When I was a resident, an EMT brought a patient in on a 100% non-rebreather mask (NRB) and I asked if she needed all that oxygen. His reply was, “Oxygen never hurt anybody.” Interesting response, and one that current data suggests may be more myth than truth.
Oxygen is one of the most common treatments in emergency medicine, being given to over a third of patients transported by ambulance and one in every six or seven hospitalized patients. Traditionally, supplemental oxygen has been used generously in the treatment of all kinds of patients, ranging from those with myocardial infarction (at least 2L of oxygen by nasal cannula even if saturations are normal) to routine trauma patients, and even patients being transported to the hospital for non-emergent, non-pulmonary symptoms. For patients with some hypoxia, 100% NRB is often applied – all under the supposition that oxygen never hurt anybody. Unfortunately, this may not be true. While low oxygen delivery puts tissues at risk of hypoxic injury, too much oxygen may also be detrimental.
All the way back in 1899, Lorain Smith demonstrated the development of fatal pneumonia in rats administered high concentration of supplemental oxygen for a mere 4 days. Then, in 1986, using radiolabeled tracers, Griffith and colleagues demonstrated that administration of 30% oxygen for 45 hours to healthy volunteers resulted in the development of leaky airways. However, these studies were thought to largely not apply to sick patients, and specifically not to patients with cardiac or pulmonary pathophysiology. However, recent data, albeit observational in nature, has consistently found an association between hyperoxia and worse outcomes (i.e either higher mortality, more morbidity, or worse neurological outcomes in survivors) in a number of conditions. In 2010, Kilgannon and colleagues reported an association between hyperoxia and worse mortality and neurological outcomes in 6300 patients after cardiopulmonary resuscitation from cardiac arrest (Kilgannon JH, et al. JAMA. 2010;303(21):2165-2171). In this study, the 1171 patients (19%) who had normoxia experienced 45% mortality compared to 57% in the 3999 patients with hypoxia (63%) and 63% in the 1156 patients (18%) with hyperoxia (defined as PaO2 above 300). Numerous subsequent reports have found similar supportive evidence of an association between hyperoxia and worse outcomes in patients after cardiac arrest (Bellomo R, et al. Crit Care. 2011;15:R90 ;Janz DR, et al. Crit Care Med. 2012;40(12):3135-3139 ) and a meta-analysis published in 2014 combining 14 studies found an association between hyperoxia (PaO2 > 300) and hospital mortality in patients post cardiac arrest with an OR of 1.4 (Wang CH, et al. Resuscitation. 2014;85(9):1142-1148). Since the initial Kilgannon report, additional studies have also demonstrated an association between hyperoxia and worse outcomes in patients with COPD, ARDS, stroke, traumatic brain injury, and critical illness requiring mechanical ventilation. Other meta-analyses found that hyperoxia was associated with a 20-40% increase in hospital mortality in a heterogeneous population of critically ill patients (Damiana E, et al. Crit Care. 2014; 18(6):711; Helmerhorst JHF, et al. Crit Care Med. 2015; 43(7):1508-1519). However, these results were tempered with a before-after study conducted in Australia that failed to find any difference in clinical outcomes when a conservative oxygen strategy, targeting oxygen saturations 90-92% was implemented (Suzuki S, et al. Crit Care Med. 2014;42(6):1414-1422.). Conducted as a pilot study, less oxygen was administered during the conservative time period, although the difference was average saturations of 95.5% compared to 98% and PaO2 of 73 compared to 108. So not huge differences in oxygen content, and certainly not the degree of hyperoxia seen in the observational studies.
All of these studies employ a study design that only allows an interpretation of the relationship between hyperoxia and worse outcomes as a mere association. Given the observational nature of the studies, the association was certainly thought provoking and even worrisome for some clinicians, but not robust enough to dramatically change practice. Randomized controlled trials were needed.
The same group from Australia conducted a small RCT (CLOSE study) in 103 adults with P/F < 300 and mechanically ventilated for at least 24 hours (Panwar R, et al. Am J Respir Crit Care Med. 2016; 193(1):43-51). Patients were randomized to a conservative strategy targeting oxygen saturations 88-92% or a liberal strategy targeting saturations greater than 96%. They did not find a difference in any clinical outcome between the groups, although, again, they did not achieve very much separation between the two arms (average PaO2 70 vs 92 mmHg).
Then last year, the results of the AVOID study were published in Circulation. This multicenter study randomized patients with ST elevation myocardial infarction to 8L oxygen by nasal cannula versus room air (Stub D, et al. Resuscitation. 2015;131:2143-2150). Supplemental oxygen resulted in larger infarct sizes, more arrhythmias, and more recurrent myocardial infarctions compared to no supplemental oxygen. More evidence, and this time in the form of a randomized controlled trial, that hyperoxia may be detrimental, albeit in myocardial infarction and not critically ill, mechanically ventilated patients.
And two months ago, Girardis presented the results of the Oxygen ICU trial at the European Society of Intensive Care Medicine (ESICM) (ESICM Hot Topics Post). This trial, now published in JAMA, is a single center, open-label, randomized trial of two different oxygen targets in almost 450 critically ill patients expected to be in the ICU for at least 72 hours (Girardis M, et al. JAMA. 2016;316(15):1583-1589). The control group was managed according to standard of care where FiO2 of at least 40% was used to target oxygen saturations of 97-100% and PaO2 up to 150 mmHg were targeted. The intervention group was managed with target oxygen saturations of 94-98% using the lowest FiO2 possible. Patients in the control group were transported with 100% oxygen while those in the intervention arm were transported on their current oxygen dose, unless saturations decreased below 94%. Again, a relatively small difference in oxygen saturation / content was achieved between the two groups (median PaO2 of 102 vs. 87 mmHg). However, the conservative oxygen management resulted in a significant survival advantage, and fewer episodes of shock, liver failure, or bacteremia. More provocative results. And a post-hoc analysis suggested lowest mortality for median time-weighted PaO2 between 87 and 93 mmHg and highest mortality for PaO2 above 107 mm Hg.
While these results again make me concerned that we may be hurting our critically ill patients with hyperoxia resulting from overgenerous use of oxygen, we also need to recognize the limitations of the data. It is a single center study so generalizability to other centers may be limited. Many patients only had one or two arterial blood gases, limiting the ability to robustly calculate a median time-weighted PaO2 value. The improved clinical outcomes seem out of proportion to the fairly minimal differences in separation of arterial oxygen content between the groups. This exaggerated effect may be due to unplanned early stopping as the study was stopped early due to a natural disaster in the area and inability to effectively continue to recruit patients.
Despite these limitations, this recent RCT, along with the preponderance of observational data, must make us at least consider the possibility that we are hurting our patients with hyperoxia. I think also playing into this consideration should be the fact that avoiding hyperoxia involves a change in practice of how we deliver oxygen, and not an expensive medication (in fact, it should also be cheaper). With all these data, I have tried to be more cognizant of high oxygen saturations in my patients, and to more aggressively wean oxygen in the critically ill patients for whom I provide care.
Let me know your thoughts and whether these emerging data have resulted in any changes in your practice.