FormalPara Take-home message

The manipulation of arterial carbon dioxide levels (PaCO2) is easy, and hyperventilation (HV) has been a common ICP-lowering strategy for over half a century. However, hyperventilation-induced vasoconstriction is a double-edged sword. It reduces cerebral blood volume and intracranial volume, and therefore, lowers ICP

We observed huge variability among centers in PaCO2 values and use of HV. Although causal inferences cannot be drawn from these observational data, our results suggest that, in patients with severe intracranial hypertension, HV is not associated with worse long-term clinical outcome

Introduction

Changes in the arterial partial pressure of carbon dioxide (PaCO2), by modifying the extravascular pH, modulate cerebrovascular tone, and hence cerebral blood flow (CBF) and cerebral blood volume (CBV) [1, 2]. Hypercapnia results in perivascular acidosis, which causes cerebral vasodilation, and consequently, an increase in intracranial volume. In patients with poor intracranial compliance, this could raise intracranial pressure (ICP). On the other hand, hyperventilation (HV) induced alkalosis reduces vascular calibre, and hence CBV, and can represent an effective measure to control intracranial hypertension, when ICP remains elevated despite first-line therapies [3,4,5,6]. However, hypocapnic cerebral vasoconstriction can also reduce CBF [7], thus posing the risk of secondary ischaemic insults [8]. In a survey across European trauma centers, the most frequently reported PaCO2 target was 36–40 mmHg in the absence of intracranial hypertension, which was reduced to 30–35 mmHg when ICP was > 20 mmHg [9]. The most recent evidence-based guidelines on TBI management provide no definitive recommendations regarding target PaCO2 levels due to the low quality of evidence available on this issue [10, 11].

Consequently, although many patients with severe TBI undergo several days of mechanical ventilation, there is little evidence-based guidance on PaCO2 targets, and clinical practice remains highly variable. A recent consensus on mechanical ventilation in patients with acute brain injury suggested aiming for a physiologic range of PaCO2 between 35 and 45 mmHg [12], and to only use hyperventilation (with an undefined PaCO2 target) as a short-term therapeutic option in patients with evidence of brain herniation. However, the document was unable to provide a recommendation on the use of hyperventilation in patients who showed significant ICP elevation, but no evidence of herniation. A management algorithm for patients with intracranial hypertension, based on expert consensus, suggested the use of HV (PaCO2 32–35 mmHg) for controlling ICP only as a second-tier treatment, did not support lower PaCO2 levels and recommended against routine hyperventilation to PaCO2 below 30 mmHg [13].

The objectives of this study were to assess, in a real-world context, PaCO2 management and the lowest target of PaCO2 in a large cohort of mechanically ventilated TBI patients and practice variability between centres to evaluate the association between the use of profound HV and 6-month clinical outcomes.

Methods

Study design and patients

The Collaborative European NeuroTrauma Effectiveness in Research in Traumatic Brain Injury (CENTER-TBI study, registered at clinicaltrials.gov NCT02210221) is a longitudinal, prospective collection of data from TBI patients across 65 centers in Europe. The study was conducted between December 19th, 2014, and December 17th, 2017 and details regarding the design and the results of the screening and enrolment process have been previously described [14,15,16].

The CENTER-TBI study was approved by the Medical Ethics Committees in all participating centers, and informed consent was obtained according to local regulations (https://www.center-tbi.eu/project/ethical-approval). This project on PaCO2 management was preregistered on the CENTER-TBI proposal platform and approved by the CENTER-TBI proposal review committee before starting the analysis (ESM Document 1). This report complies with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines (ESM Table S1).

We included all patients in the CENTER-TBI Core study who had a TBI necessitating ICU admission, required tracheal intubation and mechanical ventilation, had at least two PaCO2 measurements in the first 7 days and had been admitted to a study centre that enrolled at least ten patients.

Data collection and definitions

Detailed information on data collection is available on the study website (https://www.center-tbi.eu/data/dictionary). For the first week in ICU, the daily lowest and highest PaCO2 values from arterial blood gases and, if an ICP device was inserted, the hourly ICP measures were used for analysis.

