Background

In the early 1900s, Karsner et al. demonstrated that rabbits breathing fractional inspired oxygen concentrations (FiO2) of 0.80 for 7 day developed fibrinous bronchopneumonia [1]. Numerous pre-clinical studies subsequently confirmed this relationship between exposure to similar high oxygen concentrations and lung injury [1,2,3,4,5,6,7,8]. This substantial body of preclinical investigation informed recommendations that high FiO2s be avoided whenever possible [9,10,11,12,13]. Although recent guidelines and study protocols frequently describe the minimal arterial oxygen saturation (SaO2) or pressure (PaO2) levels that should be maintained in patients, when stipulated, FiO2 targets are typically set to not exceed 0.50 to 0.60 when possible [9,10,11,12,13,14,15,16].

Concern has grown that even low oxygen levels that are still greater than room air (that is FiO2s ≤ 0.60 but > 0.21, termed supraphysiologic here) may be harmful in critically ill patients in whom concomitant infection, systemic inflammation, and pre-existing tissue hypoxia could augment susceptibility to oxygen toxicity [17,18,19]. These concerns have been heightened by observational studies suggesting that clinicians frequently administer unneeded low but still supraphysiologic FiO2 levels in intensive care unit (ICU) patients [20, 21]. Clinical studies aiming to define acceptable low oxygen levels in critically ill patients by comparing conservative and liberal oxygen protocols have been at odds. While one systematic review comparing such studies reported that for each percentage point increase in SaO2 in the liberal group increased the relative risk of mortality, another review found no such relationship [22, 23].

Overall, the risks of low but supraphysiologic levels of oxygen remain unclear and continue to be studied clinically. The need for this work has been highlighted by the prolonged oxygen administration many patients with SARS-CoV-2 pneumonia have required [24, 25]. We were, therefore, surprised to find in an informal literature review that in contrast to the many preclinical studies of high FiO2 levels, there appeared to be few such studies examining the risks of low but supraphysiologic FiO2s. To comprehensively explore this literature, we performed a systematic review of in vivo studies in adult animal models that compared the effects of normobaric oxygen administration with FiO2s of ≤ 0.60 and > 0.21 (termed FiO2s ≤ 0.60/ > 0.21 below) vs. FiO2 = 0.21. Our primary focus was how these FiO2s altered outcomes in animals administered infectious or noninfectious inflammatory challenges but studies examining FiO2s ≤ 0.60/ > 0.21 alone were also investigated.

Methods

This systematic review was prepared using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement on guidance for literature review and data extraction (Additional file 2). It was registered with the International Prospective Register of Systematic Reviews on 12/3/2021 (PROSPERO-2021-CRD42021285138).

Literature search and study inclusion

Using search terms and strategies listed in Additional file 3, three authors (S.M., R.D., P.Q.E.) identified relevant studies published in the following databases from inception through 9/30/21: PubMed, EMBASE, and Web of Science. Recovered reports were reviewed for additional references. Studies were included for analysis if they provided data that compared the effects of exposure to FiO2s ≤ 0.60/ > 0.21 vs. an FiO2 = 0.21 for ≥ 24 h in adult in vivo animal models that either included an infectious or noninfectious inflammatory nonoxygen challenge or did not. Inflammatory challenges were defined as infectious if they included a live microbial agent or noninfectious if the challenge was nonliving (e.g., lipopolysaccharide) but typically associated with an inflammatory response. Inflammatory challenges were ones administered either into the lung, peritoneum or blood. Hyperbaric oxygen, neonatal/infant oxygen, and bronchopulmonary dysplasia studies and non-English publications were not included.

Data extraction

Three investigators (S.M., R.D, P.Q.E) independently extracted available data from reports using a standardized extraction form (Additional file 4). These data included: country and year of publication; species, strain, age and weight of animals; level, timing and duration of oxygen exposure; type, dose, route and timing of any non-oxygen inflammatory challenge; and number of animals per study-group. Data for survival and measures of lung or non-pulmonary organ injury and lung or systemic levels of immune response parameters were extracted from studies when presented and compared between study groups. When numbers or percentages of animals living or dead were not reported in studies presenting survival curves, authors of reports were contacted to obtain these data. If these data were still not available, animal numbers were calculated from presented survival curves and the total numbers of animals reportedly studied. For all other data, reported mean and median data with variances and/or levels of significance for differences in measures between study groups were recorded. If data were provided in figures alone, means or medians with variances were determined from the figures, and reported significance levels for group comparisons were recorded. To provide representative findings from exposure to higher FiO2s, similar data from groups exposed to FiO2s > 0.60 in included studies were also recorded.

