Oxygen therapy is one of the primary supportive care for patients with or at risk of developing hypoxic respiratory failure, who are admitted to the emergency department and intensive care units. To date, no clinically useful biomarker for O2 toxicity is available, and data on the effects of hyperoxia on markers of oxidative stress are equivocal.1

Numerous studies have demonstrated that hyperoxia (paO2 > 100 mmHg) can be harmful and is associated with adverse outcomes.2–6

The lungs are the first organs affected by oxygen toxicity. In previous studies lung damage and gut microbiota alteration were observed during exposure of mice to hyperoxia.7,8 Although paO2 reflects the oxygen supply in the artery, it does not necessarily provide information about the burden of oxygen in the lungs, which can be much higher than systemic levels when a high fraction of oxygen (FiO2) is used. At the same time, alveolar hyperoxia may not result in extremely high paO2 levels due to occasional existing alveolar- capillary diffusion disturbances or intrapulmonary shunting.9

Hyperoxia causes oxidative stress due to the formation of reactive oxygen species (ROS) and inflammation in the lungs,4 which may impair the surfactant system, leading to alveolar collapse and a reduction in pulmonary compliance.10 Excessive oxygen administration can lead to hyperoxic vasoconstriction in the pulmonary system and atelectasis11 as well as impair muco-ciliary clearance and the antimicrobial capacity of immune cells, contributing to the development of secondary ventilator-associated pneumonia (VAP).12 In preterm neonates, hyperoxia is associated with chronic diseases such as bronchopulmonary dysplasia.13

Several studies have shown that hyperoxia is also associated with negative effects on other organ systems, such as the central nervous system and on the cardiovascular system.14,15 Some studies report an increased risk of mortality in critically ill patients associated with hyperoxia, with the most pronounced effects seen at extreme levels of paO2.3,6,17,18 An association between hyperoxia and mortality was also suggested in critically ill children beyond the neonatal period.19

In summary, the previous studies still does not provide definitive information about the optimal paO2 levels to target during the ventilation of critically ill patients.

In this retrospective observational study, the incidence of pulmonary events and other coded diagnoses that typically occur in our critically ill patients admitted to a surgical intensive care unit were examined. Additionally, the study investigated the relationship between oxygen therapy and in- hospital mortality.

The objective of the study was to determine whether hyperoxia leads to a higher occurrence of pulmonary events such as bacterial, viral, or fungal pneumonia, atelectasis, acute respiratory failure, pyothorax, and other coded diseases, ultimately resulting in increased in- hospital mortality compared to normoxia. Furthermore, the study assessed the utility of linking paO2 levels to the onset of coded diseases in ICU- patients.

We enrolled 112 patients undergoing elective or emergency surgery, organ transplantation, patients with sepsis or septic shock, post-surgical complications, and hemorrhage that required admission to the ICU.

A total of 189 patients were identified as potential study participants. Out of these, 112 met our inclusion criteria and were eligible for analysis (Fig. 1). All patients were assigned to two groups: the normoxia group consisted of patients with paO2 ≤ 100 mmHg (n = 43), and the “mild” hyperoxia group consisted of patients with paO2 >100 mmHg (n = 69).

53 out of 69 patients in the 'mild' hyperoxia group and 33 out of 43 patients in the normoxia group required surgery and were subsequently treated postoperatively in the ICU. The average duration of surgery was 3.6 h in the hyperoxia group and 3.2 h in the normoxia group. Patients were ventilated invasively with pressure control both in the operating room and subsequently in the intensive care unit, following the current ARDS guidelines. According to our study protocol, only patients who were only invasively ventilated were evaluated in the first 24 h in the intensive care unit, ensuring that ventilatory support was not conducted for at least this period.

Secondary outcomes

There were also no statistically significant between-group differences in the incidence of non- pulmonary new infections and organ dysfunctions (DRG Codes) that occurred during the ICU stay, as shown in Table 2. In patients with hemorrhage (p = 0.53) and polytrauma (p = 0.15) as reasons for admission to the ICU, more patients received oxygen therapy resulting in paO2 > 100 mmHg.

