The Risk of Hyperoxemia in ICU Patients. Much Ado About O₂

The provision of supplemental oxygen is perhaps the most ubiquitous therapeutic intervention in critical care medicine. To prevent and treat hypoxemia, oxygen is typically administered liberally in the ICU, and patients are often exposed to a high Fi_O2 and higher than normal Pa_O2 (1–3). The association between exposure to hyperoxemia and mortality risk in critical illness has been reported in a number of studies (4–6). However, these studies did not allow for a robust assessment of any dose–response relationship. If such a dose–response relationship were demonstrable, this would increase the probability of a causal relationship between exposure to hyperoxemia and mortality risk.

In this issue of the Journal, Palmer and colleagues (pp. 1373–1380) report the findings of an observational study conducted in five university hospitals in the United Kingdom in which they examined the association between longitudinal exposure to hyperoxemia and ICU mortality (7). Hyperoxemia was defined as a Pa_O2 >13.3 kPa (100 mm Hg), on the basis that values exceeding this threshold can only be achieved with supplemental oxygen. Patients who stayed in the ICU for 24 hours or less were not included in the analysis. Another important exclusion was patients who had received cardiopulmonary resuscitation in the 24 hours preceding ICU admission. This exclusion is important because the basic science that supports the notion that exposure to hyperoxemia is harmful in hypoxic ischemic encephalopathy is comparatively strong (8) and is supported by prospective observational data (9). The aim of the study was to examine the association between longitudinal exposure to hyperoxemia, defined as time-weighted mean exposure to supraphysiologic Pa_O2, and mortality.

Regrettably, as outlined by the investigators, modeling exposure to hyperoxemia is inherently complicated for several reasons. First, patients can recover and leave the ICU or die and stop contributing data in a nonrandom fashion. Second, to measure the effect of longitudinal exposure to hyperoxemia, a window of time to observe exposure and subsequent effect is needed. Third, patients receiving more vigorous supplemental oxygen therapy are more likely both to be more severely ill and to develop hyperoxemia. Fourth, such patients are also more likely to have arterial blood gas measurements performed (surveillance bias), thus linking the probability of detecting exposure with the likelihood of being both sicker and more frequently monitored at the time of identified exposure and, probably, both before and thereafter. In an effort to account for these complexities, the investigators divided their cohort into groups with different time windows for potential exposure to hyperoxemia. A total of 77.5% of patients were exposed to hyperoxemia by Day 1, increasing to 90.6% by Day 7.

The authors found that exposure to any hyperoxemia was statistically significantly associated with increased mortality risk in patients with 3, 5, and 7 days of potential exposure, but the dose of hyperoxemia was not. In this regard, when considering the data from this study and whether they reflect a “causative pathway” or yet another association between one of the innumerable variables measured in ICU and outcome, one might want to reflect on the canonical Sir Bradford Hill criteria, which can be applied to assess the “functional relationship” between exposure and outcome (10).

In this regard, when relating hyperoxemia to mortality, the strength of association is weak, the consistency of association is limited and the specificity of association is low, the temporality is complex, the biological gradient is absent, the plausibility is unclear, the coherence is uncertain, the experimental data (outside of cardiac arrest models) are lacking, and the presence of an analogy for a similar gas-related physiological variable is absent. Thus, given the above consideration, and after acknowledging that this study provides the most detailed epidemiological data available (7), much doubt remains. Moreover, there is one more shortcoming: For clinicians considering the most appropriate dose of oxygen to administer to their patients, the risk associated with hyperoxemia is not the only salient risk. Attempting to minimize the risk for exposure to hyperoxemia with conservative oxygen therapy regimens can inadvertently increase exposure to hypoxemia (11); thus, the risks of hyperoxemia and of hypoxemia need to be considered together, not in isolation.

Despite these concerns, these doubts about hyperoxemia as a possible causative contributor to unfavorable outcomes, and the lack of a demonstrated dose–response relationship between hyperoxemia and mortality, the current study suggests that this issue is important. It also implies that clinicians should wait for data from high-quality randomized controlled trials before implementing conservative oxygen regimens in the ICU. In this regard, although a recent systematic review and meta-analysis reported reduced mortality in acutely ill adults when oxygen use was comparatively restrictive, this included relatively few studies of critically ill patients (12). Moreover, the majority of the critically ill patients included were from a single-center trial, which was stopped early at an unplanned interim analysis (13).

Two comparatively large randomized controlled trials comparing liberal versus conservative oxygen regimens in ICU adults will be reported in the near future (14, 15). These trials promise to substantially advance our understanding of the most appropriate dose of oxygen to use in ICU patients. However, given that the primary outcome for one trial is ventilator-free days (14) and that the other trial focuses on patients with hypoxic respiratory failure (15), further high-quality randomized controlled trials will be needed to establish whether conservative oxygen therapy regimens that seek to limit exposure to hyperoxemia reduce mortality risk.

The provision of supplemental oxygen is such a fundamental treatment for critically ill patients that we consider that future trials should be powered to detect a true minimally important difference in mortality. We submit that an absolute effect on mortality of 1.5 percentage points represents such a difference. A treatment effect of this magnitude would have profound global public health importance because for every 100,000 patients treated, such a difference would equate to 1,500 lives saved or lost. Moreover, in the event of a zero percentage point absolute mortality difference between treatment groups, 95% confidence intervals would be expected to exclude the possibility of an absolute increase or decrease in mortality of well under 1 percentage point. In this situation, the possibility of a clinically important effect of conservative oxygen therapy on in-hospital mortality would effectively be excluded. Until the data from such studies become available, a course of action that prudently avoids unnecessary hyperoxemia while monitoring for and protecting against hypoxemia remains the most sensible approach.