Quality of evidence

Two reviewers (S.M and P.Q.E) independently assessed included studies for quality of evidence using a modified version of the Systemic Review Centre for Laboratory Animal Experimentation (SYRCLE) grading system [26, 27]. Studies were examined to determine if the following information was provided: a primary outcome; sample size or power calculation; randomization of challenges; confirmation of baseline similarity of study groups (e.g., age, weight); blinding of challenges and outcome assessments; and randomized animal housing.

Statistical methods

For mortality, we used the odds ratios of survival to compare groups (FiO2 ≤ 0.60/ > 0.21 or FiO2 > 0.60 vs. FiO2 = 0.21). Continuous outcomes (e.g., body weight, measures of lung injury, etc.) were analyzed using standardized mean difference (SMD). Studies were combined using a random-effect models [28]. In retrieved studies in which more than one group with an increased FiO2 regimen (e.g., FiO2 = 0.40 and 0.60) was compared to a common FiO2 = 0.21 control group, if the survival results across these groups had heterogeneity (I2) with significance levels p ≥ 0.10, these results were pooled (using random-effect models) to provide a single survival effect for the study. If results from groups within studies differed with a p < 0.10, these groups were included individually in analysis. The effects of increased FiO2’s was then examined across studies employing the same animal type and then across different animal types. Influence of duration of oxygen exposure on lung injury parameters was assessed in meta-regression if there were ≥ 5 studies and/or groups available for analysis. Heterogeneity among studies was assessed using the Q statistic and I2 value. [29]. All analyses were performed using R [30](version 4.2.0) packages meta (version 5.2–0). [31]. Two-sided p values ≤ 0.05 were considered significant.

Results

Of 14,369 retrieved reports and after review of references, 24 studies met inclusion criteria (Additional file 1: Fig. S1) [32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55]. Ten studies compared the effects of FiO2s ≤ 0.60/ > 0.21 to FiO2 = 0.21 in animals challenged with an infectious (n = 6 studies) or noninfectious (n = 4) inflammatory challenge [32,33,34, 37, 38, 43, 45, 49, 51, 52] and 14 studies compared these FiO2s alone [35, 36, 39,40,41,42, 44, 46,47,48, 50, 53,54,55]. The two groups of studies were examined separately. More than 600 reports investigating only FiO2s > 0.60 were excluded.

Studies with an infectious or noninfectious inflammatory challenge

Table 1 summarizes the characteristics of the 10 studies comparing the effects of FiO2s ≤ 0.60/ > 0.21 to an FiO2 = 0.21 in animals also administered an inflammatory challenge. The FiO2s ≤ 0.60/ > 0.21 investigated included 0.40, 0.50 or 0.60 alone in two, three and two studies, respectively, and both 0.40 and 0.60 in two, or 0.27 and 0.60 in one. Five studies included animals exposed to FiO2s > 0.60 (0.70 to 1.0). The longest oxygen exposure periods were 1d in 4 studies, 4d in two studies and 6, 7, 8 and 35d in one study each. No experiment utilized mechanical ventilation. Infectious challenges included cecal ligation and puncture (CLP) in three studies and either intratracheal (IT) Legionella pneumoniae (L. pneumoniae) or IT Klebsiella pneumonia (K. pneumoniae) in one each. One study each administered intraperitoneal (IP) lipopolysaccharide (LPS), IT LPS, IT hydrochloric acid (HCL) or intracardiac (IC) oleic acid. Six studies administered the inflammatory challenge immediately before starting oxygen (0 h) and one study each administered it 4, 12, 18 or 48 h before (Table 1). Three studies included control groups exposed to FiO2s ≤ 0.60/ > 0.21 that were administered a noninflammatory control challenge. None of these studies examined whether the effects of increased FiO2s with or without an inflammatory challenge were additive or synergistic and these groups are not examined further here. Numbers of animals in experimental groups within studies are summarized in Additional file 1: Table S1.