Both study groups had similar age, BMI, lactate levels, and severity of disease scores at admission and during their ICU stay, as shown in the baseline characteristics in Table 1. Notably, the PEEP and FiO2 levels were significantly lower in the group of patients with paO2 >100 mmHg (p = 0.03 and p = 0.01, respectively). The “mild” hyperoxia group spent an average of 1.2 day longer in the ICU (p = 0.59).

Out of 69 patients in the hyperoxia group, 27 (39. 1%) and out of 43 patients in the group with PaO2 ≤ 100 mmHg, 12 (27. 9%) died during their ICU stay or further hospitalization. The Kaplan–Meier survival analysis presented in Fig. 2 shows no statistically significant differences between the groups in survival estimates (p = 0.5428).

Figure 2.

Kaplan–Meier analysis of cumulative survival from group 1 paO2 ≤100 mmHg compared with patients from group 2 paO2 >100 mmHg.

(0.19MB).

In this retrospective observational cohort study, our aim was to examine the relationship between arterial partial pressure of oxygen, a ventilation parameter, and various outcome variables, such as pulmonary events, the onset of other coded diseases, and in-hospital mortality.

Consistent with previous studies,9,22 we used paO2 as a more precise ventilation parameter for the analysis of the oxygen therapy in invasive ventilated ICU patients. This choice was made due to the S-shaped nature of the oxygen- hemoglobin dissociation curve, where even slight differences in SpO2 can lead to significant variations in paO2.

The highest paO2 observed in the “mild” hyperoxia group was 168 mmHg, a value considerably lower than the paO2 thresholds used in prior studies. In those studies, hyperoxia was typically defined as paO2 of 300 mmHg or higher. Notably, such high paO2 levels were associated with either increased mortality18,23 or worse neurological outcomes.24 In our patient cohort, we specifically examined the impact of oxygen therapy on groups with only “mild” hyperoxia, where there were no extreme differences paO2 in values. Our analysis focused on evaluating disparities in pulmonary events and in-hospital mortality within this context.

Our patient groups were homogenous regarding the severity of the disease (median APACHE score 29 vs. 31) and median paO2 levels (96 mmHg vs. 113 mmHg). Due to this homogeneity, it was likely possible to mitigate the potential effect of sicker patients receiving more oxygen, which could lead to an overestimation of the oxygen’s impact on mortality. On the other hand, nowadays, more clinicians generally recognize the potential deleterious effects of prolonged aggressive oxygen therapy, resulting in excessive paO2 levels.16,17 However, current practice still appears to involve clinicians tolerating a certain degree of hyperoxia to maintain a ‘safety buffer’ against hypoxemia.

Our results are consistent with the systematic review, meta-analysis, and trial sequential analysis conducted by Barbateskovic et al., which showed neither beneficial nor harmful effects of higher versus lower oxygenation strategies. Similarly, no evidence was found indicating that a higher oxygenation strategy versus a lower oxygenation strategy had a profound effect on the all-cause mortality.25

As of the current data, there is no consensus or established definitions for cut-off values, including both the lower and upper limits for partial pressure of oxygen, that can differentiate between beneficial and harmful effects for the patient. In this study, which involved a relatively small patient cohort, it was not feasible to establish such cut-off values. PaO2 values between 70 and 170 mmHg are likely appropriate for critically ill ventilated patients, depending on the clinical condition, given the absence of a dependable measurement of tissue oxygenation for clinical application.

Our findings contrast with those of a previous single-center randomized clinical trial,3 wherein lower mortality was noted among patients receiving conservative oxygen therapy compared to those subjected to usual oxygen therapy. The outcomes of this trial might have been influenced by initial baseline disparities among the study groups, such as age, illness severity, and organ failures. Moreover, the study was prematurely terminated after an interim analysis revealed elevated mortality in one of the study groups, possibly leading to an overestimation of effects.5

ICU- patients with hemorrhage and polytrauma as admission causes were over-represented in the hyperoxia group (28% vs. 35% p = 0.53 and 2% vs. 10% p = 0.15, respectively). These patients often receive oxygen therapy, as clinicians tend to administer oxygen more aggressively in such critical and dynamic situations with a potential increased risk of hypoxia aggravated by blood loss. This could potentially provide an explanation for this discrepancy.16

In the hyperoxia group, bacterial pneumonia occurred more often (49 % vs. 42% in the other group, p- value = 0.5593). Based on our understanding of daily practices in our ICU, we would explain this by the frequent tracheal suctioning of secretions or the performance of bronchoscopy to remove secretions. During these procedures, the fraction of inspired oxygen is typically increased to 100%. Consequently, we might have observed falsely elevated values paO2 in this group.

