To Intubate or Not Intubate, That Is the Question

The coronavirus disease (COVID-19) pandemic amplified important controversies in the management of acute hypoxemic respiratory failure. First and foremost, the role of noninvasive oxygenation strategies, such as standard oxygen, noninvasive ventilation (NIV), and high-flow nasal cannula (HFNC), has been greatly debated. Some have warned that spontaneous breathing in the setting of high respiratory drive can worsen lung injury, and thus intubation should be considered prophylactic rather than supportive care (1). In contrast, others have reasoned that the inherent complications of mechanical ventilation, namely immobility, infection, and cognitive impairment, should be avoided with the use of noninvasive oxygenation strategies. They reasoned that these approaches can reduce respiratory effort (2) and render spontaneous breathing noninjurious (3).

Furthermore, if noninvasive oxygenation strategies are used, it remains unclear what approach (NIV, HFNC, or standard oxygen), and in whom, is best. Clinical trials attempting to address this question rely on outcomes such as rates of endotracheal intubation (4–6). In an effort to reduce potential bias, the criteria for intubation are often prespecified, given that the nature of the intervention makes blinding impossible. These criteria are largely based on precedent (5) and tend to have face validity to reflect what a reasonable clinician would agree are clinically relevant thresholds to avoid unnecessary delays of life-saving invasive mechanical ventilation.

In this issue of the Journal, Yarnell and colleagues (pp. 271–282) suggest that these intubation thresholds do not reflect everyday clinical practice (7). Using two retrospective cohorts of ICU admissions to academic centers in Boston and Amsterdam, the rate of endotracheal intubation within 3 hours of meeting criteria set forth by a clinical trial (4) ranged from 9% to 13%. Although worsening hypoxia was associated with increased rates of endotracheal intubation, only 17% to 19% of the cohort were intubated within 3 hours of a Pa_O2:Fi_O2 of <80 in the Boston and Amsterdam cohorts, respectively. Interestingly, the rates of intubation within 3 hours did not seem to vary substantially based on the oxygenation strategy used (NIV, HFNC, or nonrebreather) at the time the hypoxia thresholds were met. Finally, in a Bayesian analysis, increasing age, chronic obstructive pulmonary disease (COPD), heart failure, and Black race were associated with decreased probability of endotracheal intubation at any time during the ICU stay after meeting a threshold. In contrast, the use of NIV and increased work of breathing were associated with increased probability of intubation.

There are several limitations to this study. First, these findings reflect the care practices at only two academic centers before the pandemic and may not be generalizable to many ICUs of today. As an example of practice variation, HFNC was more commonly used in the Boston cohort but not as commonly used during the pandemic (8). In addition, it remains unclear whether the individual intubation thresholds, when ignored or unrecognized, lead to worse outcomes in terms of mortality. Because no widely accepted thresholds exist, clinicians may not rely on an individual threshold to initiate invasive mechanical ventilation. Instead, they may rely on clinical trajectory, work of breathing, comorbidities, or additional organ failures, which were not reliably captured in this cohort. Although the decreased use of invasive mechanical ventilation in patients of Black race is concerning, the cohort had very limited racial diversity, and it is unknown if withholding intubation exacerbated disparities in terms of mortality. Finally, the entry criteria for the clinical trial participants (4) were strict and included tachypnea with a Pa_O2:Fi_O2 < 300 without hypercarbia, which ensured a highly selected group at risk for intubation as opposed to the broad criteria used in this observational study. Indeed, heart failure and COPD were associated with lower probability of intubation, which could be explained by the known efficacy of NIV in the prevention of intubation and mortality in patients with exacerbations of these conditions.

Regardless, if there is a disconnect between the criteria for invasive mechanical ventilation in clinical trials and everyday clinical practice, the pursuit to declare a “winner” for the best noninvasive oxygen strategy just got more challenging. The more important question is if a complex, intuitive, and individualized clinical decision such as initiating invasive mechanical ventilation can or should be protocolized? Prior research has shown that a protocolized approach to sepsis resuscitation (9) or postextubation respiratory support (10) is not better than an astute bedside provider customizing their management for individual patients. After all, critical care providers are dealing with syndromes with substantial heterogeneity and not diseases per se.

On the other hand, if nonclinical factors such as race, ethnicity, or socioeconomic status risk the influence of implicit bias on clinical decision making, perhaps an unbiased approach with standardized thresholds is warranted. These thresholds must have clinical relevance in terms of preventing delays that would otherwise lead to harm. Often a first step to define such a threshold is to develop prediction tools or analyze risk factors for failure (11–13), but these studies are in effect predicting clinician behavior and not necessarily clinical need for invasive mechanical ventilation. As shown in this analysis, these bedside clinical prediction tools were no better than simple thresholds based on severity of hypoxia. Given that clinical trajectory may play a role in decision making, perhaps techniques that use discrete-time survival analysis to derive prediction tools from large datasets may have better accuracy (14). However, an important step before proposing widespread use of any derived and validated clinical threshold would be to determine if early warning leads to better outcomes (15). Whether the early intervention used when a clinical threshold is met should include early endotracheal intubation is yet to be determined.