Table 1 Summary of O2 + nonO2 inflammatory challenge study characteristics
Full size table

Seven studies compared the effects of FiO2s ≤ 0.60/ > 0.21 vs. FiO2 = 0.21 on survival in animals administered an inflammatory challenge (Table 1, Fig. 1, Additional file 1: Fig. S2). Survival results for two studies were determined from the average of the range of animals studied and the survival curves provided (see methods). Four studies were conducted in rats, and one each in mice, rabbits or dogs. Two studies examined two FiO2s ≤ 0.60/ > 0.21, one examined the same FiO2 administered for two different time periods, and one examined the same oxygen regimen with two different LPS doses. For studies that examined two FiO2s ≤ 0.60/ > 0.21 or two LPS challenges, there was low heterogeneity across groups within each of these studies, (I2 ≤ 14%, p ≥ 0.28, Additional file 1: Fig. S2) and these groups were, therefore, pooled within studies for analysis (see methods). Across animal types studied, compared to FiO2 = 0.21, FiO2s ≤ 0.60/ > 0.21 had consistent effects on the odds ratio of survival (95%CI) (OR) that overall was on the side of harm but was not significant [0.68 (0.38, 1.23), p = 0.21; I2 = 0%, p = 0.57] (Fig. 1). Notably, in four studies oxygen was administered for only 1d and in one study for only 2 to 4d (Fig. 1). Four studies reported survival in animals administered FiO2s > 0.60. In three of these studies in rats, oxygen exposure had effects on the OR on the side of harm but which was not significant [0.40 (0.11,1.50), p = 0.18; I2 = 25%, p = 0.26)]. Across eight groups investigated in the remaining study in mice, FiO2s > 0.60 decreased the overall OR [0.04 (0.01, 0.15), p < 0.0001; I2 = 34%, p = 0.18] (Additional file 1: Fig. S2). Significant heterogeneity prevented combining these rat and mouse results for studies with FiO2s > 0.60 (I2 = 67%, p = 0.03) (Fig. 1).

Fig. 1
figure 1

Effects of either FiO2s ≤ 0.60 and > O.21 (FiO2 ≤ 0.60/ > O.21) (upper panel) or FiO2s ≥ 0.60 (lower panel) vs. FiO2s = 0.21 (controls) on the odds ratios of survival (95%CIs) (OR) in studies [author (y)] that also administered an infectious or noninfectious inflammatory challenge in animals. Animal type, increased FiO2 level and duration, and inflammatory challenge route and type employed in studies are also shown. Also noted are the numbers of animals surviving and total numbers of animals in oxygen or control groups. Data from individual regimens of oxygen or challenge that were pooled within studies based on nonsignificant heterogeneity (I2 level of significance, p ≥ 0.10) comparing the regimens are shown in Additional file 1: Fig. S2. Overall ORs were pooled across animal type and studies if the significance level for heterogeneity was p ≥ 0.10. While an overall OR could be calculated for FiO2 ≤ 0.60/ > O.21 (I2 = 0%, p = 0.54), this was not possible for FiO2s ≥ 0.60 due to high heterogeneity (I2 = 67%, p = 0.03) comparing rat and mouse. IP intraperitoneal, CLP cecal ligation and puncture, IT intratracheal, IC intracardiac, L. p. Legionella pneumoniae, HCL hydrochloric acid, OA oleic acid

