Abstract
Rationale: Supplemental oxygen is among the most commonly administered therapies in intensive care units (ICUs). High supplemental oxygen exposure has been associated with harm in observational human studies and animal models, yet no consensus exists regarding which dose and duration of high oxygen constitutes harmful hyperoxemia, and little is known regarding the clinical factors that predict potentially injurious exposure. Objectives: To determine the dose and duration of the arterial partial pressure of oxygen (Pa(O(2))) associated with mortality among mechanically ventilated patients and to identify the clinical factors that predict this exposure. Methods: We performed a retrospective cohort study of patients who received invasive mechanical ventilation at a single academic institution in 2017 and 2018. We used a generalized additive model to visualize the relationship between the measured Pa(O(2)) via arterial blood gas measurements and 30-day mortality. We used multivariable logistic regression to identify patient- and hospital-level factors that predict exposure to harmful hyperoxemia. Results: We analyzed 2,133 patients with 33,310 arterial blood gas measurements obtained during mechanical ventilation. We identified a U-shaped relationship between Pa(O(2)) and mortality, in which Pa(O(2)) was positively correlated with mortality above a threshold of 200 mm Hg. A total of 1,184 patients (55.5%) had at least one Pa(O(2)) measurement above this threshold. If patients spent an entire day exposed to Pa(O(2)) > 200 mm Hg, they had 2.19 (95% confidence interval [CI], 1.33-3.60; P = 0.002) greater odds of 30-day mortality in an adjusted analysis. Any exposure to severe hyperoxemia (Pa(O(2)) > 200 mm Hg) was associated with mortality (odds ratio, 1.29; 95% CI, 1.04-1.59; P = 0.021). The strongest clinical predictor of severe hyperoxemia exposure was the identity of the ICU in which mechanical ventilation was delivered. Conclusions: Exposure to high arterial oxygen concentrations is common among mechanically ventilated patients, and the dose and duration of Pa(O(2)) ⩾ 200 mm Hg is associated with mortality. Severe hyperoxemia is highly variable across ICUs and is far more common in clinical practice than in recent randomized trials of oxygen-targeting strategies. Efforts to minimize this common and injurious exposure are needed.