trials in TBI (IMPACT) lab prognostic model [9] and any major extracranial injury [defined as an Abbreviated Injury Scale (AIS) > 3] [10]. The likelihood ratio test was used to determine the significance of the between-center variation, comparing a model with and without a random effect for center. The corresponding p values require a mixture distribution since the null hypothesis is on the boundary of the parameter space) [11].

Statistical analyses were performed in the R statistical software [12]. Multiple imputation was used to handle missing values, with use of the mice package in R [13]. These analyses were based on Version 2.0 of the CENTER-TBI core dataset, accessed using a bespoke data management tool, ‘Neurobot’ (; RRID: SCR_01700).


Patient characteristics

A total of 4509 patients were enrolled in the CENTER-TBI study, 2138 of whom were admitted to the ICU and included in this study. Patients were mostly men (73%). The median age was 49 years (IQR 29–65). A minority were children younger than 18 years (132, 6%), 552 (26%) were older than 65 years and 94 (4%) older than 80 years. Patients with severe TBI constituted 48% of the ICU admissions, while 720 cases (36%) were classified as mild. Major extra-cranial injuries were present in 1174 (55%) patients. (Table 1). More than half of the 54 ICUs have a neuro-ICU available (35, 65%). The median number of ICU beds available was 35 (28–45). Thirty-eight ICUs had a step-down unit available (70%). (Table S1) The median number of ICU patients recruited was 28 with an IQR of 15–50 (range 1–140). The median length of stay for the entire ICU cohort was 11 (IQR 3–26) days.

Table 1 Baseline characteristics

ICU mortality and discharge rates were high in the first 72 h, but declined over time (Figs. 1, 2 ). There were 128 (6%) early deaths, 636 (30%) short-stay, and 1372 (64%) long-stay cases (Fig. 2).

Fig. 1
figure 1

ICU patient flow over time. a Plot of the dynamic states of patients with TBI that were admitted to the ICU during the first seven days after ICU admission. The y-axis represents the probability to be in one of the possible states (i.e., alive or dead or discharged from ICU) at each time point from ICU admission. *Died after ICU discharge. b Plot of the dynamic states of patients with TBI that were admitted to the ICU during the first 6 months after ICU admission. The y-axis represents the probability to be in one of the possible states (i.e., alive or dead or discharged from ICU) at each point from ICU admission. *Still in ICU

Fig. 2
figure 2

Flowchart of ICU patients. This figure shows the flow of patients at the ICU, based on their length of stay. *Patients who died within 72 h at the ICU

Early death patients had a higher median age (62 years) and more severe injuries, both intracranial and extracranial, compared to survivors. Demographic features were comparable between short-stay and long-stay groups, while significant differences were identified with respect to injury severity, CT findings, and pre-admission insults (Table 1). The main cause of mortality in early death patients was due to initial head injury (78, 81%) (Fig. S2).

The most frequent reasons for admission in short-stay patients were the need for frequent neurological observations (340; 54%) and mechanical ventilation (154; 24%) (Fig. S3). The long-stay patients included 319 patients (25%) classified as mild TBI in whom similar reasons for admission were mentioned (the need for neurological observations (152, 48%), mechanical ventilation (96, 30%).

Monitoring and treatment

Mechanical ventilation for at least 24 h was most often applied in long-stay patients and in patients who died early, when compared to short-stay patients (1164 [85%] and 91 [71%]; versus 201 [32%], respectively). A large difference was found in the use of ICP monitoring between long-stay and short-stay cases (837; 62% versus: 41; 7%, respectively). The main indication for ICP monitoring in short-stay patients was surveillance after intracranial operation (31, 76%). Invasive blood pressure monitoring was used in the majority of long-stay patients (1227; 90%) and in early deaths (113; 89%); but less frequently (388; 62%) in short-stay patients (Table S2).

Both neurosurgical interventions and extracranial surgery were more common in long-stay patients (634; 47% and 467; 34%, respectively) when compared to short-stay patients (139; 22% and 122; 19%, respectively). Patients in the short stay group rarely (≤ 5%) received aggressive ICP treatments (i.e., decompressive craniectomy, metabolic suppression, hypothermia, or intensive hypocapnia) (Table S2).

Complications and outcome

Long-stay patients suffered more complications compared with short-stay patients: most commonly ventilator acquired pneumonia (276; 21% versus 3; 0.5%) and cardiovascular complications (125; 9.3% versus 9; 1.5). The overall median hospital length of stay was 11 days (IQR: 3.4–26), while the median hospital length of stay for long-stay patients was 18 days (IQR: 7.7–35). When compared to long-stay patients, short-stay patients were less often discharged to a step-down unit (86 [14%] vs 255 [21%], respectively), and more often transferred to the ward (486 [78%] versus 616 [51%]). Long-stay patients were also often discharged to other hospitals (174; 14%) and rehabilitation units (95; 8%); while, other discharge locations (such as home, other ICU, or nursing home) were rare (Table 2).