Full size image

Three or four studies compared the effects of one or more regimen of FiO2s ≤ 0.60/ > 0.21 vs. FiO2 = 0.21 on lung injury measures including lung weights, lavage protein concentrations, pathology scores and/or arterial oxygen pressures. The animal types, oxygen regimens, inflammatory challenges and the measures, variances and units provided in studies are presented in Additional file 1: Table S2. These data were used to compare standardized mean differences (95%CI) (SMD) in these measures within studies and across animal types (Figs. 2, 3). Compared to FiO2 = 0.21, FiO2 ≤ 0.60/ > 0.21 increased SMD in lung weights across four studies in trends approaching significance and with low heterogeneity [0.47 (− 0.07,1.00), p = 0.09; I2 = 0%, p = 0.50). One study in mice reported that FiO2 = 0.50 for 4d in mice challenged with IT L. pneumophilia significantly increased lung weight but was not analyzed, because the type of variance was not identified (Additional file 1: Table S2).[45]. Differences in the effects of FiO2 ≤ 0.60/ > 0.21 between the species studied for lung lavage protein (mouse vs. rat, I2 = 43%, p = 0.17) and arterial oxygen measures (mouse vs. rat vs. rabbit vs. dog, I2 = 91%, p < 0.01) prevented estimation of overall SMDs (Figs. 2, 3). Notably, two studies in which FiO2 ≤ 0.60/ > 0.21 had effects on oxygen measures on the side of harm (i.e., favoring control) may have reported arterial oxygen measures in animals receiving some level of oxygen support thereby blunting the possible adverse effects of these FiO2s [34, 43]. While overall lung pathology scores did not differ comparing species (p = 0.36), significant heterogeneity across studies (I2 = 86%, p < 0.01) also prevented estimation of an overall SMD (Fig. 3). For parameters with more than five studies for analysis, meta-regressions analysis did not show a strong relationship [slope (± SE)] between duration of oxygen exposure and the effects of FiO2 ≤ 0.60/ > 0.21 on lung pathology scores [0.067 (± 0.035), p = 0.06; residual heterogeneity I2 = 67%] and showed no relationship with oxygenation [p = 0.49 for the slope; residual heterogeneity I2 = 92%].

Fig. 2
figure 2

Effects of FiO2sT ≤ 0.60 and > O.21 (FiO2 ≤ 0.60/ > O.21) vs. FiO2s = 0.21 (controls) on standardized mean differences (± SD) (SMD) in lung weights and lung lavage protein in studies [author (y)] that also administered animals an infectious or noninfectious inflammatory challenge. Animal type, FiO2 level and duration, inflammatory challenge route and type employed in studies are shown. Data from studies used to calculate SMDs for these parameters and parameter units are shown in Additional file 1: Table S2. Open circles represent results from individual regimens of oxygen or challenge within studies examining more than one regimen, that could be pooled (I2 level of significance, p ≥ 0.10) to report an overall SMD for the study, shown by solid circles. Results shown by solid circles were then used to determine whether SMDs could be pooled across studies in the same species and then across all studies. Also shown are the number of animals (n) in study groups. While an overall SMD could be calculated across studies for lung weights (I2 = 0%, p = 0.50; inverted triangle), differences between mouse and rat studies prevented this for lung protein (I2 = 43%, p = 0.07). CLP cecal ligation and puncture, IT intratracheal, IC intracardiac, K. p. Klebsiella pneumoniae, HCL hydrochloric acid, LPS lipopolysaccharide, OA oleic acid

Full size image
Fig. 3
figure 3

Effects of FiO2s ≤ 0.60 and > O.21 (FiO2 ≤ 0.60/ > O.21) vs. FiO2s = 0.21 (controls) on standardized mean differences (± SD) (SMD) in pathology scores and arterial oxygen levels in studies [author (y)] that also administered animals an infectious or noninfectious inflammatory challenge. Animal type, FiO2 level and duration, and inflammatory challenge route and type employed in studies are shown. Data from studies used to calculate SMDs for these parameters and parameter units are shown in Additional file 1: Table S2. Open circles represent results from individual regimens of oxygen or challenge within studies examining more than one regimen, that could be pooled (I2 level of significance, p ≥ 0.10) to report an overall SMD for the study, shown by solid circles. Results shown by the solid circles were then used to determine whether SMDs could be pooled across studies in the same species and then across all studies. Also shown are the number of animals (n) in study groups. Differences across studies for lung pathology scores (I2 = 86%, p < 0.01) and across species for arterial oxygen levels (I2 = 91%, p < 0.01) prevented estimation of overall SMDs for either parameter. CLP cecal ligation and puncture, IT intratracheal, IC intracardiac, K. p. Klebsiella pneumoniae, HCL hydrochloric acid, LPS lipopolysaccharide, OA oleic acid

Full size image

The effects of FiO2 ≤ 0.60/ > 0.21 on measures of lung injury reported on in only one or two studies were not analyzed but are presented in Additional file 1: Table S2. Measures of lung injury with FiO2s > 0.60 were only reported on in one or two studies (Additional file 1: Table S3).

