Extubation Failure in Brain-injured Patients: Risk Factors and Development of a Prediction Score in a Preliminary Prospective Cohort Study

THE decision to extubate brain-injured patients with residual impaired consciousness holds a high degree of uncertainty of success and undesirability of incorrect prediction.^1,2  Risk factors of extubation failure are common in this setting: severity of initial critical illness, emergent, and often prehospital tracheal intubation that favors stridor, prolonged mechanical ventilation (MV), altered consciousness with impaired airway protective reflexes such as cough and deglutition, neuromuscular weakness or paralysis, and hypersecretion.³  In general critical care medicine, it is usually assumed that restored conscious behavior is a prerequisite to extubation.^4–6  While separation from MV is generally easily acquired in brain-injured patients without other comorbidity,⁷  extubation failure is frequent.⁸  Nevertheless, some patients with severe alteration of consciousness could be extubated with success, and burden may be associated with extubation delay.⁹  Few studies investigated predictive factors implicated in extubation failure with focalization on different components of the problem: ventilatory parameters or global neurologic evaluation without clear predictors.^1,8,10–13  For example, Glasgow Coma Scale (GCS), in which low values often appear as limitative factors to extubation, is difficult to evaluate in intubated patients and patients’ ability to be an indicator of precise neurologic function is limited.^14,15  As a consequence, no guideline exists in this specific population and practice variations are frequent between institutions.¹  To characterize risk factors associated with extubation failure in brain-injured patients, we conducted a prospective, monocentric, observational study with multiparametric assessment of demographic, neurologic, hemodynamic, and respiratory functions. The objective of the study was to develop and internally validate a simplified pragmatic score predictive of extubation failure in this category of patients. Some of the results of this study have been previously reported in the form of an abstract.^16,17 

Additional details are provided in the online Supplemental Digital Content (https://links.lww.com/ALN/B322). Clinical trial is registered with http://www.clinicaltrials.gov (NCT 02235376).

Protocol was approved by Regional Ethics’ Committee (Comité d’Ethique des Centres d’Investigation Clinique, Rhône-Alpes-Auvergne, Grenoble, France, IRB 5921) on June 19, 2013. Because of the observational design of this study, which consisted of routine care in the studied intensive care units (ICUs), the need for written consent was waived. An information letter concerning the study was given to the patient or a next of kin after recovery.

The study was performed in a 13-bed neuro-ICU and 2 general ICUs (17 and 15 beds, respectively) of a university hospital. All consecutives brain-injured adult patients with initial GCS less than or equal to 12 (before tracheal intubation), intubated for neurologic reason and ventilated for more than 48 h, were screened between June 2013 and February 2015 (18-month period). Patients with brain structural lesions (isolated traumatic brain injury, subarachnoid hemorrhage, supra- or infratentorial spontaneous intracerebral hematoma, supra- or infratentorial acute ischemic stroke, or hypoxic–ischemic encephalopathy due to cardiac arrest) eligible for extubation were included. Patients with spinal cord injury, status epilepticus, disorder of consciousness caused by alcohol or other intoxication, central nervous system infection, tracheostomy, autoextubation, and withdrawal of care due to ethical reason were not included. Follow-up of included patients was 6 months to assess Glasgow Outcome Scale.

Detailed protocols are provided in the online Supplemental Digital Content (https://links.lww.com/ALN/B322).

Medical, paramedical, and physiotherapist staff were aware of the study protocol, which consisted of routine care in the studied ICUs.

Since all three ICUs were affiliated with the Department of Perioperative Medicine of the University Hospital of Clermont-Ferrand, Clermont-Ferrand, France, with the same medical and paramedical leadership and stewardship, protocols related to brain-injured patients were the same. The weaning protocol followed European guidelines of weaning from MV for general ICU patients.⁴  Notably, it was assumed that no tracheostomy was performed before any extubation attempt, unless the patient failed more than three spontaneous breathing trials (SBT). After resolution of acute organ dysfunctions notably increased intracranial pressure and sedative drugs withdrawal, eligibility for a SBT was daily assessed. Patients were extubated when they succeeded SBT irrespective of their neurologic status and upper-airway function.⁹  At the end of a successful SBT, previous ventilatory parameters were resumed during clinical evaluation related to the study. Extubation and respiratory care followed regular guidelines and were provided by a respiratory therapist during daytime. No prophylactic noninvasive ventilation (NIV) was used. If needed, standard oxygen therapy was initiated after extubation without high-flow devices. Time between the end of a successful SBT and extubation did not last more than 1 h. Local standard of care prevented delayed extubation after a passed SBT. All SBT procedures were reviewed on electronic patient records by two investigators not in charge of patient care (J.M.C. and E.F.) in order to look for possible delayed extubations. Additionally, data from tracheostomized patients were also reviewed in order to verify they previously failed at least three SBTs.

