Navigating the Nuances around Extubation Decisions and Observational Evidence

It is without controversy that patients with severe brain injury often require intubation and mechanical ventilation. Yet, significant controversy exists regarding when to extubate them. Accordingly, previous studies have demonstrated substantial variation in clinical practice (1). Patients with primary neurological disease (e.g., subarachnoid hemorrhage, traumatic brain injuries, or ischemic stroke) often require intubation because of an inability to protect their airway resulting from depressed mental status, inadequate cough, or neuromuscular respiratory insufficiency rather than type I or II respiratory failure. Many critical care practitioners are taught that before extubation, or sometimes even before attempting a spontaneous breathing trial (SBT), patients should be calm, cooperative, awake, and able to follow commands or at least have a mental status better than an arbitrary cutpoint of Glasgow Coma Scale (GCS) score >8 (2–4).

However, these standard extubation criteria are unattainable in many patients with brain injury (5, 6). The GCS is a reliable and reproducible measure with which to perform a focused neurological assessment, but, like any tool, it has limitations and is often confounded by specific neurological diseases. For example, a patient with a left middle cerebral artery stroke is as likely to protect their airway as a patient with an identical contralateral stroke, but their aphasia and difficulty following commands will result in lower scores on the GCS. Had the patient experienced a right middle cerebral artery stroke, the patient would be more likely to follow commands but might be unable to open their eyes and so would appear less alert. This will similarly lower their GCS score without affecting their actual level of consciousness or airway protection (7).

In 2000, Coplin published an important study demonstrating that many patients with brain injury experience delayed extubation after meeting typical extubation criteria, defined as having a stable neurological status and meeting conventional cardiorespiratory parameters, and that this delay was associated with adverse outcomes and greater costs (1). There was substantial variation in the timing of extubation among patients with impaired consciousness. Yet, many patients with low GCS scores underwent prompt and successful extubation. Since then, multiple investigations have attempted to identify predictors of extubation failure in this population, but most have been small and conducted at a single center, limiting generalizability (8–11). In this issue of the Journal, Angriman and colleagues (pp. 339–346) advance this literature by comparing prompt versus delayed extubation strategies among a sizable cohort of 1,406 patients with brain injury admitted to eight different ICUs (12). The investigators applied target trial emulation, a modern framework for observational comparative effectiveness studies that reduces the risk of several common sources of error (13). The authors incorporate design features from the planned but not completed NEURO-ETT trial (Neurologically Impaired Extubation Timing Trial) into their observational analysis. Accordingly, they set their eligibility criteria to include adults with acute brain injury and a Sedation-Agitation Scale score >1 who pass a standardized SBT. Fifty-seven percent of included patients were extubated on the day they first passed an SBT. After accounting for a reasonable list of potential confounders, the authors found that patients extubated promptly after meeting eligibility criteria experienced more ventilator-free days and lower 28-day mortality than patients for whom extubation was delayed, despite higher rates of reintubation. These findings were consistent between their unadjusted and adjusted analyses and in several sensitivity analyses designed to test the robustness of their assumptions and approach.

Key strengths of the study include a large population of patients with diverse brain injuries and the inclusion of patients from multiple ICUs. The various ICUs followed a standardized protocol for SBT screening and performance, which is shared alongside the study. Although observational, their study’s adherence to the target trial emulation framework reduces the risk of several biases that can occur when observational comparative effectiveness studies do not follow study design principles of randomized trials (including immortal time bias and certain forms of selection bias) (14). They also employ good practice for confounder adjustment, incorporating clinical knowledge into an a priori causal directed acyclic graph and presenting it for consideration alongside their study.

Where might this study go wrong? Put another way, why might NEURO-ETT, if completed, yield a contradictory result, or why might we experience differing outcomes from an “early extubation” policy in our own ICUs? For one, target trial emulation does not guarantee adequate control of confounding. Prior studies and our own clinical experience suggest that there may be other confounders, such as the presence of a robust spontaneous cough, a suspected difficult airway, brainstem dysfunction/posterior fossae lesions, and clinical trajectory, that were not included in their propensity score model. If these potential confounders are strongly related to a clinician’s decision to extubate and the resulting risk of reintubation or death, accounting for them (or randomizing patients to early or delayed extubation) may result in different observed treatment effects. This study is also susceptible to the same concerns regarding generalizability that randomized controlled trials face. Most pertinent to this study, “extubation” is not a specific treatment that occurs at one time point. It also includes postextubation management, often with respiratory support and team-based care until the risk for immediate reintubation has passed. An inability to replicate the version of treatment (including postextubation management) that occurred in the study ICUs would yield a different effect of early versus delayed extubation in subsequent studies or in clinical practice.

So, what principles should guide clinicians facing the difficult task of making decisions about extubation readiness in patients with acute brain injury? Importantly, this and prior observational studies suggest that delaying extubation in patients with brain injury may be unnecessary and harmful. The present work implies that most patients who are at least somewhat responsive (Sedation-Agitation Scale score >1) and neurologically and hemodynamically stable warrant a trial of extubation after passing an SBT, even though patients with brain injury are at higher risk of extubation failure (15). At the same time, there are considerable nuances to the decision and large areas of remaining uncertainty. In particular, the degree to which providers should consider the interaction of depressed mental status with other factors, such as secretion burden, cough strength, and lesion localization, warrants further study. In our own practice, we are much more likely to attempt extubation in a patient with a low GCS score if a strong, spontaneous cough and minimal secretions are present. In contrast, we are more likely to defer a marginal extubation if a patient’s neurological status is rapidly improving and so they may be likely to have a greater probability of success within a few days. A patient and family’s wishes regarding provision of life-sustaining treatment also influences the decision, because specific wishes, such as a strong preference not to be reintubated and/or have a tracheostomy placed, will influence the aggressiveness with which we pursue a trial of extubation.

Unfortunately, no spontaneous test of airway protection/secretion clearance analogous to an SBT exists for patients with brain injury, so decisions regarding extubating patients with altered neurological function caused by brain injury will always require nuanced clinical judgment. A multicenter randomized trial of a clearly protocolized brain injury extubation procedure would provide high-quality data to guide practice. Until then, the present study demonstrates that delaying extubation may cause more harm than good and that well-designed observational studies can offer valuable insights into clinical care when trial data are unavailable.