HV was defined as moderate for PaCO2 ranging between 30 and 35 mmHg and profound for PaCO2 < 30 mmHg [10, 13]. Therapy intensity level (TIL) was calculated according to the most recent TIL scale [17]. Patients with invasive ICP monitoring during the first week of ICU stay were classified as ICPm, while those who did not receive ICP monitoring during ICU stay as no-ICPm. Intracranial hypertension was defined as ICP > 20 mmHg.

Objectives

The aims of this study are:

  1. 1.

    to describe the PaCO2 values in the first week from ICU admission in mechanically ventilated TBI patients, and to evaluate practice variability across centers, particularly focusing on the lowest targets of PaCO2;

  2. 2.

    to assess at a center level the PaCO2 management in patients with/without ICP monitoring and with/without intracranial hypertension;

  3. 3.

    to evaluate the association between patient outcomes and center propensity to use profound HV.

Outcomes

Mortality and functional outcome (measured as the Extended Glasgow Outcome Score, GOSE) were assessed at 6 months. All responses were obtained by study personnel from patients or from a proxy (where impaired cognitive capacity prevented patient interview), during a face-to-face visit, by telephone interview, or by postal questionnaire around 6 months after injury [18]. All evaluators had received training in the use of the GOSE. An unfavourable outcome was defined as GOSE ≤ 4, which includes both poor functional outcome and mortality.

Statistical methods

Patient characteristics were described by means (± standard deviation, SD), medians (I–III quartiles, Q1–Q3) and counts or proportions, as appropriate. The comparison of baseline features according to ICP monitoring was performed using Mann–Whitney U test, t test and Chi-square test as appropriate. We used the median odds ratio (MOR) to estimate the between-centre heterogeneity in targeting a PaCO2 of 35–45 mmHg. MOR was obtained from a longitudinal logistic mixed-effect model on daily lowest PaCO2 adjusted for the IMPACT core covariates [19], ICP monitoring, and daily evidence of elevated ICP (at least one ICP > 20 mmHg during the day); and with a hierarchical random intercept effect’s structure (i.e., patients within centers). The same model architecture was used to quantify between-centres heterogeneity in the use of profound HV.

We resorted to an instrumental variable approach to evaluate the association between HV and 6-month outcomes, trying to minimize the potential measured and unmeasured confounding acting in this complex observational study [20]. This was done by considering the propensity of centres to apply profound HV, measured as the proportion of daily lowest PaCO2 < 30 mmHg, as an instrument in the logistic regression model with a random intercept for centers. This model was adjusted for some subject-specific covariates that included IMPACT core covariates at baseline, ICP monitoring and dose of intracranial hypertension, calculated as the area under the ICP profile above 20 mmHg, named AUC ICP > 20[21]. The assumptions underlying the IV approach were assessed (ESM-Statistical methods).

Tests were performed with a two-sided significance level of 5%. All analyses were conducted using R statistical software (version 4.03).

Results

Of the 4509 patients included in the CENTER-TBI dataset, 2138 patients with TBI from 51 centers in Europe were admitted to ICU. Among these, 1176 required mechanical ventilation and had at least two PaCO2 measurements within the first 7 days from ICU admission. Excluding the centres that enrolled less than ten patients, 1100 patients from 36 centers were available for the analysis (ESM Fig. 1). During the first week of ICU admission, a total of 11,791 measurements of PaCO2 were available (5931 lowest and 5860 highest daily values).

Fig. 1
figure 1

(a) Distributions of the daily lowest PaCO2 recorded in the first 7 days of ICU in each participating centre (coloured by country) and overall (grey area). These distributions were estimated by a Gaussian kernel density. (b) Centre-specific mean values (coloured by country) of daily lowest PaCO2 with the corresponding 95% confidence intervals. The solid vertical line represents the overall mean of daily lowest PaCO2 values, and the size of the dots is proportional to the number of patients in the centre. PaCO2 the partial pressure of carbon dioxide, AT Austria, BE Belgium, DE Germany, ES Spain, FI Finland, FR France, HU Hungary, IT Italy, LT Lithuania, NL Netherlands, NO Norway, SE Serbia, UK United Kingdom

Full size image

Patient characteristics

Patient characteristics at hospital admission in the overall population and stratified according to the presence (n = 751) or not (n = 349) of ICP monitoring, are summarized in Table 1. The median age was 48 years (Q1–Q3 = 29–64), and most patients were male (74%). 64.7% of patients presented with a severe TBI (Glasgow Coma Scale, GCS ≤ 8) and 12.5% of cases were complicated by thoracic trauma. In 727 (97%) ICPm patients, ICP was inserted by the second day of ICU admission.