Table 2 Outcome and complications

In-hospital mortality for the ICU stratum was 15%; and at 6 months, mortality rose to 21% (data available for 1846 cases), which was lower than the expected mortality based on the IMPACT model (30%). Six-month mortality was higher in the long-stay-patient group compared with the short-stay group (20% versus 5.5%) (Fig. 3, Table 2).

Fig. 3
figure 3

Six-month Glasgow outcome scale extended. This figure shows the distribution of the functional outcomes at the GOSE after 6 months for all ICU patients, short-stay patients, and long-stay patients

An unfavorable outcome at 6 months (GOSE < 5) was observed in 43% in the total ICU stratum, 50% (590) in long-stay group, and in 15% in short-stay group (77). The unfavorable outcome rate in the total ICU stratum was similar to the expected rate based on the IMPACT model (49%).

Between-center differences

Substantial between-center differences were found in the proportion of short stay, long stay and early deaths (MOR: 2.3, p < 0.001, Fig. 4). When adjusted for case-mix and random variation, between-center variation in the proportions of patients in the short-stay versus long-stay and early death groups was still substantial (MOR: 2.3, p < 0.001).

Fig. 4
figure 4

Between-center differences in ICU policies and outcome. This panel shows the adjusted differences (adjusted for case-mix with the IMPACT prognostic model) between centers by considering. a The proportion of patients with a short stay (≤ 72 h in the ICU) versus long stay (> 72 h) and early deaths (≤ 72 h); long stay and early deaths were treated as one group, since they resemble more severe patients and we aimed to study the proportion in each center of short-stay patients that were discharged alive within 72 h. b GOSE at 6 months for total ICU population. c ICP monitoring. d Aggressive therapy (any use of decompressive craniectomy, metabolic suppression, hypothermia therapy or intensive hypocapnia during ICU stay). A random-effect regression model was used to correct for random variation and adjusted for case-mix severity using the IMPACT variables and the presence of any major extracranial injury. The MOR reflects the between-center variation; a MOR equal to 1 represents no variation, the larger the MOR, the larger the variation. Significant differences (p value < 0.001) are present for data shown in a, c, and d for b (p = 0.01). GOSE Glasgow Outcome Scale extended, ICP intracranial pressure, MOR median odds ratio

Regarding ICP monitoring, after adjustment for case-mix, substantial and significant between-center variation persisted in the use of ICP monitoring (MOR: 2.5, p < 0.001, Fig. 4). A sensitivity analysis (with a subset of patient with a GCS ≤ 8 and CT abnormalities) confirmed this between-center variation (MOR: 2.6, p < 0.001). After case-mix adjustment, significant between-center differences were also found in the use of aggressive therapies (MOR: 2.9, p < 0.001, Fig. 4).

Between-center variation in outcome was smaller compared to the variation in treatment. The MOR in the total ICU population for six-month GOSE was 1.2 (p = 0.01, Fig. 4).


The aims of this study were to describe ICU admission policies, selected management aspects, and outcome in TBI patients across Europe both at the patient and center levels. A substantial proportion of patients admitted to the ICU were classified on presentation as having a mild or moderate TBI. This is in strong contrast with historical TBI series, such as the USA Traumatic Coma Data Bank study [14] and other studies [15]. However, those series included only severe TBI patients, so that any evaluation of the general ICU admission policies at that time for milder cases is impossible. A more recent study, which analyzed data from 1648 mild TBI patients in 11 US level I trauma centers, showed that about 24 percent of them required admission to the ICU at some stage [16].

Even when compared to these latter data, our findings indicate quite liberal ICU admission rates for less severe cases. This is consistent with the strategies declared by the majority of centers participating in CENTER TBI. When centers were asked (in the Provider Profiling survey; see [5]) if they would admit “patients with a Glasgow Come Score (GCS) between 13 and 15 without CT abnormalities but with other risk factors”, 68% of responders reported this as consistent with their center policy.