Lung or systemic immune response measures were reported with one or more regimen of FiO2 > 0.60 and an inflammatory challenge in three studies and with FiO2s ≤ 0.60/ > 0.21 in six studies and are presented in Additional file 1: Tables S3 and S4, respectively. Except for one study in which polymorphonuclear cell depletion reduced oxygen induced lung injury with IT LPS challenge [32], it is difficult to determine how changes in the various immune response measures reported in these studies contributed to the effects of increased FiO2s on outcomes.

Studies without an inflammatory challenge

Table 2 summarizes characteristics of the 14 studies comparing the effects of FiO2 ≤ 0.60/ > 0.21 to an FiO2 = 0.21 and not combined with an inflammatory challenge. The FiO2s ≤ 0.60/ > 0.21 investigated included 0.60 alone in ten, 0.50 alone in three, and 0.30, 0.40, 0.50 and 0.60 in one. The longest oxygen exposure period investigated was 3d in two studies, 7d in three studies, 21d in two studies and 2.5, 3.75, 8, 14, 42 or 90d in one study each. Ten studies included animals exposed to FiO2s > 0.60 (0.65 to 1.0). One study [40] utilized mechanical ventilation, but only after 1w of oxygen exposure and for the purpose of conducting mechanical studies. Numbers of animals in experimental groups within studies are summarized in Additional file 1: Table S5.

Table 2 Summary of O2 only study characteristics
Full size table

Two studies [31, 36] compared survival time in rats exposed to FiO2 = 1.0 immediately after animals had previously been exposed to either FiO2 = 0.21or 0.60 for 7d or FiO2 = 0.21 or 0.50 for 42d. Compared to FiO2 = 0.21, survival time was reduced after prior exposure to FiO2 = 0.60 [median survival time 66 h vs. 48 h, respectively (no IQRs reported, n = 40 animals per group), p < 0.01 as reported)] or FiO2 = 0.50 (mean ± SD survival time, 67.2 ± 1.1 h vs. 55.5 ± 3.3 h, p < 0.001 as reported). One study each in rats or mice noted that all animals survived after exposure to FiO2 = 0.60 for 7d or 3d, respectively [41, 53].

Five studies in rats and one each in mice and hamsters, provided data that could be used to compare SMDs in body weight for FiO2s ≤ 0.60/ > 0.21 vs. FiO2 = 0.21 (Fig. 4, Additional file 1: Table S6). There was substantial heterogeneity across five groups receiving increasingly longer regimens of oxygen in one rat study (I2 = 50%, p = 0.09) and three groups in the hamster study (I2 = 79%, p < 0.01) and these groups could not be pooled within each study, Although FiO2s ≤ 0.60/0.21 did not increase the SMD for body weight significantly in any study or group but decreased it significantly in one study and in two groups in another study, there was significant heterogeneity in these SMDs across studies (I2 = 71%, p < 0.01). Meta-regression showed a negative relationship [slope (SE)] between duration of oxygen exposure and body weight [− 0.036 (± 0.016), p = 0.03] but residual heterogeneity was high (I2 = 70%). Five studies in rats and one in mice provided body weight data in animals administered FiO2 > 0.60 (Fig. 4, Additional file 1: Table S6). One rat study examined two different FiO2s, each for five different time periods, but the results of these 10 groups could not be pooled (I2 = 69%, p < 0.01). While FiO2 > 0.60 decreased the SMD in body weight significantly in 9 individual studies or groups and did not increase it in any, there was significant heterogeneity in its overall effects (I2 = 78%, p < 0.01). There was no significant relationship between duration of exposure to FiO2 > 0.60 and body weight [0.007 (± 0.048), p = 0.89; residual heterogeneity I2 = 82%].

Fig. 4
figure 4

Effects of FiO2s ≤ 0.60 and > O.21 (FiO2 ≤ 0.60/ > O.21) (upper panel) or FiO2 > 0.60 (lower panel) vs. FiO2s = 0.21 (controls) on standardized mean differences (± SD) (SMD) in body weights in grams in studies [author (y)] that did not include an infectious or noninfectious inflammatory challenge. Animal type, increased FiO2 level, and duration employed in studies or groups are shown. Also shown for each study or group are the standardized means (± SD) and numbers of animals (n) used to calculate individual SMDs. Data from studies used to calculate SMDs for these parameters and parameter units are shown in Additional file 1: Table S6. Analysis for two studies (Coursin and Nylen) that included more than one regimen of FiO2 ≤ 0.60/ > O.21 or FiO2 > 0.60 showed that the levels of significance for heterogeneity (I2 ≥ 50%, p ≤ 0.09) across these regimens prevented pooling groups and these groups are shown as solid circles and used in analysis individually here. Heterogeneity across studies and groups also prevented calculating overall SMDs for FiO2 ≤ 0.60/ > O.21 (I2 = 71%, p < 0.01) and FiO2 > 0.60 (I2 = 78%, p < 0.01)

Full size image

Eight and three studies compared the effects of one or more regimen FiO2 ≤ 0.60/ > 0.21 vs. FiO2 = 0.21 on measures of lung weight and lavage protein concentrations, respectively (Fig. 5, Additional file 1: Table S7). For lung weights, there was substantial heterogeneity across three groups in one study with hamsters (I2 = 79%, p < 0.01). Although overall lung weights tended to be increased with FiO2 ≤ 0.60/ > 0.21, heterogeneity across studies and groups was high and significant (I2 = 73%, p < 0.01). Meta-regression did not show a significant relationship between oxygen exposure time and lung weight [− 0.027 (± 0.012), p = 0.03; residual heterogeneity I2 = 70%]. However, for lavage protein, the effects of FiO2 ≤ 0.60/ > 0.21 were consistent across studies and on the side of being increased [0.43 (− 0.23, 1.09), p = 0.20; I2 = 0%, p = 0.40)]. The effects of FiO2 ≤ 0.60/ > 0.21 on other potential measures of lung injury but only reported in one or two studies and not analyzed further are presented in (Additional file 1: Table S7). Two studies reported electron microscopic changes with FiO2 ≤ 0.60/ > 0.21 that were difficult to summarize as to an overall effect [40, 47]. Six and three studies examined the effects of one or more regimen of FiO2 > 0.60 on lung weight and lavage protein levels, respectively (Fig. 5, Additional file 1: Table S8). For lung weights, there was substantial heterogeneity across two groups in a mouse study (I2 = 81%, p = 0.02). Overall, FiO2 > 0.60 produced SMDs for lung weight across studies on the side of being increased, but there was heterogeneity across groups and studies (I2 = 89%, p < 0.01). There was no relationship between duration of oxygen exposure and lung weight [0.06 (± 0.49), p = 0.90; residual heterogeneity I2 = 93%]. FiO2 > 0.60 increased the SMD for lung lavage protein significantly in all three studies measuring it, but again there was substantial heterogeneity for these effects across studies (I2 = 76%, p = 0.02) (Fig. 5). The effects of FiO2 > 0.60 on other lung injury measures reported in only one or two studies are presented in Additional file 1: Table S8.

Fig. 5
figure 5

Effects of FiO2s ≤ 0.60 and > O.21 (FiO2 ≤ 0.60/ > O.21) (Panel A) or FiO2s > 0.60 (Panel B) vs. FiO2s = 0.21 (controls) on standardized mean differences (± SD) (SMD) in two measures of lung injury, lung weight and lung lavage protein concentration, in studies [author (y)] that did not include an infectious or noninfectious inflammatory challenge. Animal type and increased FiO2 level and duration employed in studies and groups are shown. Also shown for each study or group are the standardized means (± SD) and numbers of animals (n) used to calculate individual SMDs. Data from studies used to calculate SMDs for these parameters and parameter units are shown in Additional file 1: Table S7. Analysis for two studies (Nylen and Lagishetty) that included more than one regimen of FiO2 ≤ 0.60/ > O.21 or FiO2 > 0.60 showed that the level of significance for heterogeneity across these regimens prevented pooling groups (I2 ≥ 79%, p < 0.01; I2 ≥ 81%, p < 0.01) and these are shown as solid circles and used in analysis individually here. While it was possible to calculate an overall SMD for lavage protein concentration in studies with FiO2 ≤ 0.60/ > 0.21 (I2 = 0%, p = 0.40), heterogeneity prevented an overall calculation for lung weights for FiO2 ≤ 0.60/ > 0.21 (I2 = 73%, p < 0.01) or for lung weights and lavage protein concentration for FiO2 > 0.60 (I2 = 89%, p < 0.01 and I2 = 76%, p = 0.02, respectively)

Full size image

Lung or systemic immune response measures were reported with FiO2s ≤ 0.60/ > 0.21 and with FiO2 > 0.60 in 8 studies. These measures are available for review in Additional file 1: Tables S8 and S9.

Quality of evidence

Most studies examining oxygen combined with an inflammatory challenge or alone reportedly matched animals based on age and/or weight (Additional file 1: Table S10). However, information regarding sample size calculations, group randomization procedures, blinding to results assessment, animal removal and randomized animal housing was unclear in more than half of each type of study.

Discussion

A large body of evidence from animal studies showed that high FiO2s have injurious pulmonary effects and has been an important basis for avoiding these levels clinically when possible [1,2,3,4,5,6,7,8]. The present literature search retrieved more than 600 animal studies investigating only FiO2 > 0.60. By contrast, despite ongoing questions regarding the potential risks of lower FiO2s in critically ill patients, this search retrieved only 10 studies examining the impact of FiO2 ≤ 0.60 and > 0.21 (FiO2 ≤ 0.60/ > 0.21) on survival or lung injury in animals administered an infectious or noninfectious inflammatory challenge [32,33,34, 37, 38, 43, 45, 49, 51, 52] and 14 studies examining these FiO2s alone [35, 36, 39,40,41,42, 44, 46,47,48, 50, 53,54,55]. The differing study designs and parameters measured and sometimes inconsistent effects across the limited numbers of studies examining similar or related parameters provide no firm conclusion as to the overall effects of FiO2 ≤ 0.60/ > 0.21 in either group of studies. However, findings here do support concerns raised about the potential adverse effects of FiO2s levels that are traditionally considered non-toxic and suggest the types of further animal studies that might help inform clinical use of these FiO2s [17,18,19, 22, 23].

Across seven studies examining FiO2 ≤ 0.60/0.21 with an accompanying inflammatory challenge, increased oxygen had effects on survival that, while not significant, were highly consistent and on the side of harm. However, in four of these studies, animals were exposed to oxygen for only 24 h. Overall, 6 of the 10 studies administered oxygen and observed animals following an inflammatory challenge for ≤ 4d and only five included an infectious inflammatory challenge. Despite these short exposure and observation periods, among the four studies examining lung weights, a measure of lung injury, FiO2s ≤ 0.60/0.21 also had consistent effects that were on the sides of harm. A fifth study that could not be used in analysis reported a similar result [35].

The 14 studies examining FiO2s ≤ 0.60/0.21 without an accompanying inflammatory challenge are less informative since clinically relevant questions relate largely to whether lower FiO2s aggravate existing lung injury. Notably though, exposure to FiO2s = 0.50 and 0.60 alone for 7 days in two studies did decrease survival when animals were subsequently exposed to FiO2 = 1.0. While lung lavage protein concentrations were consistent and on the side of harm in the three studies measuring them, body weight and lung weight changes with FiO2 ≤ 0.60/0.21 were inconsistent across studies. Consistent with the impact animal studies have had on the avoidance of FiO2s > 0.60 clinically, even in the small group of studies investigated here the effects of FiO2s > 0.60 on body and lung weights and lavage protein concentrations while variable, were well on the side of harm.

Taken together though, these two groups of studies support the possibility that FiO2 ≤ 0.60/0.21 can have harmful effects. When combined with another inflammatory challenge, these effects were apparent with as little as 24 h of exposure. However, sensitivity to hyperoxia varies across and within species and how these limited findings apply clinically is unclear [18, 22, 23]. However, results from controlled studies in healthy humans [56] and observational studies in patients [57,58,59,60] were consistent with findings from animal studies regarding the harmful pulmonary effects of FiO2s > 0.60. It is likely that appropriately designed animal studies examining FiO2 ≤ 0.60/0.21 would be informative as well.

To maximize the benefit and minimize the risks of oxygen therapy in patients with pneumonia or other pulmonary injuries, support is routinely titrated based on arterial oxygen saturation levels or blood oxygen levels on arterial blood gases. Mechanical ventilation is also often required along with oxygen therapy to prevent morbidity or mortality from ventilatory failure as opposed to hypoxemia. The ideal animal model to examine the impact of FiO2 levels would be one that included a nonoxygen inflammatory pulmonary challenge and compared the risks and benefits of oxygen therapy titrated over lower or higher ranges. Such a model would also include mechanical ventilation when necessary to prevent confounding by hypoventilation. Such a study would likely require a large animal model and considerable other resources which few laboratories could provide. However, since oxygen titration is standard clinical practice, animal models investigating the risks and benefits of oxygen therapy must attempt to develop the methods to include this type of titration if these models are going to inform clinical practice.

Short of animal models allowing titration of oxygen and ventilatory support, the present review combined with published clinical experience suggest several ways animal models examining the impact of lower FiO2s would be most informative clinically. First, these models should emphasize the type of accompanying pulmonary inflammatory challenge typically seen in patients. For medical intensive care units, this would include either bacterial or viral pneumonia. While two studies here employed a pulmonary infectious challenge (L. pneumoniae and K. pneumoniae), other bacteria such as S. pneumoniae, S. aureus and H. influenza that commonly require ICU admission and oxygen support should be a consideration. Importantly, despite the prevalence of influenza pneumonia and the rapid rise in ICU admissions for SARS-CoV-2 pneumonia, no model examined the impact of FiO2s ≤ 0.60/ > 0.21 on a viral pulmonary challenge. Examination of how lower FiO2s effect viral pulmonary pathology, especially coronaviruses-like SARS-CoV-2, appears essential [25]. Second, models should include oxygen exposure periods long enough to simulate those critically ill patients are exposed to. For patients with severe enough lung injury who require noninvasive or invasive mechanical ventilation, observational studies suggest that these periods should be at least 5–7 days [61,62,63]. Experience in patients with SARS-CoV-2 suggest that these periods should be longer [24]. Third, since several studies presented here suggest that FiO2 ≤ 0.60/ > 0.21 alone may cause some level of pulmonary injury, studies examining how potential injury with these FiO2s interacts with inflammatory pulmonary challenges, i.e., are these effects additive or synergistic, would be most informative. Such studies would require sufficient subject numbers to test for these interactions. Fourth, consensus and consistent use of measures of lung injury considered most informative in studies examining the effects of oxygen therapy on inflammatory lung injury would allow more reliable analysis across studies. Finally, quality of the studies analyzed here was weak. It was unclear in most cases whether studies included sample size calculations, animal randomization procedures, blinding of study results, removal of animals during oxygen exposure periods and randomized animal housing. Future preclinical studies examining the effects of FiO2 ≤ 0.60/ > 0.21 on inflammatory lung injury would be strengthened by providing explicit information about these study design components.

There are potential limitations to this systematic review. First, while our search terms were broad and we included additional reports after reviewing references from studies undergoing full paper review, we may have failed to retrieve all studies meeting our inclusion criteria. Second, sensitivity analysis examining sources of heterogeneity was restricted due to the limited numbers of studies available for review for individual parameters. Furthermore, given the heterogeneity of the models used and the outcomes measured and reported, interpretation and generalizability of these results and analysis are limited. Third, in studies where differing regimens of oxygen could not be pooled for analysis, control groups were employed repetitively. Fourth, for almost all studies analyzed it was not possible to determine how reported changes in immune response measures contributed to outcomes. Finally, while oxygen measures when available were employed as a measure of lung injury, increased FiO2s in animals at the time of measurement may have blunted reductions in this parameter making them less informative.

In conclusion, while the potential impact of lower FiO2s on lung injury in critically ill patients continues to be a concern, few preclinical studies have addressed this question. Those that have, have been limited in terms of the oxygen exposure periods and types of accompanying inflammatory lung injury studied. Given the impact animals studies have had on recommendations regarding the avoidance of toxic higher FiO2s clinically, additional animal studies appear warranted to explore how lower FiO2s effect lung injury in patients.

Take home message While hyperoxia with FiO2 > 0.60 is avoided in the Intensive Care Unit due in large part to animal studies showing harm, less is known about the effects of low but still supraphysiologic (FiO2 ≤ 0.60 but > 0.21) oxygen supplementation. This review highlights the need for more well-designed animal models to evaluate the effects of low but still supraphysiologic (FiO2 ≤ 0.60 but > 0.21) oxygen supplementation with a concurrent inflammatory insult similar to patients seen in the Intensive Care Unit.