Table 1 Baseline demographic and clinical characteristics, including trauma characteristics, clinical presentation, and neuroimaging features at ICU admission in the overall population and stratified according to the presence or not of ICP monitoring
Full size table

In the overall population, the mean PaCO2 at ICU admission was 39.1 (± 6) mmHg, and the no-ICPm group had higher PaCO2 mean values compared to the ICPm patients (39.9 ± 6.8 vs 38.7 ± 5.6 mmHg, p < 0.002).

Lowest PaCO2 targets according to centers

Daily minimum PaCO2 distribution during the first week for the whole population, and separated by the centre, are presented in Fig. 1a. The overall mean lowest PaCO2 was 35.2 ± 5.4 mmHg with substantial heterogeneity between centres, whose means ranged from 32.3 (± 3.7) to 38.7 mmHg (± 5.9). This result seems to be related more to different management strategies at the centre level, rather than reflecting national policies (Fig. 1b). For example, among the UK centers (in yellow), two centers had a mean PaCO2 value of 32.3 and 36.4 mmHg.

Only 144 (13%) patients had all PaCO2 measurements between 35 and 45 mmHg, while 588 (53%) patients had at least half of the total PaCO2 measurements in this range. Using MOR to quantify between-centre differences in targeting the suggested PaCO2 range of 35–45 mmHg, we found that, after correction for patient and trauma characteristics, there was a 1.72-fold difference in the odds of having a PaCO2 range of 35–45 mmHg between centres with the highest and lowest rates. After excluding 390 patients with intracranial hypertension, the percentage of patients with all and at least half of the total PaCO2 measurements between 35 and 45 mmHg raised to 19% (111/593) and 64% (380/593), while MOR decreased to 1.4.

Lowest PaCO2 targets in the presence or not of ICP monitoring

Mean minimum PaCO2 values were significantly lower in ICPm patients compared to no-ICPm (34.7 ± 4.9 mmHg vs 36.8 ± 5.7 mmHg, p < 0.001). Large variability was observed among centers in the management of PaCO2 targets in both subgroups (Fig. 2 and ESM Fig. 2). Some centres showed no differences in target PaCO2 when ICPm was used (i.e. data points near the line of identity in Fig. 2a), but most hospitals tended to adopt lower PaCO2 targets when ICP was monitored (i.e. data points that deviate substantially from the line of identity in Fig. 2a). For example, three centers showed a reduction greater than 4 mmHg in the mean daily lowest PaCO2 when ICP monitoring was available (from 38–38.4 mmHg to 33.1–34.2 mmHg).

Fig. 2
figure 2

(a): Scatterplot of the mean daily lowest PaCO2 values in no-ICPm vs ICPm patients in each participating centre (coloured by country). The dashed line represents the line of identity, and a data point on or close to the line indicates that PaCO2 targets in that centre were not affected by the presence of ICP monitoring. The gradient of grey zones on either side of the grey area indicates increasing deviations from this line of identity between values in no-ICPm vs ICPm patients. Each gradation in shade representing one unit change (mmHg). The size of the dots is proportional to the number of ICPm patients at a centre. The outlier centre from Hungary included only two no-ICPm patients, out of a total of 12 patients, with only two measurements each before ending ventilation. (b) Mean of the daily lowest PaCO2 values in ICPm patients with no episodes of elevated ICP (ICP ≤ 20 mmHg) vs ICPm patients with at least one episode of elevated ICP (> 20 mmHg) in each participating centre (coloured by country). The dashed line represents the line of identity, and the size of the dot is proportional to the number of ICPm patients with elevated ICP. PaCO2 the partial pressure of carbon dioxide, AT Austria, BE Belgium, DE Germany, ES Spain, FI Finland, FR France, HU Hungary, IT Italy, LT Lithuania, NL Netherlands, NO Norway, SE Serbia, UK United Kingdom