Extubation failure was defined as the need for ventilatory support after extubation using tracheal intubation or NIV⁴  during ICU stay. Respiratory failure necessitating reventilation was defined as the occurrence of at least two signs among oxygen therapy greater than 9 L·min⁻¹ to maintain oxygen saturation measured by pulse oximetry greater than 90%, respiratory rate greater than 35 min⁻¹ with accessory respiratory muscles involvement, respiratory or cardiac arrest, major tracheal secretions with inadequate cough, Paco₂ greater than 50 mmHg with pH less than 7.35, heart rate greater than 120 min⁻¹, systolic blood pressure greater than 200 mmHg or less than 90 mmHg. Analyses of extubation failure before 48 h as classically defined⁴  and at any time during ICU stay were performed. Justifications of those timings are presented in the Discussion.

Numerous clinical and paraclinical data were collected before extubation (see Methods and Data Collection in Supplemental Digital Content, https://links.lww.com/ALN/B322, which extensively expose collected clinical and paraclinical data). Of note, neurologic assessment included GCS with one point for verbal (total score on 10 points due to inability to assess verbal component of the score with tracheal intubation),¹⁸  Full Outline of UnResponsiveness (FOUR) score with three components (eyes, motor, and brainstem reflexes)¹⁹  (respiration item systematically rated 1 with tracheal intubation and breathes above ventilator rate for every patient in our cohort since they sustain pressure support ventilation), and Coma Recovery Scale-Revised (CRS-R) with its six components (auditory, visual, motor, oromotor/verbal, communication, and arousal),²⁰  with specificity related to inability to vocalize due to the endotracheal tube (item 2 of the oromotor/verbal function scale [vocalization/oral movement] was validated if oral movement compatible with vocalization attempt was observed and item 3 [intelligible verbalization] was validated if one could recognize words on patient lips or if the patient was able to write words). Data collection before extubation and follow-up were exclusively done by four senior intensivists working in the three ICUs (R.C., T.G., S.K., and J.M.) in a specifically designed and standardized case report form.

It seemed difficult to propose a sample size estimation according to literature in order to develop and validate a simplified pragmatic score predictive of extubation failure in this category of patients. Numerous rules of thumb have been suggested for determining the minimum number of subjects required to conduct multiple regression analyses, but they are heterogeneous and are often with minimal empirical evidence. For multiple regression models, some authors suggested variable ratios of 15:1 or 30:1 when generalization was critical.^21–24  Considering these works and expected extubation failure rate between 20 and 30%, we proposed to include at least 120 subjects to highlight three to five predictive factors. All analyses were performed using Stata software (version 13; StataCorp, USA) and done for a two-sided type I error of α = 5%. Patients’ characteristics were described by numbers and percentages for categorical parameters. For quantitative values, mean and SDs or median with interquartile range were calculated and presented according to statistical distribution (normal distribution of quantitative values was checked by Shapiro–Wilk test). Categorical data were compared using chi-square test. Quantitative data were compared between independent groups (extubation success/failure) using Student’s t test or Mann–Whitney U test when assumptions of t test were not met (normality studied using Shapiro–Wilk test and homoscedasticity using Fisher–Snedecor test). A multivariate analysis was performed using logistic regression models by stepwise approach according to univariate results (P < 0.10)^25,26  and clinical relevance.^27,28  Results were expressed with odds ratios (OR) and 95% CI. The final model was validated by a two-step bootstrapping process. For each step, bootstrap samples with replacements (n = 1,000) were generated from the training set. In the first phase, the percentage of models including each initial variable was determined by usual stepwise approach. Then, in the second phase, parameters of generalized linear regression (logistic for dichotomous-dependent variable) of the final model were independently estimated. The bootstrap estimates associated with each covariate regression coefficient, and their associated standard errors (SEs), were finally averaged from replicates. Log-likelihood measured the goodness-of-fit of a model. After these multivariate analyses, a receiver operating characteristic (ROC) curve was plotted for the final model, and area under the curve (AUC) was estimated.²⁹  A score predicting the extubation failure was estimated according to OR values. The threshold value of this score was determined according to usual recommendations by estimating several indexes as Youden, Liu, and efficiency. Sensitivity, specificity, and negative/positive predictive values were presented with 95% CI. A sensitivity analysis was performed to study patterns of patients with missing data and considered after analyses as not missing at random. An analysis of extubation failure before 48 h was also performed. Our study conforms to the recent set of reporting guidelines: Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis.³⁰ 

