In the chaotic early days of the COVID-19 pandemic, many potential shortages loomed large, but none seemed as terrifying and immediate as the potential shortage of mechanical ventilators, and the inability to support the lives of infected individuals who could not adequately breathe on their own (1). At the time of the initial pandemic surge, the United States had a finite supply of conventional full-function ventilators (i.e., those used for an intubated patient in the ICU) and basic ventilators (i.e., nonconventional ventilators [NCVs] including, portable transport ventilators and noninvasive ventilators), and an additional back-up supply in the National Strategic Stockpile. If and when there was a shortage of full-function conventional ventilators, use of basic ventilators seemed an obvious option. However NCVs, by design, do not provide all the features of full-function ventilators, and models may vary in settings and ability to provide consistent respiratory support in patients with changing pulmonary mechanics (2). Although initial projections of mechanical ventilator needs—and shortages—fortunately did not materialize nationally, supplies were exhausted in certain hospitals at certain times. The potential and realized ventilator shortages motivate the question: How did patients fare under this particular crisis standard of care? In addition, how does this inform our national preparedness strategy for inevitable future respiratory pandemics? In this issue of Critical Care Medicine, Gondhalekar et al (3) evaluate outcomes associated with the emergency authorized use of NCVs in the treatment of COVID-19 acute respiratory distress syndrome (ARDS). This retrospective cohort study was completed in a 12-hospital healthcare system in New York, the U.S. pandemic epicenter, during the early pandemic days of March and April 2020 when local conventional ventilator supplies were exhausted. Adult patients with confirmed COVID-19 and Berlin-definition ARDS requiring invasive mechanical ventilation were eligible for inclusion. Patients were classified based on ventilation exclusively with NCVs (exposed cohort) versus any exposure to conventional ventilators (control cohort). NCVs were defined as any nonfull-function ventilators including transport ventilators and noninvasive devices. A secondary analysis classified patients (NCV only vs. any exposure to conventional ventilator) based on ventilator exposure in the first 7 days after intubation. Allocation of NCVs was described as "first come, first served." The primary outcome was 28-day in-hospital all-cause mortality, which was modeled by logistic regression adjusting for baseline demographics, comorbidities, COVID-19-related therapies, acute physiology, respiratory parameters, ICU crowdedness, and longitudinal time. Treatment in a temporary ICU—defined as treatment in a hospital location that was not an ICU before the pandemic—was included as an additional adjustment factor. The authors comment that all patients, regardless of treatment location, had a 2:1 patient-to-nurse ratio, a board-certified intensivist supervising care, and daily rounding by a respiratory therapist. In a secondary analysis, additional propensity score matching was used to further address residual confounding between exposure groups. The final analysis included 1980 patients of which 452 (23%) were ventilated exclusively with NCVs. There were several important differences between the exposure groups. Patients in the NCV cohort had lower rates of certain comorbidities (i.e., hypertension, cancer, chronic neurologic disease), received less treatment with prone positioning, were more often cared for in temporary ICUs (notably at a rate of 58% vs. 34%), and had higher unadjusted/observed mortality. In the fully adjusted model, patients ventilated exclusively with NCVs had higher 28-day in-hospital all-cause mortality (odds ratio [OR] 1.41; 95% CI, 1.07–1.86). The secondary matched analysis showed similar results. When restricted to the early ventilation period (first 7 d of invasive mechanical ventilation), there was no observed difference in mortality (OR 0.88; 95% CI, 0.69–1.20). There are important limitations to consider in the interpretation of these results. The early pandemic study period is when NCVs were in highest use and includes a study population not exposed to emergent viral variants, evidence-based COVID-19 therapeutics, vaccination, and immunity from past infection. Perhaps most relevant to the question at hand, the early pandemic time was marked by very high (and in retrospect, likely maladaptive) rates of intubation and ICU admission, and high observed mortality, all of which changed dramatically and quickly as COVID-19 care evolved, even before targeted evidence-based therapies (4). If the United States ever saw a COVID-19 case volume that again prompts use of NCVs, the intubated population would be far different than the one studied here. Residual confounding is the biggest threat to validity and interpretation of the results in this (and most) observational studies. The use of an NCV for the sustained management of a patient with respiratory failure is a highly irregular clinical