Full size image

Lowest PaCO2 in the presence of intracranial hypertension

In the subgroup of patients with ICP monitoring, we also explored the attitude of centres in response to episodes of intracranial hypertension (n = 3646). Some centres showed no differences in target PaCO2 when ICP was elevated (i.e. data points near the line of identity in Fig. 2b), but most hospitals tended to adopt lower PaCO2 targets when ICP was monitored (i.e. data points that deviate substantially from the line of identity in Fig. 2b). The mean minimum PaCO2 was significantly lower in 398 patients with at least one episode of intracranial hypertension compared to the 240 who did not experience increased ICP (34.1 vs 35.6 mmHg, p < 0.001). Within the group of patients with ICP monitoring in place, significant inter-centre differences were observed in the mean lowest PaCO2, both in the absence and presence of intracranial hypertension (ESM Fig. 3).

Profound hyperventilation

An episode of profound HV (PaCO2 < 30 mmHg) was recorded on 727 occasions during the first week of ICU admission in 397 (36%) patients (57% had one, 22% two and 10% three occurrences). Results from the longitudinal mixed-effects model show notable heterogeneity between centres on the use of HV, even after adjusting for patient and trauma characteristics, with a MOR of 2.04 (Fig. 3, ESM Table 1). We found a significant positive association between the occurrence of increased ICP and the use of HV. Among ICPm patients, even after correction for covariates, the odds of HV in a day with elevated ICP was nearly three times that in a day with controlled ICP (OR = 4.34 95% CI = 4.25-4.44, p value < 0.0001 vs OR = 1.47 95% CI = 0.97-2.22, p value = 0.03167). Finally, HV was less applied from day 1 to 7 (OR of HV per day = 0.83; 95% CI = 0.82–0.84, p value < 0.0001).

Fig. 3
figure 3

Caterpillar plot of between-centre variation in using profound HV. The figure shows the predicted random intercepts for each centre, on the log-odds scale, along with their 95% prediction intervals. Higher values indicate a higher propensity to use profound HV. A longitudinal random effect logistic model was used to correct for random variation and adjusted for the core IMPACT covariates and elevated ICP. The MOR summarises the between-centre variation: a MOR = 1 indicates no variation, while the larger the MOR is, the larger the variation present. The median odds ratio (MOR = 2.04) refers to the odds of using profound HV between two randomly selected centres for patients with the same covariates and (comparable) random effects

Full size image

Neuromonitoring

Indirect CBF monitoring, using jugular bulb venous oxygen saturation or brain tissue oxygenation, was not used frequently. No differences were found in their use in patients receiving profoundly HV (jugular bulb venous oxygen saturation, SjvO2: 2.4% vs profound HV 3.5%, p value = 0.380; brain tissue oxygenation, PbtO2: 14.2% vs profound HV 13.9%, p value = 0.937). However, the use of profound HV was associated with significantly higher use of more aggressive treatment, expressed as mean TIL (9.7 vs 6.3 p value < 0.001). In particular, patients who received profound hyperventilation were more likely to have decompressive surgery (8.6 vs 4.8, p value < 0.001) and hyperosmolar therapy (low dose 12.7 vs 5.5, p value < 0.001; high dose 16.8 vs 5.7, p value < 0.001).

6 months mortality and neurological outcome

Overall, of the 1100 patient cohort, 165 died before ICU discharge (15%). Of the 970 patients for whom 6-month outcomes were available, 246 (25.4%) died, and 529 (54.5%) experienced unfavourable functional outcomes (GOSE ≤ 4). The 6 months mortality rate was 29% in patients who had at least one episode of profound HV and 23% in those who did not (p value = 0.045), while the rates of unfavourable GOSE were 64% vs 49% in the two groups, respectively (p value < 0.001). The percentage of patients who received profound HV in the first seven days from admission ranged from 1 to 30% between hospitals. In the IV analysis, the propensity to apply profound HV (defined by the use of PaCO2 < 30 mmHg) did not significantly increase mortality or unfavourable functional outcome, after adjusting for the dose of intracranial hypertension. Patients in hospitals that used 10% more profound HV had 1.06 higher odds of mortality compared to hospitals where profound HV was applied less often (95% CI = 0.77–1.45, p value = 0.7166) and the OR for the same comparison was 1.12 (95% CI = 0.90–1.38, p value = 0.3138) for an unfavourable functional outcome (Table 2).