See the Supplemental Digital Content (https://links.lww.com/ALN/B322) for more information. One hundred and forty patients eligible to extubation were included between June 2013 and February 2015 (fig. 1). Extubation failure occurred in 43 (31%) patients (31 patients [24%] before 48 h). Data of extubation failure before 48 h are presented in online supplementary material (Tables E1 and E2, Supplemental Digital Content, https://links.lww.com/ALN/B322, presenting data about patients with extubation failure before 48 h). Further presented data correspond to extubation failure at any time during ICU stay. Missing data were investigated. Among the failure group, 361 of 4,730 (7.6%) and among the success group, 915 of 10,670 (8.6%) data were missing (P= 0.45). Total percentage of missing data was 8.3%. These aspects had no impact on results. No data were missing for the primary outcome.

There was no difference related to demographic data, general and neurologic initial severity scores, type of neurologic injury, pupillary status, brainstem reflexes, comorbidity, or characteristic of tracheal intubation between success and failure groups. More patients had alcohol abuse in the failure group (table 1). Characteristics of patients on successful SBT day are presented in Table E3 (Supplemental Digital Content, https://links.lww.com/ALN/B322, presenting characteristics of patients on successful SBT day). No difference between characteristics of patients was observed on successful SBT day. Intercurrent events (neurologic, respiratory, or hemodynamic) and notably intercurrent pneumonia or adult respiratory distress syndrome had no effect on extubation outcome. Duration of MV, number of failed SBT, and arterial blood gases had no impact.

Patients’ outcomes are presented in table 2. ICU mortality and length of stay were increased in the failure group. There was no difference in hospital length of stay. Glasgow Outcome Scale was significantly higher (meaning better recovery) in the extubation success group at ICU discharge and at 6 months.

Causes of extubation failure are presented in Table E4 (Supplemental Digital Content, https://links.lww.com/ALN/B322, presenting causes of extubation failure). Hypersecretion was the main reported reason accounting for 67% of extubation failure. Stridor was the cause in 14%. No patient was reintubated due to any acute neurologic complication.

Extubation failure rates associated with total GCS and eye and motor subscales are presented in Figure E1 (Supplemental Digital Content, https://links.lww.com/ALN/B322, presenting extubation failure rates associated with total and eye and motor subscales of GCS). Some patients with GCS as low as 3 could be extubated with success.

In univariate analysis, as shown in table 3, assessment of ocular functions in FOUR and CRS-R scores significantly differentiated success and failure. None of the “motor” responses was significant irrespective to scores. “Communication” and “oromotor” responses from CRS-R did not appear discriminative. Brainstem and arousal capabilities assessed by FOUR, GCS, and CRS-R were associated with extubation failure. Agitation and pain assessed by Richmond Agitation and Sedation Scale and Behavioral Pain Scale, respectively, did not accurately predict extubation outcome. Confusion of patients, as assessed by confusion assessment method for the intensive care unit, was significantly associated with extubation failure. Classical respiratory and general parameters like respiratory rate, rapid shallow breathing index, weight variation, and heart rate were not significant. Assessment of airway management criteria, illustrated by the capability to cough, the deglutition ability, and the gag reflex, were strongly associated with extubation failure when absent.

