Among intubated patients, we can often control the pO2, pCO2, and pH. Sounds great. Unfortunately, we also have no idea which numbers we should be targeting. Should we target a normal pH, or permissive hypercapnia? Should we shoot for a normoxia, try to maximize the oxygen delivery (hyperoxia), or allow for permissive hypoxemia? Nobody knows. We check ABGs and tweak the ventilator, but it's often dubious whether this helps our patients.
Although oxygen is necessary for life, it's also involved in burning stuff (whether that may be a forest fire or oxidative tissue damage). There is increasing recognition that hyperoxia may be harmful, particularly following cardiac arrest and myocardial infarction (e.g., the AVOID trial). This leads to the current study: is targeting a high oxygen saturation harmful in critically ill patients?
Background: Oxygen physiology and the silliness of targeting a high oxygen saturation.
The amount of oxygen delivered to the body (the DO2) is calculated as a product of the cardiac output, oxygen saturation, and hemoglobin:
DO2 = 13.4(CO)(Arterial O2 sat)(Hemoglobin concentration)
To survive, the DO2 should be comfortably above the body's oxygen consumption (VO2). Now, let's consider the typical range of variation in these factors for a patient in the ICU:
- Cardiac output may vary from ~2-8 L/min (4-fold variation)
- Hemoglobin usually varies from ~7-14 mg/dL (2-fold variation)
- VO2 may vary ~3 fold depending on temperature and paralysis
- Arterial oxygen saturation usually varies from 85% to 100% (1.2-fold variation)
Thus, the amount of variation in oxygen saturation is minimal compared to the other variables which make up DO2 and VO2. Changes in oxygen saturation within the 85-100% range will have relatively little impact on the DO2/VO2 balance (compared to other variables, which are largely unmeasured). We focus on the oxygen saturation because it's on the monitor and blinking in red, but it is generally a minor player in oxygen balance.
Targeting an excessively high oxygen saturation can be harmful
Given that oxygen saturation doesn't have much impact on oxygen delivery, excessive focus on saturation can actually be harmful. For example, if we crank up the PEEP to increase the oxygen saturation, this may actually drop the cardiac output, with an overall negative effect on the DO2 (more on this here).
For ventilated patients without severe hypoxemia, targeting a high oxygen saturation may still be harmful. Targeting a higher saturation leads to higher FiO2 and PEEP. Even if these settings are only minimally elevated (e.g. FiO2 55% and PEEP of 8 cm), this may impair ventilator weaning. Most weaning protocols require that the patient meet certain FiO2 and PEEP cutoffs prior to weaning, so even a trivial increase in these parameters may delay extubation (1).
Girardis M et al: Effect of conservative vs. conventional oxygen therapy on mortality among patients in an intensive care unit: The Oxygen-ICU randomized controlled trial.
This is a RCT involving all adults admitted to a single medical-surgical ICU with an expected length of stay >72 hours. Patients were randomized into two groups:
- Conservative group: Target saturation 94-98% or PaO2= 70-100 mm
- Conventional group: Target saturation 97-100% with PaO2<150 mm
434 patients were included in the final analysis. The time-weighted average PaO2 was indeed lower in the conservative group (87mm vs. 102mm). The conservative group had a lower ICU mortality (the primary endpoint) as well as fewer episodes of shock, liver failure, and bacteremia:
- It was terminated early following unplanned interim analysis in the wake of an earthquake that damaged the hospital, impairing enrollment.
- It is a single-center study, which may amplify the effects of how a particular group of physicians and respiratory therapists respond to oxygen levels.
- Patients with COPD exacerbation or severe ARDS were excluded, which could limit generalizability.
- There were baseline imbalances between the groups, with patients in the conventional group being sicker.
- The endpoint of ICU mortality may be less robust than hospital mortality or 90-day mortality endpoint.
Why I tend to believe the conclusions of this study
This study is believable because it generally meets my expectations (2). First, I feel that targeting an oxygen saturation of 97-100% is an unreasonably high saturation target. For more opinions, I performed a twitter poll (below). People responding to this poll were also usually comfortable with lower levels of oxygen:
As discussed above, my expectation would be that targeting a saturation of 97-100% would cause problems among patients on mechanical ventilation, including delayed ventilator weaning. This seems to have happened. Patients in the conservative oxygen group were weaned faster (with an average of 24 more ventilator-free hours, p=0.02).
As might be expected, the benefit of a conservative oxygen strategy was driven primarily by patients who were intubated at admission:
A 9% absolute mortality benefit is surprisingly high, but the 95% confidence intervals of the mortality benefit are wide (1.7%-15%). Mortality benefit may have been increased artificially by baseline imbalance between the groups and early termination of the study.
Is hyperoxia dangerous?
Unfortunately, this study doesn't answer whether hyperoxia itself is harmful. Targeting a high oxygen saturation among intubated patients seems to be harmful, but the mechanism of harm is unknown:
This depends on your current practices regarding oxygen:
- If your center is currently targeting an oxygen saturation of 97-100%, then this study would suggest that you could reduce your oxygen target. However, it seems that relatively few centers are targeting oxygen levels this high. Furthermore, there is already ample evidence in the literature that targeting lower oxygen levels is safe.
- If your center is currently targeting oxygen levels below the 97-100% level, then this study supports your current practice. It seems that most units will be included in this second group.
However, limitations in study design suggest that this study by itself shouldn't necessarily drive changes in practice. Hopefully the multi-center ICU-ROX study by Dr. Paul Young of the ANZICS study group will bring more definitive answers.
- The Oxygen-ICU trial found benefit from a conservative (94-98%) oxygen target when compared to a liberal (97-100%) oxygen target.
- It is conceivable that targeting a lower oxygen saturation led to lower PEEP and FiO2 levels, facilitating earlier extubation.
- This study suggests that targeting a saturation of 97-100% may be too high. However, most clinicians may already targeting levels lower than this.
- Girardis M et al. Effect of conservative vs. conventional oxygen therapy on mortality among patients in an intensive care unit: The oxygen-ICU randomized trial. JAMA 2016; 316(15) 1583-1589
- ARDS physiology: what is our oxygenation goal? (PulmCrit)
- HFNC for early ventilator weaning in hypoxemic respiratory failure (PulmCrit)
- The case of the Erratic Pendulum (EMNerd explores early study termination)
- ICU-ROX trial discussion with Paul Young (Critical Care Reviews Podcast).
Update (4/17/18) – Fresh tweetorial by Gabriel Bosslet with more evidence & updated studies:
Recently in our #MICU, I took pics of the oxygen sats of patients on oxygen supplementation.
What is the optimal oxygen supplementation strategy for a clinically stable patient in the ICU?
— Gabriel Bosslet (@gbosslet) April 16, 2019
And, if you want to skip this entire post & tweetorial, all you really need is this one tweet:
Agree. The point is to set an upper limit as well so patients don’t hang out with sats of 98% on 50% fio2 all day long!!
Order should read “keep oxygen saturation 89-96%”
— Gabriel Bosslet (@gbosslet) April 17, 2019
- I'm not saying that this should happen. One of my tricks for getting patients off the ventilator is accepting a lower oxygen saturation. Patients who are hypoxemic but otherwise ready to be extubated (e.g. from a work-of-breathing standpoint) can frequently be extubated to high-flow nasal cannula, as discussed previously here.
- I will go out on a limb to say that this isn't a case of circular logic, but rather Bayesian reasoning (this allows me to term my biases as “prior probabilities”).
Image credit: Opening image from www.freeimages.com/photo/fire-1519428