Table 2 Results of the logistic mixed-effect model on 6-month outcomes by the instrumental variable approach with complete data (n = 919)
Full size table

Discussion

The current literature is inconclusive regarding the optimal ventilatory strategy to adopt in patients with TBI and, though there is increasing caution surrounding the use of HV, the translation of expert consensus recommendations into clinical practice remains uncertain. This study examined the PaCO2 management during mechanical ventilation at a centre level in prospectively collected observational data from a large multicentre cohort of TBI patients, focusing on the use of HV.

Our main findings are:

  • there is substantial practice variation among countries and centers regarding PaCO2 levels and the lowest PaCO2 adopted in TBI patients;

  • patients who received ICP monitoring were managed at lower PaCO2 compared to patients in whom such monitoring was not used;

  • patients who did receive ICP monitoring and experienced episodes of increased ICP were managed at lower PaCO2 levels than those who did not have ICP elevations; profound HV was commonly used in such patients;

  • we observed no association between the risk of mortality or unfavourable functional outcome and more frequent use of profound hyperventilation (PaCO2 < 30 mmHg).

Appropriate management of PaCO2 is a critical requirement in mechanically ventilated patients with TBI, since carbon dioxide is one of the major determinants of cerebral vascular physiology, and therefore cerebral blood flow and volume. The effect of the interplay between carbon dioxide and perfusion pressure on the cerebral circulation results in a sophisticated modulation of cerebrovascular resistance and tone, with hypercapnia causing cerebral vasodilation, and hypocapnia, vasoconstriction.

The only randomized controlled trial [22] addressing the benefit of prophylactic hyperventilation was conducted thirty years ago, and randomised TBI patients into three categories: control (n = 41), hyperventilation (n = 36), and HV + tromethamine (an H+ acceptor used to treat metabolic acidosis; n = 36). This setting is different from the current context, as the putatively normoventilated controls had PaCO2 values in the hypocapnic range (35 mmHg), and the HV utilized was profound (PaCO2 25 mmHg). These discordances with current practice, the limited number of patients, and the low incidence of episodes of intracranial hypertension make the results difficult to interpret.

A recent consensus still recommends targeting a normal range of PaCO2 values in the absence of increased ICP [12]. However, in the case of increased ICP, no agreement was achieved regarding the role of HV, providing evidence of the current uncertainty in this area [12]. Although induced hypocapnia is considered an efficient second line measure to reduce ICP, clinicians remain worried about potential cerebral ischemic complications of hyperventilation [8, 23]. Coles et al. used positron emission tomography in a cohort of 30 patients to show that the acute application of HV resulted in a reduction of cerebral blood flow and an increase in oxygen extraction fraction and the ischemic brain volume [23]. These results have left an indelible imprint on the way HV is perceived by intensivists, but they do not represent a randomized trial. Other authors suggest that mild HV may reduce ICP without leading to pathological changes of brain metabolism and oxygenation measured through cerebral microdialysis and PbtO2 [24] or energy failure. Moreover, Diringer et al. demonstrated that HV reduces global cerebral blood flow while increased oxygen extraction fraction leaving cerebral metabolic rate for oxygen unchanged, concluding that it is unlikely that HV causes neurological injury [25, 26].

Although some concerns still exist, PaCO2 reduction is still widely used in the clinical setting for ICP control. The most common PaCO2 target declared by clinicians in the absence of intracranial hypertension (35–40 mmHg) is higher than in the case of raised ICP (30–35 mmHg) [9]. Similarly, in a retrospective study of 151 patients with TBI, the PaCO2 target adopted in clinically stable ICP was 36 ± 5.7 mmHg, whereas in the case of increased ICP it was 34 ± 5.4 mmHg [27]. Besides, a recent consensus on ICP treatment suggested considering HV to PaCO2 of 30–32 mmHg when ICP is elevated in patients not responding to Tier 1 and 2 treatment [13].