In multivariate analyses, GCS and FOUR as total scores or as their different components, as well as alcohol consumption history, were not significant. CRS-R subscales were collinear. Related to practical ability in intubated patients, the decision was made to keep CRS-R “visual.” Multivariate analysis was computed, and results are presented in table 4. According to statistical distribution and clinical relevance, CRS-R “visual” subscore was dichotomized as presented in Figure E2 (Supplemental Digital Content, https://links.lww.com/ALN/B322, presenting CRS-R “visual” subscore dichotomization).

Starting from this model, we created a score with weighting related to ORs and ranging from 1 to 14 (table 5). Bootstrap validation was performed for the construction of ROC curve presented in figure 2. AUC was 0.82 (95% CI, 0.73 to 0.91) for CRS-R “visual” subscale-based multivariate model. In our cohort, at the cutoff point of 9 determined by ROC analysis, positive and negative predictive values for extubation failure were 89 and 66% with a sensitivity of 84% and a specificity of 75%, respectively (table 6).

Extubation failure rates across the original cohort are presented in figure 3. Scores beyond presented cutoff of 9 show low extubation failure incidences. Patients presenting with at least two operating airway components succeeded extubation in 90 versus 10% if less (fig. 4A). In each subgroup of operating airway functions (0, 1, 2, or 3), extubation success rates were 38, 32, 67, and 90%, respectively, independent of the type of operating function: gag reflex, cough, or deglutition (fig. 4B).

ROC curve of model including only airway’s function has an AUC of 0.79 and was not significantly different from ROC curve of model integrating neurologic status (AUC, 0.82). Notwithstanding, this last model is more parsimonious with lower log-likelihood (55 vs. 60). When considering patients with low consciousness levels (CRS-R “visual” scores 0, 1, and 2), predictions of extubation success were 38 versus 85% when considering operating airway’s functions (fig. 5). No extubation delays or erroneous primary tracheostomy indications were revealed by retrospective analysis of electronic patient records.

This study identified risk factors associated with extubation failure in a cohort of neurocritical care patients with severe brain injuries. A pragmatic predictive clinical score, easy to perform at the bedside, was elaborated and validated.

Weaning of MV requires two successive steps: weaning of pressure support (ventilator) and liberation of the airway from the endotracheal tube. Brain-injured patients, usually not affected by cardiopulmonary incompetency as a cause of their critical illness, generally succeed in the first step, but extubation per se remains challenging.

In our cohort, eligibility for the first SBT was prolonged related to the severity of the patients. Comparable duration of MV is classical in this type of population.³¹  As expected and despite prolonged MV, autonomy from the ventilator could be obtained easily with less than two failed SBT before extubation, and no difference between groups (Table E3, Supplemental Digital Content, https://links.lww.com/ALN/B322, presenting characteristics of patients on successful SBT day). Extubation failure in our work reached 31% (24% within 48 h). This rate is consequent and is in accordance with literature with reported values of 20 to 40% in neurocritical care.^7,8  As a comparison, extubation failure in general ICU patients ranges between 10 and 20%.³  It has to be mentioned that some studies in this field found some surprising low rates in extubation failure.^10,13  It could be related to population disparities, for example, elective neurosurgical patients with short duration of MV or exclusions of tracheostomy without any extubation attempt for severe patients in some studies. In our cohort, no tracheostomy was performed unless the patient was not able to sustain SBT. It happened in four nonincluded patients unable to tolerate multiple weaning trials related to infratentorial ischemic stroke, with bulbar respiratory drive palsy in three patients and prolonged neuromuscular weakness in one patient.

In this population, from extended epidemiologic, clinical, and biologic criteria concerning neurologic, hemodynamic, and ventilatory functions, we identified few independently associated with extubation failure: predominantly loss of upper-airway protective reflexes and to a lesser extent loss of minimal behavioral clinical evidence of consciousness.

Coplin et al.⁹  demonstrated that brain-injured patients meeting standard weaning criteria could be extubated irrespective of their upper-airway function and their mental status could be evaluated with the GCS. In their cohort, some patients with a GCS as low as 4 tolerated extubation. Extubation’s delay was associated with increased risk of pneumonia and prolonged length of stay. In our cohort, patients with low GCS could also sustain extubation.

