The Association between Supraphysiologic Arterial Oxygen Levels and Mortality in Critically Ill Patients: A Multicenter Observational Cohort Study

Edward Palmer, Benjamin Post, Roman Klapaukh, Giampiero Marra, Niall MacCallum, David Brealey, Ari Ercole, Andrew Jones, Simon Ashworth, Peter Watkinson, Richard Beale, Stephen J. Brett, Duncan Young, Claire Black, Aasiyah Rashan, Daniel Martín, Mervyn Singer, Steve HarrisAmerican 2019 · DOI

TL;DR

This study investigated whether high levels of oxygen (hyperoxemia) in critically ill patients are linked to increased mortality. While they found an association between any exposure to hyperoxemia and higher ICU mortality, they did not observe a clear dose-response relationship, which tempers causal interpretations and highlights the need for further experimental research.

What Problem Does This Paper Address?

Oxygen is a vital drug in critical care, but its potential for toxicity, particularly at supraphysiologic levels (hyperoxemia), has been a subject of debate. Previous research has yielded conflicting evidence, with some studies suggesting harm and others finding no effect. A key challenge in this field is accurately measuring oxygen exposure, as it's often treated as a single event rather than a longitudinal exposure. Confounding factors, such as the reason for administering high oxygen levels (e.g., sicker patients needing more oxygen), and patient dropout from studies over time, also complicate interpretation. This paper aims to address these issues by examining the association between longitudinal hyperoxemia exposure and mortality in a large, real-world ICU cohort.

Methods

The researchers conducted a retrospective observational cohort study using data from the National Institute of Health Research Critical Care Health Informatics Collaborative. They analyzed ICU admissions from five UK university hospitals between January 2014 and December 2018.

A novel aspect of their approach was how they quantified "hyperoxemia dose." Instead of relying on single blood gas measurements, they calculated the area under the arterial partial pressure of oxygen (PaO₂) time curve above a threshold of 13.3 kPa (100 mm Hg) for different exposure windows (1, 3, 5, and 7 days). This "dose" was then divided by the duration of the exposure window. This method aimed to capture the cumulative effect of sustained high oxygen levels.

Multivariable logistic regression was used to model the association between this hyperoxemia dose and ICU mortality. They also included other relevant covariates such as age, sex, weight, admission type, respiratory diagnosis, mechanical ventilation status, and APACHE II score. To account for the possibility that patients with no hyperoxemia exposure might be a distinct group, they included an indicator variable for "any hyperoxemia exposure" alongside the continuous dose variable. They also explored potential interaction effects between hyperoxemia and respiratory diagnosis or mechanical ventilation.

Datasets

The study utilized data from the National Institute of Health Research Critical Care Health Informatics Collaborative (NIHR CCHIC). This collaborative dataset comprises ICU admissions from five United Kingdom university hospitals, collected between January 31, 2014, and December 31, 2018. The analysis included a primary cohort of 19,515 patients with at least a 1-day ICU stay, further nested into subsets with longer ICU stays (up to 3, 5, and 7 days) to allow for longer exposure windows.

Key Insights and Ideas

  • Beyond Single Measurements: The study emphasizes the importance of looking at oxygen exposure over time, rather than relying on single blood gas readings, which can be misleading.
  • Defining "Harmful" Oxygen: By setting a threshold of 13.3 kPa (100 mm Hg) for defining hyperoxemia, the researchers aimed to isolate the effects of supraphysiologic oxygen, independent of the need for oxygen to maintain basic physiological function.
  • The "Any Exposure" Effect: The finding that any exposure to hyperoxemia, even for a short duration, was associated with increased mortality is significant, suggesting that even brief periods might carry risk.
  • Dose-Response Weakness: The lack of a clear dose-response relationship is a crucial finding. It suggests that either the relationship is not linear, or the effect plateaus, or that other factors are masking a true dose-dependent effect. This significantly tempers the ability to draw strong causal conclusions.

Key Results

  • Increased Mortality with Hyperoxemia: The study found a statistically significant association between exposure to hyperoxemia and increased ICU mortality across different exposure windows:
  • 0-3 days: Odds Ratio (OR) of 1.35 (95% CI, 1.04–1.74; P=0.02)
    • 0-5 days: OR of 1.5 (95% CI, 1.07–2.13; P=0.02)
    • 0-7 days: OR of 1.74 (95% CI, 1.11–2.72; P=0.02)
    • The association for the 0-1 day window was not statistically significant (OR 1.15, P=0.15).
  • No Dose-Response: Despite the association with "any" hyperoxemia, the "hyperoxemia dose" variable itself did not show a statistically significant association with mortality in most models, and the odds ratios for the dose variable were consistently close to 1.
  • No Differential Effect: There was no evidence to suggest that hyperoxemia had a different impact on mortality based on whether the patient had a primary respiratory diagnosis or was mechanically ventilated.
  • Average Treatment Effect: The estimated average treatment effect (difference in mortality risk between observed hyperoxemia and no hyperoxemia) ranged from 0.4% for the 0-1 day window to 2.7% for the 0-7 day window.