Our data document a divergence between suggestions from literature and practice: nearly half of the daily lowest PaCO2 measurements in the first week were < 35 mmHg. Moreover, in presence of ICP monitoring, clinicians use a lower target of PaCO2. However, we also saw wide variability in PaCO2 levels between centres, both in terms of the overall values, and the lowest levels of PaCO2 observed. These differences were seen not just across the whole study cohort, but also in subgroups of patients with and without ICP monitoring, and those with and without episodes of intracranial hypertension in the first week. HV in presence of high ICP was frequently used, particularly in the first few days after admission, and was often combined with other ICP-lowering therapies such as osmotic agents and decompressive craniectomy. Interestingly, centres that used HV more frequently were not more likely to routinely apply more advanced neuromonitoring techniques for early detection of impaired cerebral blood flow and cerebral oxygen availability.

There is no strong evidence regarding the possible benefits or harms of profound HV on patient outcomes. However, a single retrospective analysis of 251 brain-injured patients [28] reported that, when compared to controls, patients who underwent prolonged HV (PaCO2: 25–30 mmHg; mean duration = 10, min–max = 5–41 h) experienced lower mortality (9.8 vs. 32.8%) but a higher rate of poor functional outcome.

We found that being treated in a centre where profound hypocapnia is more frequently used compared to centers where it is rarely used was not significantly associated with a higher rate of mortality or poor functional outcome.

In summary, our results suggest that moderate HV is widely used in severely brain-injured patients, especially when ICP is monitored, and in case of elevated ICP.

Limitations

Although our results may provide useful context with an important clinical message for physicians, we believe they should be interpreted with caution for several reasons. First, 6 months GOSE and mortality are influenced by several other factors, such as systemic and ICU complications, as well as post-ICU events. Therefore, based on observational data, it is speculative to draw a direct causal relationship between PaCO2 and outcome: further randomized controlled studies are needed to assess the effect of PaCO2 more precisely and in particular HV, on the outcome. Second, this is an analysis of data from a large study, which primarily addressed the epidemiology, clinical care and outcome of TBI. However, as respiratory management was not a primary focus of the study, more specific data on ventilatory management of these patients are missing, and hence unavailable to strengthen our analysis. Data on the incidence and timing of pulmonary complications such as acute respiratory distress respiratory syndrome and ventilator-associated pneumonia, the use of ventilatory strategies used to manipulate PaCO2, and the ventilator settings used in our study population are unavailable. Third, the outcome was evaluated at 6 months, which can be considered as an early measurement of outcome after TBI, and further long-term evaluations would have been desirable. Fourth, we did not specifically take into consideration the temperature management of the patients, which can importantly affect PaCO2 values. However, the measurements of PaCO2 are automatically corrected for temperature from the arterial blood gases machines, and we aimed to assess the targets of PaCO2 achieved, regardless of the effects of different factors on its final value.

Finally, in our dataset only the daily lowest and highest PaCO2 values were collected, thus missing possible changes in PaCO2 and pulmonary function parameters that may occur suddenly and repeatedly during the day. However, our analysis includes data on daily PaCO2, thus providing a longitudinal view of PaCO2 management over time.

Conclusions

In a large cohort of mechanically ventilated TBI patients, we found substantial between-centre variations in PaCO2, but with a large proportion of patients being managed at PaCO2 levels below those suggested by expert consensus statements. On average, patients who had ICP monitors in place had significantly lower PaCO2 levels than those that did not, and amongst ICP monitored patients, PaCO2 levels were lower in patients who had episodes of intracranial hypertension—suggesting that HV is still used for ICP management. Profound hyperventilation (PaCO2 < 30 mmHg) was not uncommon. However, a centre that had a greater propensity to use profound HV did not worsen 6-month mortality or functional outcome. Notwithstanding this, we believe that the available evidence still makes the case for caution in the use of HV, with careful consideration of risks and benefits on a case-by-case basis. Our data provide no basis for dismissing continuing concerns regarding prophylactic or profound hyperventilation. We need randomized controlled trials and high-level evidence guidelines to support rational choices regarding optimal ventilation management and PaCO2 targets in patients with TBI.