Nevertheless, Namen et al.⁸  identified GCS to be the best independent factor associated with extubation failure. ROC curve analysis identified a cutoff beyond GCS greater than or equal to 8 for extubation success (AUC, 0.681; OR, 4.9; 95% CI, 2.8 to 8.3; P < 0.001). Other studies found GCS with a threshold value of 8 to be a good indicator of extubation tolerance,^32–34  and American guidelines suggest weaning when adequate mentation defined as GCS greater than or equal to 13 is present.⁵  However, other studies did not recognize GCS as a predictor,^13,35  and in our cohort, GCS was not independently associated with extubation failure.

Indeed, GCS lacks information to differentiate subtle disorders of consciousness, does not assess brainstem reflexes, and is not evaluable in intubated patients.¹⁵  Identical GCS with total sum of scale components could indicate very different neurologic conditions.¹⁸  Therefore, results based on the global GCS in the aforementioned studies could be questionable. FOUR score that evaluates brainstem reflexes, respiratory function, and nonverbal signs of consciousness was developed for intubated patients in the ICU.¹⁹  FOUR was not independently associated with extubation failure in our cohort as in the study by Ko et al.¹⁰ 

CRS-R²⁰  allows clinical diagnosis of vegetative state, minimally conscious state (MCS), emergence of consciousness, and locked-in syndrome by means of behavioral assessment in patients with depressed mental status. It is a reference scale in neurorehabilitation. Even if CRS-R could be difficult to use in an acute care setting,^36,37  its validation in different countries and languages frequently included ICU patients.^38–40  Notably, Schnakers et al.³⁷  included 22 intubated patients in ICU among 60 patients with disorders of consciousness. In this study, the authors concluded that FOUR score was not as accurate as CRS-R to diagnose vegetative state and MCS even in an acute care setting. In our cohort, CRS-R total score was not statistically significant in multivariate analysis. CRS-R subscales “visual” and “arousal” were significant but collinear, meaning they bear identical information. Due to clinical relevance with facility to discriminate these clinical signs, “visual” assessment was integrated in our model. We found a change in extubation tolerance related to neurologic status when the patient was able to sustain visual pursuit (follow a mirror without loss of fixation). Recent works speculated that visual pursuit was a key surrogate of neurologic progression to MCS.^41,42 

Ability to follow stereotyped commands was a strong predictor of extubation success in a study by Anderson et al.¹³  Nevertheless, neither motor item of the GCS, FOUR, or CRS-R was predictive of extubation tolerance in our cohort. This type of response involves upper cognitive processing. Our different and more severe population could explain this discrepancy.

Airway function and especially ability to clear secretions by cough and swallow is fundamental to succeed weaning after liberation from the endotracheal tube.⁴³  These functions are frequently impaired in this setting due to inherent neurologic lesions⁴⁴  and prolonged critical illness under MV with artificial airway.^45,46 

Few components of upper-airway control have been studied in the literature, and quality of data are scarce.³  The ability to cough on demand,¹³  spontaneously,⁹  or during endotracheal suctioning^12,47  was associated with better extubation outcomes. One could argue that cough on demand only reflects neurologic status. A landmark study from Coplin et al.⁹  reported that comatose patients (GCS less than or equal to 8) with absent or weak gag and/or cough reflex sustained extubation, while the presence of spontaneous cough and low suctioning frequency were associated with better extubation outcomes.⁹  In order to have pragmatic objective clinical factors, we decided not to monitor volume and quality of secretions or suctioning frequency as reported elsewhere.^9,13,48  Spontaneous deglutition screening is used especially in stroke patients to assess dysphagia and aspiration risk.⁴⁹  Gag reflex could be a simple indicator of aspiration risk⁵⁰  but has been assessed in few studies.^9,13  No exact detail of evaluation technique was provided. Besides being challenging to test in intubated patients, a previous study reported that gag reflex was absent in 37% of healthy subjects despite preserved pharyngeal sensations.⁵¹  Such an observation might create confusion in interpretation of results of published studies, and further investigation is needed.

As with cough, we did not evaluate qualitative aspects of gag reflex and deglutition. We simply assessed airway factors on a yes/no basis more compatible with daytime clinical practice.