Limitations and Future Work

The authors acknowledge several limitations:

  • Absence of Dose-Response: The lack of a clear dose-response relationship weakens causal interpretation. This could imply the effect is small, plateaus, or that unmeasured confounding is at play.
  • Definition of Hyperoxemia: While the chosen threshold of 13.3 kPa (100 mm Hg) helps isolate supraphysiologic levels, it might reduce the number of cases available for analysis and potentially miss effects at slightly lower, but still elevated, levels.
  • Longitudinal Data Complexity: The study used a method to collapse longitudinal PaO₂ data into a single "dose" measure, which inherently involves a loss of statistical information. They couldn't fully capture the nuances of oxygen exposure profiles over time.
  • Unmeasured Confounding: Despite efforts to control for confounders, the possibility of unmeasured factors influencing both hyperoxemia exposure and mortality remains. For instance, patients requiring frequent transfers or procedures might be sicker, experience more hyperoxemia, and have higher mortality, creating an apparent association.
  • Potential for Interaction Effects: While they found no evidence for differential effects of hyperoxemia based on respiratory diagnosis or mechanical ventilation, the study might have lacked the statistical power to detect such effects.

Future Work: The authors strongly advocate for further experimental research, particularly prospective randomized trials, to definitively elucidate the causal relationship between supraphysiologic oxygen levels and mortality in critically ill patients. They suggest that a small but meaningful reduction in mortality might be achievable by avoiding hyperoxemia, but this requires more robust evidence.

Summary updated

Abstract

<b>Rationale:</b> There is conflicting evidence on harm related to exposure to supraphysiologic Pa<sub>O<sub>2</sub></sub> (hyperoxemia) in critically ill patients.<b>Objectives:</b> To examine the association between longitudinal exposure to hyperoxemia and mortality in patients admitted to ICUs in five United Kingdom university hospitals.<b>Methods:</b> A retrospective cohort of ICU admissions between January 31, 2014, and December 31, 2018, from the National Institute of Health Research Critical Care Health Informatics Collaborative was studied. Multivariable logistic regression modeled death in ICU by exposure to hyperoxemia.<b>Measurements and Main Results:</b> Subsets with oxygen exposure windows of 0 to 1, 0 to 3, 0 to 5, and 0 to 7 days were evaluated, capturing 19,515, 10,525, 6,360, and 4,296 patients, respectively. Hyperoxemia dose was defined as the area between the Pa<sub>O<sub>2</sub></sub> time curve and a boundary of 13.3 kPa (100 mm Hg) divided by the hours of potential exposure (24, 72, 120, or 168 h). An association was found between exposure to hyperoxemia and ICU mortality for exposure windows of 0 to 1 days (odds ratio [OR], 1.15; 95% compatibility interval [CI], 0.95-1.38; <i>P</i> = 0.15), 0 to 3 days (OR 1.35; 95% CI, 1.04-1.74; <i>P</i> = 0.02), 0 to 5 days (OR, 1.5; 95% CI, 1.07-2.13; <i>P</i> = 0.02), and 0 to 7 days (OR, 1.74; 95% CI, 1.11-2.72; <i>P</i> = 0.02). However, a dose-response relationship was not observed. There was no evidence to support a differential effect between hyperoxemia and either a respiratory diagnosis or mechanical ventilation.<b>Conclusions:</b> An association between hyperoxemia and mortality was observed in our large, unselected multicenter cohort. The absence of a dose-response relationship weakens causal interpretation. Further experimental research is warranted to elucidate this important question.
BibTeX
@inproceedings{palmer2019association,
  author = {Edward Palmer and Benjamin Post and Roman Klapaukh and Giampiero Marra and Niall MacCallum and David Brealey and Ari Ercole and Andrew Jones and Simon Ashworth and Peter Watkinson and Richard Beale and Stephen J. Brett and Duncan Young and Claire Black and Aasiyah Rashan and Daniel Martín and Mervyn Singer and Steve Harris},
  title = {{The Association between Supraphysiologic Arterial Oxygen Levels and Mortality in Critically Ill Patients: A Multicenter Observational Cohort Study}},
  booktitle = {American},
  year = {2019},
  doi = {10.1164/rccm.201904-0849oc},
}
Figure