Among the cases admitted, we looked at three different patient groups. Around 6% of patients died in the first 3 days after admission, with clearly severe intracranial and extra-cranial injuries. Patients in this group were significantly older, and only approximately half of those with documented intracranial mass lesions in this group received an operation. In survivors, we studied two distinct groups: those with a brief transition through the ICU and the second characterized by a prolonged ICU treatment. We selected the first 72 h as criterion to separate these two patient streams, triggered by the high ICU discharge rate during the first 3 days. This separation identified patients with different clinical characteristics, care pathways, and outcomes: long-stay patients were more severely injured, required more frequent invasive monitoring (including ICP) and therapies (both surgical and medical), and suffered a worse outcome. In contrast, short-stay patients were less severely injured, received less monitoring and treatments, and achieved better outcomes. The most frequently indicated reasons for ICU admission in this latter group were the need for strict neurological observation and mechanical ventilation (which, however, was continued for at least 24 h only in a third of cases). This may reflect current policy of early intubation at the scene of accident, and/or during initial assessment and evaluation. Cranial and extra-cranial surgery could also have been alternative indications for a short period of intense post-operative observation in the ICU.

These data can be interpreted in one of the two ways. On the one hand, the observed practice may represent a prudent strategy, offering close surveillance and assistance to patients at relatively low risk, but with the opportunity to ensure consistently good outcomes. The risk of deterioration in mild TBI is low but non-negligible. A recent meta-analysis, including 45 studies (for a total of 65,724 patients), estimated a 12% incidence of neurological deterioration and 3.5% neurosurgical intervention in mild TBI (characterized as GCS 13–15) [17]. Alternatively, the observed admission strategies may represent costly over-triage, because the ICU is an expensive resource, which should be used wisely. The fact that 11 patients in the short-stay group were discharged home directly from the ICU raises strong reservations on their need for intensive care. A previous study in mild TBI patients in the ICU in the USA showed that 17% of cases were over-triaged, with over triaged patients defined as “ICU stay ≤ 1 day; hospital stay ≤ 2 days; no intubation; no neurosurgery; and discharged to home” [18]. Our data on ICU admission of mild TBI patients are partially concordant with these findings, and while they do not permit accurate cost–benefit analysis, they clearly indicate a trend in ICU admission policies that deserves attention.

After adjustment for case-mix and random variation between centers, we found significant between-center proportion of short-stay patients discharged alive within 72 h. This confirms the results of earlier studies that found large variation in admission and discharge policies, primarily for mild TBI patients [5, 18]. This variation might reflect various factors: a search towards more individualized management [2], a lower adherence to guidelines [19], different availability of resources, or various combinations of these different factors. As for monitoring and management variations among centers, heterogeneity was not unexpected: previous studies [19,20,21] and surveys [22,23,24] found profound dissimilarities between centers in monitoring and treatment policies similar to our study.

The MOR for outcome between centers (1.2) was significant (p = 0.01), but smaller than the MOR for case-mix, ICP monitoring and aggressive therapies (2.5–2.9). This may reflect the small proportion of outcome variance modifiable by differences in management, and/or that differences in individual aspects of management may be discordant and make any outcome impact less easily detectable. Further, between-center variations in outcome that we demonstrated were smaller than previously reported [25, 26]. This may be because previous analyses were based on older data, collected across multiple studies, and heterogeneity in time and location explained the larger outcome variance in these older reports. It is also possible that over time, a more homogeneous standard of treatment has evolved in Europe and Israel.

Strengths and limitations

The CENTER-TBI study is unique for its extensive data collection in multiple centers, enrolling TBI patients with varying injury severity across a wide range of European centers. Limitations include that we focused on the ICU while an individual patient’s fate, and policies of the center at which treatment is delivered, depends on the continuum of care (from pre-hospital to rehabilitation). Second, the centers differed in their ICU characteristics, which might potentially contribute to between-center differences in ICU stay, treatment and outcome. In addition, we might have missed some important case-mix variables in the models that might have contributed to differences between centers (instead of true differences in policies). Third, the low number and non-consecutive enrolment in some centers could result in non-representative recruitment with reference to local ICU admission policy and introduce selection bias. Finally, all centers participating in CENTER TBI are characterized by their commitment to TBI research. They might represent a selected sample of the neuro-trauma centers in Europe limiting generalizability.

Future directions

The observed between-center differences in ICU policies require further research on whether these differences impact patient outcome. Comparative effectiveness research (CER) can be used for this purpose [27], requiring adequate covariate adjustment to account for confounders, and adjustment for other treatment policies that might differ between the centers. Variation in ICU performance also provides opportunities for future benchmarking and quality initiatives.


Our results confirm that the current ICU patient population admitted with Traumatic Brain Injury across Europe has changed, compared to previous data, and now includes older patients and a substantial proportion of mild and moderate cases. Sub-populations of patients (which we defined as short-stay, long-stay, and early mortality groups), are clearly different in injury severities, indications for ICU admission, care pathways, ICU resource utilization, and outcome. Our per-center analysis identified differences in the proportion of short-stay patients and interventions, for instance in the use of ICP monitoring and aggressive therapy, while there were only small differences in outcome.