Criteria associated with hemodynamics, medical history, and interestingly respiratory functions did not correlate with extubation failure in this population. Indeed, as already reported, standard weaning parameters do not predict extubation failure of neurocritical care patients.¹⁰ 

Finally, our study highlights that extubation of brain-injured patients relies on a minimal level of consciousness and more importantly on maintenance of airway protective reflexes (gag reflex, cough, and deglutition). This observation corroborates the work by Manno et al.,⁵²  where neuro-ICU patients who sustained SBT and had favorable airway characteristics according to the airway care score (sputum quantity, character and viscosity, cough to suction, and suctioning frequency) were randomized to early or delayed extubation regardless of GCS. They observed that early extubation of patients with impaired mental status was feasible, without increase in neither reintubation rate nor mortality. This is in accordance with our results and results of others⁹ ; if airway is functional, low consciousness level does not alter extubation tolerance.

We used an extended time definition of extubation failure. Indeed, current trends tend to enlarge the definition to 5 to 7 days³  compared to the classical 48 h even if variability in definition is frequent.⁵³  We considered this cutoff more appropriate for this population. Early failure might explore cardiorespiratory incompetency, whereas delayed reintubation could reflect neurologic and airway impairment. However, we performed subgroup analyses before and beyond 48 h to stick with international guidelines⁴  and found no difference (Tables E1 and E2, Supplemental Digital Content, https://links.lww.com/ALN/B322, presenting data about patients with extubation failure before 48 h).

Relatively low observed rates of stridor in our study could be explained by differences in endpoint definition and patients’ selection from other studies.^12,54  Notably, extended timepoint beyond 48 h was taken to define extubation failure, and no selection was done on the cuff leak test or ability to cough and deal with bronchial secretions. Strict monitoring every 4 h of the cuff pressure was a part of the protocol in the three ICUs. Even lower stridor rates could be found in the neurocritical care setting.^13,55 

Our study has several limitations that deserve discussion. First, observational and monocentric design limits generalization of results. However, the initial goal was the development of a prediction score. Even if statistical bootstrap validation is an accepted method, prospective, multicentric validation on a larger cohort remains the definitive standard and is urgently mandated. However, because it does not require distributional assumptions (such as normally distributed errors), the bootstrap can provide more accurate inferences when the data are not well behaved or when the sample size is small.⁵⁶  The lack of an external cohort validation is the main limitation of the study. However, this preliminary study was intended to explore inherent bias, feasability of the generalization to a larger multicentric trial. Moreover, this preliminary study allowed the selection of pertinent criteria involved in the extubation success among a very large panel of hypothetic ones. Second, deglutition was not assessed with paraclinic examinations such as fibroscopy or videofluoroscopy, which could have been more sensitive⁴⁵  but not routinely available and difficult with orotracheal intubation. Third, cough was determined as present spontaneously or during suctioning. Few studies reported the use of peak flow systems to evaluate cough strength and found a good correlation with extubation success.^47,48  Quantitative evaluation on demand was limited in our severe brain-injured patients. Fourth, the incidence of pneumonia was high (greater than 70%). In a recent study, Asehnoune et al.³¹  reported an incidence of 68% in a cohort of severe brain-injured patients with comparable MV duration. Fifth, physicians assessed patients at a unique time point even if fluctuations notably in neurologic examination might exist.⁵⁷  Sixth, NIV in brain-injured patients might be considered inadequate related to upper-airway dysfunction.⁵⁸  Twelve patients (28%) in the extubation failure group were concerned. This therapy was administered with strict clinical observation. Not surprisingly, every patient treated with NIV was promptly intubated (within a few hours). Ultimately, one could argue that early tracheostomy before any extubation attempt could be beneficial in this population, but optimal timing still has to be demonstrated.¹ 

In this preliminary study, extubation of brain-injured patients with impaired mental status seems reasonably possible providing a tight control and functioning of protective airway reflexes. The use of a pragmatic predictive clinical scoring system could help decision-making. External validation in a large, multicentric, prospective fashion is urgently needed.

The authors acknowledge the nurses, respiratory therapists, and critical care fellows and staff who participated in patients’ management and care from the Department of Perioperative Medicine, University Hospital of Clermont-Ferrand, Clermont-Ferrand, France.

Support was provided solely from institutional and/or departmental sources.

The authors declare no competing interests.