On the other hand, there is evidence that prolonged aggressive oxygen therapy can impair pulmonary innate immunity and the capacity for bacterial phagocytosis. In a study by Six et al., a clear association between hyperoxia (remarkably defined as paO2 > 120 mmHg) at ICU admission and ventilator- associated pneumonia (VAP) as a secondary pulmonary infection was suggested.12

Studies on animal models have shown an association between hyperoxia and bacterial infection. Hyperoxia-induced suppression of macrophage phagocytosis of Pseudomonas Aeruginosa causes an increased susceptibility to bacterial infection, which can lead to higher mortality.26,27 Another study found that hyperoxia increased mortality in mice with Acinobacter Pneumonia; however, the administration of procysteine was able to improve survival by enhancing the phagocytic activity of alveolar macrophages in mice kept under hyperoxic conditions.28

In the recent study by Damiani et al. involving patients with SARS- CoV-2 viral pneumonia, “the prevalence of hyperoxemia was independently associated with a higher risk of ICU mortality, as well as a greater risk of developing VAP”.29 In our patient cohort only cases of herpes simplex virus pneumonia were observed as viral pneumonia, and the overall incidence of this was low.

Interestingly, with regard to non- pulmonary diagnoses, no cases of mesenteric ischemia occurred in either patient group.

Our retrospective study has demonstrated that relating paO2 to outcomes and coded diseases, particularly in surgical ICU patients, is relatively challenging.

Our evaluation certainly has limitations. Firstly, this is a retrospective single-center study, and the patient cohort, consisting of 112 enrolled surgical critical care patients, is relatively small. However, our patient groups were well-balanced and homogenous in terms of the median APACHE score (29 vs. 31) and the median paO2 levels. Furthermore, a comprehensive analysis was conducted regarding the interaction between all relevant DRG-Codes in intensive care medicine, such as sepsis, septic shock, organ transplantation, delirium, prolonged weaning, and so on.

Secondly, in our study, we focused on mortality during the hospital stay without follow-up, unlike other studies, which means that mortality in both groups might be underestimated.9,30 In this study, we explored a detailed list of different diagnoses typically occurring during an ICU stay and their correlation with paO2. Specifically, pulmonary diseases were investigated, including various lung infections.

In this study, we observed no differences in pulmonary events, other coded diseases, or in-hospital mortality related to oxygen therapy in “mild” hyperoxia group compared to the normoxia group. However, further prospective multicentre studies involving a larger number of patients are necessary to conduct a precise protocol- based analysis of arterial partial pressure and other ventilation parameters, and to establish a connection between these factors and the outcomes and coded diseases in critically ill patients.

From now, while the “optimal level” remains unknown and may vary based on specific clinical conditions, it is recommended to target PaO2 values within the normal range.

In other words, carefully titrating PaO2 is advised to prevent both hypoxemia and excessive hyperoxemia.

This retrospective exploratory cohort study received approval from Medical Ethics Commission of the Medical Faculty of Heidelberg University, Heidelberg, Germany (Approval ID: Votum S-357/2020). This study was carried out in compliance with the ethical standards outlined in the latest version of the Helsinki Declaration (2013).30

Requirement for written informed consent and consent for publication was waived by the Ethics Commission.

All data are stored for at least 10 years. The datasets analysed during the current study are available from the corresponding author on reasonable request.

The authors declare that they have no competing interests.

The participating institutions obtained no public or private funding for this work.

LS helped design the work, analysed and interpreted the data, was a major contributor in writing the manuscript, and critically reviewed and revised the final manuscript. DN and FU helped design the work and critically reviewed and revised the final manuscript. TB helped design the work, analysed the data, critically reviewed and revised the final manuscript. MW and AK helped design the work, critically reviewed and revised the final manuscript. MOFK helped design the work, analysed and interpreted the data, drafted the manuscript, and critically reviewed and revised the final manuscript.

During the preparation of this work the authors used ChatGPT in order to improve readability and language. After using this chatbot, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.