practice compared with usual care—a crisis standard of care. The study reports that ventilators were used on a first-come, first-served basis, but we know that patients do not arrive at a hospital one at a time. Clinicians faced with fewer conventional ventilators than patients awaiting intubation would almost certainly, and reasonably, take into account patient-level factors when allocating those conventional ventilators—and in turn, when allocating the next-up NCVs. This may have taken the form of giving NCVs to patients with a higher perceived likelihood of survival regardless of ventilator type (e.g., lower acuity, fewer comorbidities, and younger age); or alternatively, NCVs could have preferentially been given to patients who were perceived to be unlikely to survive despite ventilator type (e.g., highest acuity, comorbidity burden, and older age). Pooled patient characteristics by exposure group may not adequately reveal these potential sources of bias and the use of traditional risk adjustment including propensity matching is unlikely to be a perfect solution. Because conventional ventilators, if available, would always be (perceived) superior to NCVs, it is reasonable to assume that NCV use was concentrated at times of extreme capacity strain (known to be associated with worse outcomes [5]). This is also consistent with the higher rates of treatment with NCVs in temporary ICUs, an additional, highly irregular, management decision with its own confounding by indication and association with poor outcomes (6). It is not clear that adjustments for ICU crowdedness and treatment location are sufficient to overcome these analytic challenges. All that said, the authors should be commended for a valiant effort to address an important question in the face of analytic challenges and limitations largely inherent to the study question itself. Scarce resource allocation and considerations of life support equipment including mechanical ventilators, is an essential component of crisis preparedness for respiratory viral epidemics and pandemics, natural disasters, attacks on the public, and other mass casualty events (7). Resource allocation guidelines (8–11) aim generally to balance societal benefit for the greatest number of people with the needs of individual patients during periods when demand may overwhelm supply (12). But even thoughtful guidelines that explicitly address allocation of scarce mechanical ventilators and discuss use of NCVs to augment surge capacity, such as the 2015 New York State Department of Health Ventilator Allocation Guidelines published in response to Hurricane Sandy, often do not differentiate tiers of ventilators in allocation schema (11). The study by Gondhalekar et al (3) highlights two important and related issues during crises: the use of nonstandard therapies, and research during public health emergencies. Although all observational research is vulnerable to confounding, retrospective research on nonstandard therapies used during a prior public health emergency—whether focused on nonconventional devices or investigational or repurposed pharmaceuticals—is particularly fraught, as discussed. Nonrandom patient selection leads to confounding by indication and concomitant management decisions and situational factors, such as capacity strain, introduce other potentially impactful factors without a clear path to effective adjustment and unbiased modeling. Safety and equity considerations also motivate the use of nonstandard therapies within ethically designed trials (13). Use of a nonstandard therapy during crises as part of a prospective randomized trial is a theoretical solution, but one with its own set of barriers. In a public health emergency, trial enrollment may compete with out-of-study access to a given therapy (such as emergency use authorization, compassionate use, or off-label use), and swiftly-created inclusion and exclusion criteria may miss the mark in the setting of partial and rapidly changing information. Trials may not achieve adequate power to provide the chance of finding a meaningful result, equipoise can be lost, and the standard of care comparator can evolve (13). Use of newer trial designs such as adaptive platform trials that allow for precrisis infrastructure builds and in-crisis simultaneous testing of multiple therapeutics, ongoing data analysis, and results-driven addition and removal of therapies, appear to be an essential piece of the solution (14). Reconsidering the study by Gondhalekar et al (3), could a platform trial have provided a result with fewer bias concerns? As an exercise in theoretical study design, certainly, but logistical barriers abound. For now, we are left with hard work and a result-with-asterisk that patients managed exclusively with NCVs had worse outcomes. Whether this should motivate a change to U.S. ventilator stockpile strategies, as the authors suggest considering, requires complex risk modeling, a more thorough COVID-19 pandemic after-event evaluation, and stockpile cost and maintenance considerations.
No takes yet. Share an insight, caveat, or question.
Agarwal et al. (2024) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: