Classification and effectiveness of different oxygenation goals in mechanically ventilated critically ill patients: network meta-analysis of randomised controlled trials

Xu Zhao, Huaping Xiao, Feng Dai, Daniel Brodie, Lingzhong Meng
European Respiratory Journal 2021 58: 2002928; DOI: 10.1183/13993003.02928-2020

Abstract

Background The optimal oxygenation in mechanically ventilated critically ill patients remains unclear.

Methods We performed a systematic review of randomised controlled trials (RCTs) with the aim to classify oxygenation goals and investigate their relative effectiveness. RCTs investigating different oxygenation goal-directed mechanical ventilation in critically ill adult patients were eligible for the analysis. The trinary classification classified oxygenation goals into conservative (partial pressure of arterial oxygen (PaO2) 55–90 mmHg), moderate (PaO2 90–150 mmHg) and liberal (PaO2 >150 mmHg). The quadruple classification further divided the conservative goal from the trinary classification into far-conservative (PaO2 55–70 mmHg) and conservative (PaO2 70–90 mmHg) goals. The primary outcome was 30-day mortality. The secondary outcomes included intensive care unit, hospital and 90-day mortalities. The effectiveness was estimated by the relative risk and 95% credible interval (CrI) using network meta-analysis and visualised using surface under the cumulative ranking curve (SUCRA) scores and survival curves.

Results We identified eight eligible studies involving 2532 patients. There were no differences between conservative and moderate goals (relative risk 1.08, 95% CrI 0.85–1.36; moderate quality), between moderate and liberal goals (relative risk 0.83, 95% CrI 0.61–1.10; low quality) or between conservative and liberal goals (relative risk 0.89, 95% CrI 0.61–1.30; low quality) based on the trinary classification. There were no differences in secondary outcomes among the different goals. The results were consistent between the trinary and quadruple classifications. The SUCRA scores and survival curves suggested that the moderate goal in the trinary and quadruple classifications and the conservative goal in the quadruple classification may be superior to the liberal and far-conservative goals.

Conclusions Different oxygenation goals do not lead to different mortalities in mechanically ventilated critically ill patients. The potential superiority of maintaining PaO2 in the range 70–150 mmHg remains to be validated.

Abstract

The available evidence does not suggest different effects on mortality among different oxygenation goals in mechanically ventilated critically ill patients. The PaO2 range of 70–150 mmHg is potentially superior but remains to be validated. https://bit.ly/3qtMc1D

Introduction

Mechanical ventilation in an intensive care unit (ICU) is associated with mortality rates up to 30–50%, depending on the aetiology and severity of the disease [1, 2]. Endeavours to improve survival in mechanically ventilated patients are thus of paramount importance. Among the multiple potential causes of these poor outcomes, inappropriate goals of arterial blood oxygenation used during the management of mechanical ventilation may play a role [3].

Multiple randomised controlled trials (RCTs) have been performed to compare the relative effectiveness between different oxygenation goals in mechanically ventilated critically ill patients. Girardis et al. [4] compared the goal of maintaining partial pressure of arterial oxygen (PaO2) at 70–100 mmHg and pulse oxygen saturation (SpO2) at 94–98% with the goal of maintaining PaO2 up to 150 mmHg and SpO2 at 97–100%. They showed that the lower oxygenation goal was associated with a lower ICU mortality (11.6% versus 20.2%). In comparison, Barrot et al. [5] compared the goal of maintaining PaO2 at 55–70 mmHg and SpO2 at 88–92% with the goal of maintaining PaO2 at 90–105 mmHg and SpO2 ≥96%. They found that the lower oxygenation goal was associated with a higher ICU mortality (36.4% versus 26.5%). Not only are these results contrasting, but the goals used by these two studies were also different. While the higher oxygenation goals utilised by these two studies were similar, the lower oxygenation goals were PaO2 70–100 mmHg [4] and 55–70 mmHg [5], respectively, which do not even overlap. These two studies highlight the importance of standardising oxygenation goals for the purposes of both research and clinical practice.

A recent systematic review concluded that in acutely ill adults, liberal oxygen therapy increases mortality and supplemental oxygen might become unfavourable if SpO2 is above the range of 94–96% [6]. In contrast, another recent systematic review was unable to draw conclusions on the relative effectiveness between higher and lower fractions of inspired oxygen (FIO2) or targets of arterial oxygenation in adult ICU patients [7]. Due to the heterogeneity in patient characteristics, methods of oxygen therapy and intervention end-points, it is difficult to interpret the results of these systematic reviews with pairwise meta-analysis. To date, there is no network meta-analysis investigating the effectiveness of different oxygen therapies.

It is our hypothesis that there is an optimal goal of arterial blood oxygenation in mechanically ventilated critically ill patients. The aims of this systematic review of RCTs are to classify oxygenation goals and investigate their relative effectiveness in terms of mortalities in mechanically ventilated ICU patients using network meta-analyses and the surface under the cumulative ranking curve (SUCRA).

Methods

Protocol and registration

We designed and wrote the study according to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) Extension Statement for Reporting of Systematic Reviews Incorporating Network Meta-analyses of Health Care Interventions (PRISMA-NMA) guidelines [8]. The study was prospectively registered in the International Prospective Register of Systematic Reviews (PROSPERO: CRD42020180392).

Eligibility criteria

RCTs comparing the relative effectiveness between different oxygenation goals in mechanically ventilated critically ill patients were eligible. The inclusion criteria included 1) adult patients (≥18 years old); 2) ICU setting; 3) >50% of study participants received mechanical ventilation; 3) comparison between different oxygenation goals defined by PaO2, arterial oxygen saturation (SaO2) or SpO2; 4) mechanical ventilation management guided by different oxygenation goals for at least 24 h; 5) actual oxygenation levels reported; and 6) mortality reported. We considered studies which, although comparing different FIO2 management methods, reported the actual oxygenation levels. We excluded studies that were performed intra-operatively on surgical patients or during the resuscitation of cardiopulmonary arrested patients. The primary outcome was 30-day (including 28-day) mortality. The secondary outcomes included ICU, hospital and 90-day mortalities. There were no restrictions on publication year or language. Both full-text articles and abstracts were eligible.

Information sources and search

We systematically searched Ovid MEDLINE, Ovid Embase, Web of Science and the Cochrane database from inception to 17 April 2020 (search strategy included in the supplementary material). The reference lists of all relevant articles were manually screened to supplement the systematic search.

Study selection and data collection

We used EndNote version 8.0 (Clarivate, Philadelphia, PA, USA) for study deduplication and selection. Two investigators (X.Z. and H.X.) independently screened all deduplicated titles and abstracts derived from the systematic search, evaluated the full candidate articles to determine their eligibility, and performed data extraction using a predesigned data form. Between-investigator disagreements were resolved via team discussion.

Data items

The data items extracted from each eligible study were: 1) authors and year of publication; 2) study design, patient characteristics and number of patients receiving mechanical ventilation; 3) goals of oxygenation defined by the protocol and the actual oxygenation levels; and 4) outcome measures and results.

Risk of bias of individual studies

The Cochrane Collaboration's tool was used to assess the risk of bias in each study [9]. A study was rated as having a high risk of bias overall if one or more domains were rated as having a high risk of bias. A study was rated as having an unclear risk of bias overall if one or more domains were rated as having an unclear risk of bias while the other domains were rated as having a low risk of bias. Otherwise, a study was rated as having a low risk of bias.

Network geometry

We constructed network geometries to visualise the comparisons between the different oxygenation goals. The oxygenation goals were classified per the following trinary classification system: 1) conservative (defined as PaO2 55–90 mmHg and SaO2/SpO2 88–97%), 2) moderate (defined as PaO2 90–150 mmHg and SaO2/SpO2 97–100%) and 3) liberal (defined as PaO2 >150 mmHg) (figure 1a). Accordingly, the network geometry had three nodes corresponding to the different oxygenation goals. We used the actual oxygenation level, instead of the targeted oxygenation level, to determine the oxygenation goal's class for each randomisation group of individual studies. Each randomisation group was assigned to a node according to its oxygenation goal classification. The size of the node was determined by the total number of patients. Different nodes were connected by edges, with the size of the edge determined by the total number of direct comparisons.

FIGURE 1
FIGURE 1

Classification of oxygenation goals: a) trinary classification and b) quadruple classification. SaO2: arterial oxygen saturation; PaO2: partial pressure of arterial oxygen. The conservative goal in the a) trinary classification is further divided into conservative and far-conservative goals in the b) quadruple classification. The oxyhaemoglobin dissociation curve was based on the equation developed by Severinghaus [53]: SaO2=(23 400×(PaO23+150×PaO2)−1+1)−1.

Summary measures

We used the relative risk, 95% confidence interval (CI) and 95% credible interval (CrI) to measure the relative effectiveness between the different oxygenation goals. The efficacy hierarchy was visualised using the SUCRA score, which is a metric used to evaluate which treatment in a network is likely to be the most effective [10].

Planned methods of analysis

For the network meta-analysis, the pooled relative risk for a given comparison was based on direct evidence derived from individual studies and indirect evidence derived from the network. We used a Bayesian hierarchical random effects model with a binomial likelihood and log link function to simulate the probability of events [11]. The pooled estimates were derived using the Markov Chain Monte Carlo method. Three chains of 100 000 iterations were used after a burn-in period of 50 000 iterations, in which initial iterations were discarded to ensure that the final estimates were based on stable posterior sampling. Convergence was assessed using the Brooks–Gelman–Rubin statistic. The network meta-analysis was performed using the R package “gemtc” in R version 3.5.3 (R Foundation for Statistical Computing, Vienna, Austria) [12].

For the pairwise meta-analysis, we used fixed effects models if the total number of studies was less than five and random effects models otherwise using the R package “metafor” [13]. We chose fixed effects models because if fewer than five studies are included in a meta-analysis, the between-study variance cannot be estimated reliably using random effects models [14]. The I2 statistic was calculated to measure the heterogeneity of the included studies. Trial sequential analysis was used to calculate the required information size to control for type I (false-positive) and type II (false-negative) errors [15]. TSA software version 0.9.5.10 beta (Copenhagen Trial Unit, Copenhagen, Denmark) was used for the trial sequential analysis [16, 17].

Assessment of inconsistency

The node splitting method, separating the evidence for each comparison into direct and indirect evidence, was used to appraise the inconsistency of the results [18].

Publication bias across studies

Funnel plots and Egger's test were used to analyse the potential publication bias for direct comparisons based on three or more studies [19].

Quality of evidence

The quality of evidence for each network estimate was assessed according to the GRADE (Grading of Recommendations Assessment, Development and Evaluation) guidelines, which appraise the quality of a body of evidence on the basis of study limitations, imprecision, inconsistency, indirectness and publication bias for the targeted outcome [20].

Additional analysis

We investigated the relative effectiveness of different oxygenation goals using a quadruple classification system, which serves as a sensitivity test for the trinary classification system. The oxygenation goals were classified as 1) far-conservative (defined as PaO2 55–70 mmHg and SaO2/SpO2 88–94%), 2) conservative (defined as PaO2 70–90 mmHg and SaO2/SpO2 94–97%), 3) moderate (defined as PaO2 90–150 mmHg and SaO2/SpO2 97–100%) and 4) liberal (defined as PaO2 >150 mmHg) (figure 1b). This quadruple classification divided the conservative goal (i.e. PaO2 55–90 mmHg) in the trinary classification into far-conservative (i.e. PaO2 55–70 mmHg) and conservative (i.e. PaO2 70–90 mmHg) goals. In the quadruple classification, the far-conservative goal was consistent with the goal used in the RCTs performed by Barrot et al. [5] and Panwar et al. [21], while the conservative goal was consistent with the goal used by the RCTs performed by Mackle et al. [22] and Girardis et al. [4].

We performed additional meta-analysis based on the patient-level time-to-event data to corroborate the primary analysis based on the aggregate mortality outcome data, similar to the previous report by Chu et al. [6]. The patient-level data were extracted from publications that reported Kaplan–Meier curves using DigitizeIt software (I. Bormann, Braunschweig, Germany) and the algorithm described by Guyot et al. [23]. New Kaplan–Meier curves corresponding to the different oxygenation goals defined by the trinary or quadruple classification were constructed using pooled extracted data [24]. The log-rank test was used to compare the survival distributions of these samples. Cox regression models with the study treated as a random effects variable were fitted to compare the effects of different oxygenation goals on mortality, with the relative effectiveness measured by the hazard ratio (HR) with 95% CI.

Results

Study selection

The study selection process and results are detailed in figure 2. We identified eight eligible studies published between 2014 and 2020 [4, 5, 21, 22, 25–28].

FIGURE 2
FIGURE 2

Study identification and selection flow diagram.

Characteristics of individual studies

Of the eligible studies, six were multicentre trials [5, 21, 22, 26–28], while two were single-centre trials (table 1) [4, 25]. One was a feasibility study [27] and three were pilot studies [21, 25, 26]. Three studies were terminated early due to safety concerns [27], safety and futility concerns [5], and earthquake-related slow enrolment [4], respectively. Three studies used 28-day mortality as the primary outcome [5, 25, 28], one study used ICU mortality as the primary outcome [4], while the remaining four studies used a primary outcome that was not a form of mortality [21, 22, 26, 27]. Five studies reported 30-day mortality [5, 22, 25, 26, 28], four studies reported 90-day mortality [5, 21, 22, 28], five studies reported ICU mortality [4, 5, 21, 22, 25] and three studies reported hospital mortality [4, 22, 27] (table 2). Six studies found no between-group difference in mortality [21, 22, 25–28], while one study found reduced ICU and hospital mortalities in patients treated with a lower oxygenation goal [4], and one study found reduced 90-day mortality in patients treated with a higher oxygenation goal [5]. Details of interventions reported by individual studies are presented in supplementary table E1.

View this table:
TABLE 1

Characteristics of individual studies with oxygenation goals classified

View this table:
TABLE 2

Outcome measures and results of individual studies

Risk of bias of individual studies

The risks of bias of individual studies are presented in supplementary table E2. Four studies had a low risk of bias [5, 21, 22, 26] and four studies had an unclear risk of bias [4, 25, 27, 28]. The pooled risks of bias are presented in supplementary figure E1.

Patient characteristics

The eight eligible studies involved a total of 2532 ICU patients (ranging from 17 to 965 patients). One study involved patients with septic shock [28], one study involved patients with acute respiratory distress syndrome [5], two studies involved patients suffering from out-of-hospital cardiac arrest [26, 27] and four studies involved patients with diverse critical conditions [4, 21, 22, 25]. The median or mean age of participants was in the range 58–67 years. The range of male participants was 57–94%. The participants in six studies were all mechanically ventilated [5, 21, 22, 26–28], while 67% of the participants in one study [4] and 84% of the participants in another study [25] were mechanically ventilated. Four studies reported baseline PaO2/FIO2 ratios, ranging from 117 to 259 mmHg [5, 21, 22, 28]. More details of inclusion and exclusion criteria, respiratory diseases, and comorbidities reported by individual studies are presented in supplementary table E1.

Summary of classification

Based on the trinary classification, the conservative goal was investigated by six studies [4, 5, 21, 22, 25, 27], the moderate goal by all eight studies [4, 5, 21, 22, 25–28] and the liberal goal by two studies [26, 28]. Based on the quadruple classification, the far-conservative goal was investigated by two studies [5, 21], the conservative goal by four studies [4, 22, 25, 27], and the moderate and liberal goals by the same studies as the trinary classification [26, 28].

Summary of the network geometry

The five studies that reported 30-day mortality had a total of 10 randomisation groups, which were assigned to three nodes corresponding to the different oxygenation goals (figure 3a). The four studies that reported 90-day mortality had a total of eight randomisation groups, which were assigned to three nodes corresponding to the different oxygenation goals (supplementary figure E2a). We were unable to construct a network geometry for the five studies that reported ICU mortality and for the three studies that reported hospital mortality because these studies only compared the conservative goal with the moderate goal.

FIGURE 3
FIGURE 3

Effects of different oxygenation goal-directed mechanical ventilation management methods on 30-day mortality based on the trinary classification: a) network geometry, b) results of network meta-analysis and c) surface under the cumulative ranking curve (SUCRA) scores. CrI: credible interval.

Synthesis of the results

The results based on the trinary classification are presented in table 3, figure 3b and supplementary figure E2b. The 30-day mortality was 31.3% (214 out of 683), 30.7% (299 out of 975) and 40.9% (113 out of 276) for the conservative, moderate and liberal goals, respectively, based on a simple pooling of the data from five studies (table 2). There were no differences between the conservative and moderate goals (relative risk 1.08, 95% CrI 0.85–1.36; moderate quality), between the moderate and liberal goals (relative risk 0.83, 95% CrI 0.61–1.10; low quality) or between the conservative and liberal goals (relative risk 0.89, 95% CrI 0.61–1.30; low quality). The 90-day mortality was 36.7% (231 out of 630), 34.8% (296 out of 850) and 47.9% (104 out of 217) for the conservative, moderate and liberal goals, respectively, based on a simple pooling of the data from four studies (table 2). There were no differences between the conservative and moderate goals (relative risk 1.17, 95% CrI 0.88–1.62; moderate quality), between the moderate and liberal goals (relative risk 0.86, 95% CrI 0.54–1.40; very low quality) or between the conservative and liberal goals (relative risk 1.00, 95% CrI 0.59–1.80; very low quality). There were no differences between the conservative and moderate goals for ICU mortality (relative risk 0.93, 95% CI 0.69–1.26; low quality) and hospital mortality (relative risk 0.97, 95% CI 0.83–1.13; very low quality) based on pairwise meta-analysis. The moderate goal was likely the most effective for reducing 30-day mortality (SUCRA 82.5%; figure 3c) and 90-day mortality (SUCRA 83.9%; supplementary figure E2c).

View this table:
TABLE 3

Estimates of effects and quality of evidence based on the trinary classification

Inconsistency, publication bias across studies and quality of evidence

There was no need to explore the inconsistency because no comparison had both direct and indirect evidence. No publication bias was found (supplementary table E3 and supplementary figure E3). The quality of evidence is presented in table 3 and supplementary table E3.

Results of the additional analyses

The results of the analyses based on the quadruple classification are presented in supplementary figures E4–E7 and supplementary tables E4 and E5. There were no differences among the far-conservative, conservative, moderate and liberal goals for the 30-day, ICU, hospital and 90-day mortalities. The likely most effective goals for 30-day mortality reduction were the moderate (SUCRA 75.2%) and conservative (SUCRA 73.1%) goals. The likely most effective goals for 90-day mortality reduction were the moderate (SUCRA 77.7%) and conservative (SUCRA 59.9%) goals.

Six studies reported Kaplan–Meier curves involving 2351 patients. As per the trinary classification, five studies reported survival data related to conservative goal-directed care [4, 5, 21, 22, 25], six studies reported survival data related to moderate goal-directed care [4, 5, 21, 22, 25, 28] and only one study reported survival data related to liberal goal-directed care [28]. The survival analysis based on the extracted patient-level data from these six studies showed that the different oxygenation goal-directed invasive mechanical ventilation management methods might have different levels of effectiveness in terms of mortality (figure 4a). The conservative and moderate goals, although having comparable effectiveness, might both be superior to the liberal goal for mortality reduction. The survival analysis based on the quadruple classification also suggested that different oxygenation goals might have different levels of effectiveness in terms of mortality, and the conservative and moderate goals might both be superior to the liberal and far-conservative goals (figure 4b).

FIGURE 4
FIGURE 4

Survival probability for patients treated with different oxygenation goal-directed care based on the a) trinary and b) quadruple classifications. HR: hazard ratio; CI: confidence interval. The Kaplan–Meier curves were constructed using the pooled patient-level data extracted from six randomised controlled trials. We used the survival data up to 90 days when available from the following studies: one study up to 28 days [25], one study up to 60 days [4], one study up to 80 days [21] and three studies up to 90 days [5, 22, 28].

Because some patients were not intubated in two studies [4, 29], we performed additional analyses excluding these two studies. The results of the analyses with and without these two studies were consistent (supplementary figure E8). The trial sequential analysis showed that the z-curves neither reached the required sample size nor surpassed the O'Brien–Fleming monitoring boundaries, suggesting the false-negative possibility of the results (figure 5).

FIGURE 5
FIGURE 5

Results of trial sequential analysis based on the trinary classification. a) Conservative versus moderate goal for 30-day mortality, b) moderate versus liberal goal for 30-day mortality, c) conservative versus moderate goal for 90-day mortality, d) conservative versus moderate goal for intensive care unit mortality and e) conservative versus moderate goal for hospital mortality. The abscissa indicates the number of patients (linear scale), while the ordinate indicates the cumulative z-score. The z-score is the test statistic, with a |z| of 1.96 corresponding to a p-value of 0.05, as indicated by the solid brown horizontal lines. A higher z-score corresponds to a lower p-value. The z-curve is indicated by the blue graph line. The red vertical dashed line on the right side indicates the required information size, which was calculated based on the pooled event rates in patients treated by a higher oxygenation goal, two-sided α=0.05, β=0.20 and a 20% relative risk reduction, with adjustments for heterogeneity. The O'Brien–Fleming monitoring boundaries for benefit/harm (top or bottom left-hand corners) and futility (inside the brown horizontal lines) are indicated with red dashed lines.

Discussion

Summary of the evidence

We identified a total of eight eligible RCTs investigating the clinical effectiveness of different oxygenation goal-directed invasive mechanical ventilation management strategies in critically ill adult patients. We used the actual, instead of the targeted, levels of oxygenation to classify the different goals used by the individual studies. The primary analysis was based on the trinary classification system. The quadruple classification system was used in the sensitivity analysis. The network meta-analysis showed that different oxygenation goals did not lead to different mortalities in mechanically ventilated critically ill patients. The negative results may be secondary to the inadequate sample size, as suggested by the trial sequential analysis. The SUCRA ranking suggested that the likely most effective oxygenation goals are the moderate goal (PaO2 90–150 mmHg) as per the trinary classification and the moderate (PaO2 90–150 mmHg) and conservative (PaO2 70–90 mmHg) goals as per the quadruple classification. The survival analysis suggested that the conservative goal based on the quadruple classification and the moderate goal based on the trinary and quadruple classification, although likely comparable, might both be superior to the liberal goal and the far-conservative goal. Our findings should be interpreted with caution because the quality of only two bodies of evidence is moderate while the rest is low and very low.

Comparisons with previous studies

The results of our study are in contrast to a recent systematic review with pairwise meta-analysis of RCTs reported by Chu et al. [6]. Their review concluded that there is an increased mortality in acutely ill patients receiving liberal oxygen therapy. However, there are noticeable differences between Chu et al.’s [6] review and our study. First, their review included studies that were performed in patients with a diversity of diagnoses, including acute stroke [30–36], acute myocardial infarction [37–42], surgery [43–45], resuscitation [46], septic shock [28], traumatic brain injury [47], post-out-of-hospital cardiac arrest [27] and unspecified critical condition [4, 21]. Most patients with acute stroke or acute myocardial infarction or having surgery do not have hypoxaemia. Patients in a critical condition may or may not have hypoxaemia and may have different severities of hypoxaemia if they are hypoxaemic. Moreover, the patients in most of the included studies were not endotracheally intubated [30–43], whereas in the rest of the studies they were intubated [4, 21, 27, 28, 44–47]. Second, Chu et al.’s [6] review used a binary relative approach to define different oxygen therapies, i.e. a treatment targeting a higher FIO2, PaO2, SaO2 or SpO2 was defined as liberal oxygen therapy while a treatment targeting a lower value was considered conservative oxygen therapy. The majority of the included studies in their review used FIO2 to differentiate liberal and conservative oxygen therapies. The binary relative approach can cause confusion due to the overlap between the different definitions from different studies. For example, one study used FIO2 of 0.30 and 0.21 in the liberal and conservative groups, respectively [30], while another study used FIO2 of 0.80 and 0.30 in the liberal and conservative groups, respectively [44]. Evidently, the FIO2 used in the liberal group in the former study was the same as the FIO2 used in the conservative group in the latter study. Moreover, the studies included in Chu et al.’s [6] review used different methods of oxygen delivery, with most studies using either nasal prongs [30, 33, 35, 42] or a face mask [31, 32, 34, 36–41, 43], while the remainder used invasive mechanical ventilation [4, 21, 27, 28, 44–47]. Third, Chu et al.’s [6] review did not define or assess the end-points of intervention. The oxygenation goal defined by a prespecified PaO2 or SaO2 value/range should be clarified if the purpose of oxygen therapy is to impact arterial blood oxygenation and not to simply target a specific FIO2. FIO2 and PaO2 are related but do not follow a linear relationship. How a given change in FIO2 translates into changes in PaO2 is complicated, as exemplified by the concept of the PaO2/FIO2 ratio that is largely dependent on the severity of the lung disease or gas exchange abnormality. In contrast, our current study only included ICU patients who received oxygenation goal-directed invasive mechanical ventilation, with the oxygenation goal classified by the trinary or quadruple classification system.

An association between a higher oxygenation goal and hazardous effects was suggested by some cohort studies. A large-scale retrospective study found that among patients requiring oxygen therapy, the lowest hospital mortality was observed at SpO2 in the range 94–98% and this finding was consistent across subgroup analyses [48]. A different retrospective study found that PaO2 >120 mmHg was associated with increased ICU mortality [49]. The difficulty of confounding control in cohort studies is one of the major limitations of these studies. In contrast, we only included RCTs in our study. Most importantly, the inclusion in our study of the two RCTs published in 2020 tilted the balance towards a higher oxygenation goal [5, 22]. Barrot et al. [5] showed that 28-day mortality was higher in patients treated with a conservative goal (34.3%) than a moderate goal (26.5%). Mackle et al. [22] showed that 30-day mortality was slightly higher in patients treated with a conservative goal (31.8%) than a moderate goal (29.1%) (data shared by the ICU-ROX Investigators). These results are in stark contrast to the results of most studies published before 2020. The chronological discrepancy highlights the complexity of this topic.

The goal of oxygenation

The current oxygenation goal widely adopted for patients with acute respiratory distress syndrome is PaO2 55–80 mmHg or SpO2 88–95%. This goal became a well-recognised standard practice following the publication of the ARDSNet ARMA trial in 2000 [50]. However, the study was designed to compare the effectiveness between lower and higher tidal volumes, not to validate the goal of oxygenation per se, because both groups in the study targeted the same goal. Nonetheless, the study does illustrate the early attention on the importance of determining the rightful oxygenation goal.

It is highlighted by our study that simply defining an oxygenation goal as higher or lower is not enough. We need to standardise different oxygenation goals using precise terms for the purposes of both research and clinical practice. The trinary and quadruple classifications we propose in this study deserve discussion. First, no matter how hard we try, it is impossible to define the best dividing lines that are in concordance with the criteria used by the different studies. This matter is complicated by the fact that, sometimes, discrepancies exist between the targeted goals and the actual levels. Furthermore, the actual levels seen in a group of patients often follow a wide range of distribution and overlap between groups, especially when the targeted goals have a narrow separation. We chose PaO2 90 mmHg to divide conservative and moderate goals in the trinary classification because it is not only in concordance with the most recent high-quality study [22], but it is also clinically acceptable to most practitioners. The reason we propose a quadruple classification is to use it as a sensitivity test in case the narrow range defined by PaO2 70–90 mmHg is more favourable. The reason we primarily used PaO2, instead of SaO2, to define different goals was better granularity of PaO2 in differentiating different oxygenation levels at a higher oxygenation range. However, we retained SaO2/SpO2 in the classification because they are the commonly used parameters in clinical practice.

Despite the discrepancy between our work and the previous work, there is highly likely an optimal range of oxygenation that is associated with the most favourable outcomes. This optimal range is neither too low nor too high; the question remains what this range is. If this range was relatively wide and the goals investigated by some previous studies were within it, this might have caused the failure in detecting a difference in these studies. The quality of most previous studies might also be low, as evidenced by the fact that seven of the eight studies included in our analysis were either pilot studies or were terminated early. Only the study performed by Mackle et al. [22] was regarded as a complete formal study. The currently ongoing Handling Oxygenation Targets in the Intensive Care Unit (HOT-ICU) trial (ClinicalTrials.gov: NCT03174002) is expected to provide more insights into this topic [51]. At present, it is probably prudent to adopt the PaO2 range 70–150 mmHg in mechanically ventilated critically ill patients based on the overall evidence.

Limitations

Our study has some limitations. First, the trinary and quadruple classifications have not been previously validated, although they are supported by the overall studies and are in concordance with clinical practice. Second, the quality of our meta-analysis might be adversely affected by the quality of the studies included because most of these studies were pilot studies or were terminated early. Third, the lack of studies directly comparing a conservative goal and a liberal goal created a gap in the network we constructed, which may have made our analysis less robust. Fourth, our approach of using the actual instead of the targeted oxygenation levels required judgement based on the data provided by publications. Fifth, our study included critically ill patients with a diversity of diagnoses, which therefore requires caution in generalising our findings. Sixth, the protocols of oxygenation goal-directed invasive mechanical ventilation used by different studies might have varied in ventilation mode, setting, timing and duration, which may have introduced bias into our study.

Conclusions

Our study provides the first network meta-analysis of RCTs investigating the clinical effectiveness between different oxygenation goals in mechanically ventilated critically ill adult patients. Different oxygenation goals did not lead to differences in mortality. However, this negative finding may be secondary to inadequate sample size. The SUCRA ranking and the survival analysis suggested that the moderate goal (PaO2 90–150 mmHg) based on the trinary and quadruple classifications and the conservative goal (PaO2 70–90 mmHg) based on the quadruple classification are likely to be the most effective. The inclusion of the two recent RCTs published in 2020 tilted the balance towards the moderate goal. The conflicting evidence calls for further research on this topic. At this time, it may be prudent to maintain PaO2 in the range 70–150 mmHg in mechanically ventilated critically ill patients.

Supplementary material

Supplementary Material

Please note: supplementary material is not edited by the Editorial Office, and is uploaded as it has been supplied by the author.

Supplementary material ERJ-02928-2020.SUPPLEMENT

Shareable PDF

Supplementary Material

This one-page PDF can be shared freely online.

Shareable PDF ERJ-02928-2020.Shareable

Acknowledgements

We thank the ICU-ROX Investigators for sharing their mortality results assessed under different criteria.

Footnotes

  • This article has supplementary material available from erj.ersjournals.com

  • This article has an editorial commentary: https://doi.org/10.1183/13993003.01023-2021

  • This study is registered at PROSPERO with identifier number CRD42020180392.

  • Author contributions: Concept and design: L. Meng, X. Zhao and H. Xiao. Acquisition, analysis and/or interpretation of data: all authors. Drafting of the manuscript: X. Zhao and L. Meng. Critical revision of the manuscript for important intellectual content: all authors. Statistical analysis: X. Zhao and F. Dai. Administrative, technical or material support: F. Dai and L. Meng. Supervision: L. Meng.

  • Conflict of interest: X. Zhao has nothing to disclose.

  • Conflict of interest: H. Xiao has nothing to disclose.

  • Conflict of interest: F. Dai has nothing to disclose.

  • Conflict of interest: D. Brodie reports personal fees from ALung Technologies (research support and previously medical advisory board membership), Baxter, BREETHE, Xenios and Hemovent (for medical advisory boards), outside the submitted work.

  • Conflict of interest: L. Meng reports a consultancy fee from Edwards Lifesciences.

  • Support statement: Support was provided by institutional and/or departmental sources. The funders/sponsors were not involved in the design or execution of the study or in the analysis and interpretation of the data.

  • Received July 27, 2020.
  • Accepted February 12, 2021.

References

  1. ↵
    1. Bellani G,
    2. Laffey JG,
    3. Pham T, et al.
    Epidemiology, patterns of care, and mortality for patients with acute respiratory distress syndrome in intensive care units in 50 countries. JAMA 2016; 315: 788–800. doi:10.1001/jama.2016.0291
    OpenUrlCrossRefPubMed
  2. ↵
    1. Luhr OR,
    2. Karlsson M,
    3. Thorsteinsson A, et al.
    The impact of respiratory variables on mortality in non-ARDS and ARDS patients requiring mechanical ventilation. Intensive Care Med 2000; 26: 508–517. doi:10.1007/s001340051197
    OpenUrlCrossRefPubMedWeb of Science
  3. ↵
    1. Angus DC
    . Oxygen therapy for the critically ill. Mass Med Soc 2020; 382: 1054–1056. doi:10.1056/NEJMe2000800
    OpenUrl
  4. ↵
    1. Girardis M,
    2. Busani S,
    3. Damiani E, et al.
    Effect of conservative vs conventional oxygen therapy on mortality among patients in an intensive care unit: the Oxygen-ICU randomized clinical trial. JAMA 2016; 316: 1583–1589. doi:10.1001/jama.2016.11993
    OpenUrlCrossRefPubMed
  5. ↵
    1. Barrot L,
    2. Asfar P,
    3. Mauny F, et al.
    Liberal or conservative oxygen therapy for acute respiratory distress syndrome. N Engl J Med 2020; 382: 999–1008. doi:10.1056/NEJMoa1916431
    OpenUrlCrossRefPubMed
  6. ↵
    1. Chu DK,
    2. Kim LH,
    3. Young PJ, et al.
    Mortality and morbidity in acutely ill adults treated with liberal versus conservative oxygen therapy (IOTA): a systematic review and meta-analysis. Lancet 2018; 391: 1693–1705. doi:10.1016/S0140-6736(18)30479-3
    OpenUrlCrossRefPubMed
  7. ↵
    1. Barbateskovic M,
    2. Schjorring OL,
    3. Russo Krauss S, et al.
    Higher versus lower fraction of inspired oxygen or targets of arterial oxygenation for adults admitted to the intensive care unit. Cochrane Database Syst Rev 2019; 11: CD012631. doi:10.1002/14651858.CD012631.pub2
    OpenUrl
  8. ↵
    1. Hutton B,
    2. Salanti G,
    3. Caldwell DM, et al.
    The PRISMA extension statement for reporting of systematic reviews incorporating network meta-analyses of health care interventions: checklist and explanations. Ann Intern Med 2015; 162: 777–784. doi:10.7326/M14-2385
    OpenUrlCrossRefPubMed
  9. ↵
    1. Higgins JP,
    2. Altman DG,
    3. Gøtzsche PC, et al.
    The Cochrane Collaboration's tool for assessing risk of bias in randomised trials. BMJ 2011; 343: d5928. doi:10.1136/bmj.d5928
    OpenUrlFREE Full Text
  10. ↵
    1. Salanti G,
    2. Ades AE,
    3. Ioannidis JP
    . Graphical methods and numerical summaries for presenting results from multiple-treatment meta-analysis: an overview and tutorial. J Clin Epidemiol 2011; 64: 163–171. doi:10.1016/j.jclinepi.2010.03.016
    OpenUrlCrossRefPubMedWeb of Science
  11. ↵
    1. Dias S,
    2. Sutton AJ,
    3. Ades A, et al.
    Evidence synthesis for decision making 2: a generalized linear modeling framework for pairwise and network meta-analysis of randomized controlled trials. Med Decis Making 2013; 33: 607–617. doi:10.1177/0272989X12458724
    OpenUrlCrossRefPubMedWeb of Science
  12. ↵
    1. van Valkenhoef G,
    2. Lu G,
    3. de Brock B, et al.
    Automating network meta-analysis. Res Synth Methods 2012; 3: 285–299. doi:10.1002/jrsm.1054
    OpenUrl
  13. ↵
    1. Viechtbauer W
    . Conducting meta-analyses in R with the metafor package. J Stat Softw 2010; 36: 1–48. doi:10.18637/jss.v036.i03
    OpenUrlCrossRefPubMed
  14. ↵
    1. Feller M,
    2. Snel M,
    3. Moutzouri E, et al.
    Association of thyroid hormone therapy with quality of life and thyroid-related symptoms in patients with subclinical hypothyroidism: a systematic review and meta-analysis. JAMA 2018; 320: 1349–1359. doi:10.1001/jama.2018.13770
    OpenUrlCrossRefPubMed
  15. ↵
    1. Wetterslev J,
    2. Jakobsen JC,
    3. Gluud C
    . Trial sequential analysis in systematic reviews with meta-analysis. BMC Med Res Methodol 2017; 17: 39. doi:10.1186/s12874-017-0315-7
    OpenUrlCrossRefPubMed
  16. ↵
    1. Thorlund K,
    2. Engstrøm J,
    3. Wetterslev J, et al.
    User manual for Trial Sequential Analysis (TSA). 2017. https://ctu.dk/wp-content/uploads/2020/12/tsa_manual_ENG.pdf Date last accessed: 19 February 2021.
  17. ↵
    1. Lan KKG,
    2. DeMets DL
    . Discrete sequential boundaries for clinical trials. Biometrika 1983; 70: 659–663. doi:10.1093/biomet/70.3.659
    OpenUrlCrossRefPubMedWeb of Science
  18. ↵
    1. Dias S,
    2. Welton NJ,
    3. Caldwell DM, et al.
    Checking consistency in mixed treatment comparison meta-analysis. Stat Med 2010; 29: 932–944. doi:10.1002/sim.3767
    OpenUrlCrossRefPubMedWeb of Science
  19. ↵
    1. Egger M,
    2. Smith GD,
    3. Schneider M, et al.
    Bias in meta-analysis detected by a simple, graphical test. BMJ 1997; 315: 629–634. doi:10.1136/bmj.315.7109.629
    OpenUrlAbstract/FREE Full Text
  20. ↵
    1. Puhan MA,
    2. Schünemann HJ,
    3. Murad MH, et al.
    A GRADE Working Group approach for rating the quality of treatment effect estimates from network meta-analysis. BMJ 2014; 349: g5630. doi:10.1136/bmj.g5630
    OpenUrlFREE Full Text
  21. ↵
    1. Panwar R,
    2. Hardie M,
    3. Bellomo R, et al.
    Conservative versus liberal oxygenation targets for mechanically ventilated patients: a pilot multicenter randomized controlled trial. Am J Respir Crit Care Med 2016; 193: 43–51. doi:10.1164/rccm.201505-1019OC
    OpenUrlPubMed
  22. ↵
    1. Mackle D,
    2. Bellomo R,
    3. Bailey M, et al.
    Conservative oxygen therapy during mechanical ventilation in the ICU. N Engl J Med 2020; 382: 989–9998. doi:10.1056/NEJMoa1903297
    OpenUrlPubMed
  23. ↵
    1. Guyot P,
    2. Ades A,
    3. Ouwens MJ, et al.
    Enhanced secondary analysis of survival data: reconstructing the data from published Kaplan–Meier survival curves. BMC Med Res Methodol 2012; 12: 9. doi:10.1186/1471-2288-12-9
    OpenUrlCrossRefPubMed
  24. ↵
    1. de Jong VM,
    2. Moons KG,
    3. Riley RD, et al.
    Individual participant data meta-analysis of intervention studies with time-to-event outcomes: a review of the methodology and an applied example. Res Synth Methods 2020; 11: 148–168. doi:10.1002/jrsm.1384
    OpenUrl
  25. ↵
    1. Yang X,
    2. Shang Y,
    3. Yuan S
    . Low versus high pulse oxygen saturation directed oxygen therapy in critically ill patients: a randomized controlled pilot study. J Thorac Dis 2019; 11: 4234–4240. doi:10.21037/jtd.2019.09.66
    OpenUrl
  26. ↵
    1. Jakkula P,
    2. Reinikainen M,
    3. Hastbacka J, et al.
    Targeting two different levels of both arterial carbon dioxide and arterial oxygen after cardiac arrest and resuscitation: a randomised pilot trial. Intensive Care Med 2018; 44: 2112–2121. doi:10.1007/s00134-018-5453-9
    OpenUrl
  27. ↵
    1. Young P,
    2. Bailey M,
    3. Bellomo R, et al.
    HyperOxic Therapy OR NormOxic Therapy after out-of-hospital cardiac arrest (HOT OR NOT): a randomised controlled feasibility trial. Resuscitation 2014; 85: 1686–1691. doi:10.1016/j.resuscitation.2014.09.011
    OpenUrl
  28. ↵
    1. Asfar P,
    2. Schortgen F,
    3. Boisrame-Helms J, et al.
    Hyperoxia and hypertonic saline in patients with septic shock (HYPERS2S): a two-by-two factorial, multicentre, randomised, clinical trial. Lancet Respir Med 2017; 5: 180–190. doi:10.1016/S2213-2600(17)30046-2
    OpenUrl
  29. ↵
    1. Young PJ,
    2. Mackle DM,
    3. Bailey MJ, et al.
    Intensive care unit randomised trial comparing two approaches to oxygen therapy (ICU-ROX): results of the pilot phase. Crit Care Resusc 2017; 19: 344–354.
    OpenUrl
  30. ↵
    1. Ali K,
    2. Warusevitane A,
    3. Lally F, et al.
    The stroke oxygen pilot study: a randomized controlled trial of the effects of routine oxygen supplementation early after acute stroke – effect on key outcomes at six months. PLoS One 2014; 8: e59274. doi:10.1371/journal.pone.0059274
    OpenUrlPubMed
  31. ↵
    1. Mazdeh M,
    2. Taher A,
    3. Torabian S, et al.
    Effects of normobaric hyperoxia in severe acute stroke: a randomized controlled clinical trial study. Acta Med Iran 2015; 53: 676–680.
    OpenUrl
  32. ↵
    1. Padma MV,
    2. Bhasin A,
    3. Bhatia R, et al.
    Normobaric oxygen therapy in acute ischemic stroke: a pilot study in Indian patients. Ann Indian Acad Neurol 2010; 13: 284–288. doi:10.4103/0972-2327.74203
    OpenUrlCrossRefPubMed
  33. ↵
    1. Rønning OM,
    2. Guldvog B
    . Should stroke victims routinely receive supplemental oxygen? A quasi-randomized controlled trial. Stroke 1999; 30: 2033–2037. doi:10.1161/01.STR.30.10.2033
    OpenUrlAbstract/FREE Full Text
  34. ↵
    1. Singhal AB,
    2. Benner T,
    3. Roccatagliata L, et al.
    A pilot study of normobaric oxygen therapy in acute ischemic stroke. Stroke 2005; 36: 797–802. doi:10.1161/01.STR.0000158914.66827.2e
    OpenUrlAbstract/FREE Full Text
  35. ↵
    1. Roffe C,
    2. Nevatte T,
    3. Sim J, et al.
    Effect of routine low-dose oxygen supplementation on death and disability in adults with acute stroke: the stroke oxygen study randomized clinical trial. JAMA 2017; 318: 1125–1135. doi:10.1001/jama.2017.11463
    OpenUrlCrossRefPubMed
  36. ↵
    1. Shi S,
    2. Qi Z,
    3. Ma Q, et al.
    Normobaric hyperoxia reduces blood occludin fragments in rats and patients with acute ischemic stroke. Stroke 2017; 48: 2848–2854. doi:10.1161/STROKEAHA.117.017713
    OpenUrlAbstract/FREE Full Text
  37. ↵
    1. Hofmann R,
    2. James SK,
    3. Jernberg T, et al.
    Oxygen therapy in suspected acute myocardial infarction. N Engl J Med 2017; 377: 1240–1249. doi:10.1056/NEJMoa1706222
    OpenUrlCrossRefPubMed
    1. Khoshnood A,
    2. Carlsson M,
    3. Akbarzadeh M, et al.
    The effects of oxygen therapy on myocardial salvage in ST elevation myocardial infarction treated with acute percutaneous coronary intervention: the Supplemental Oxygen in Catheterized Coronary Emergency Reperfusion (SOCCER) Study. Cardiology 2015; 132: 16–21. doi:10.1159/000398786
    OpenUrl
    1. Ranchord AM,
    2. Argyle R,
    3. Beynon R, et al.
    High-concentration versus titrated oxygen therapy in ST-elevation myocardial infarction: a pilot randomized controlled trial. Am Heart J 2012; 163: 168–175. doi:10.1016/j.ahj.2011.10.013
    OpenUrlCrossRefPubMed
    1. Rawles JM,
    2. Kenmure AC
    . Controlled trial of oxygen in uncomplicated myocardial infarction. Br Med J 1976; 1: 1121–1123.
    OpenUrlAbstract/FREE Full Text
  41. ↵
    1. Stub D,
    2. Smith K,
    3. Bernard S, et al.
    A randomized controlled trial of oxygen therapy in acute myocardial infarction Air Verses Oxygen In myocarDial infarction study (AVOID Study). Am Heart J 2012; 163: 339–345. doi:10.1016/j.ahj.2011.11.011
    OpenUrlCrossRefPubMedWeb of Science
  42. ↵
    1. Ukholkina GB,
    2. Kostianov I,
    3. Kuchkina NV, et al.
    [Effect of oxygenotherapy used in combination with reperfusion in patients with acute myocardial infarction]. Kardiologiia 2005; 45: 59.
    OpenUrlPubMed
  43. ↵
    1. Butler CM,
    2. Ham RO,
    3. Lafferty K, et al.
    The effect of adjuvant oxygen therapy on transcutaneous pO2 and healing in the below-knee amputee. Prosthet Orthot Int 1987; 11: 10–16. doi:10.3109/03093648709079373
    OpenUrlCrossRefPubMed
  44. ↵
    1. Schietroma M,
    2. Cecilia EM,
    3. De Santis G, et al.
    Supplemental peri-operative oxygen and incision site infection after surgery for perforated peptic ulcer: a randomized, double-blind monocentric trial. Surg Infect 2016; 17: 106–113. doi:10.1089/sur.2013.132
    OpenUrl
  45. ↵
    1. Bickel A,
    2. Gurevits M,
    3. Vamos R, et al.
    Perioperative hyperoxygenation and wound site infection following surgery for acute appendicitis: a randomized, prospective, controlled trial. Arch Surg 2011; 146: 464–470. doi:10.1001/archsurg.2011.65
    OpenUrlCrossRefPubMedWeb of Science
  46. ↵
    1. Kuisma M,
    2. Boyd J,
    3. Voipio V, et al.
    Comparison of 30 and the 100% inspired oxygen concentrations during early post-resuscitation period: a randomised controlled pilot study. Resuscitation 2006; 69: 199–206. doi:10.1016/j.resuscitation.2005.08.010
    OpenUrlCrossRefPubMedWeb of Science
  47. ↵
    1. Taher A,
    2. Pilehvari Z,
    3. Poorolajal J, et al.
    Effects of normobaric hyperoxia in traumatic brain injury: a randomized controlled clinical trial. Trauma Monthly 2016; 21: e26772. doi:10.5812/traumamon.26772
    OpenUrl
  48. ↵
    1. van den Boom W,
    2. Hoy M,
    3. Sankaran J, et al.
    The search for optimal oxygen saturation targets in critically ill patients: observational data from large ICU databases. Chest 2020; 157: 566–573. doi:10.1016/j.chest.2019.09.015
    OpenUrlCrossRef
  49. ↵
    1. Schjørring OL,
    2. Jensen AK,
    3. Nielsen CG, et al.
    Arterial oxygen tensions in mechanically ventilated ICU patients and mortality: a retrospective, multicentre, observational cohort study. Br J Anaesth 2020; 124: 420–429. doi:10.1016/j.bja.2019.12.039
    OpenUrl
  50. ↵
    1. Brower RG,
    2. Matthay MA,
    3. Morris A, et al.
    Ventilation with lower tidal volumes as compared with traditional tidal volumes for acute lung injury and the acute respiratory distress syndrome. N Engl J Med 2000; 342: 1301–1308. doi:10.1056/NEJM200005043421801
    OpenUrlCrossRefPubMedWeb of Science
  51. ↵
    1. Schjørring OL,
    2. Perner A,
    3. Wetterslev J, et al.
    Handling Oxygenation Targets in the Intensive Care Unit (HOT-ICU) – protocol for a randomised clinical trial comparing a lower vs a higher oxygenation target in adults with acute hypoxaemic respiratory failure. Acta Anaesthesiol Scand 2019; 63: 956–965.
    OpenUrl
    1. Guyatt G,
    2. Oxman AD,
    3. Akl EA, et al.
    GRADE guidelines: 1. Introduction – GRADE evidence profiles and summary of findings tables. J Clin Epidemiol 2011; 64: 383–394. doi:10.1016/j.jclinepi.2010.04.026
    OpenUrlCrossRefPubMed
  53. ↵
    1. Severinghaus JW
    . Simple, accurate equations for human blood O2 dissociation computations. J Appl Physiol Respir Environ Exerc Physiol 1979; 46: 599–602.
    OpenUrlPubMedWeb of Science
Back to top
View this article with LENS
Email
Print
Citation Tools

Classification and effectiveness of different oxygenation goals in mechanically ventilated critically ill patients: network meta-analysis of randomised controlled trials
Xu Zhao, Huaping Xiao, Feng Dai, Daniel Brodie, Lingzhong Meng
European Respiratory Journal Sep 2021, 58 (3) 2002928; DOI: 10.1183/13993003.02928-2020
del.icio.us logo Digg logo Reddit logo Technorati logo Twitter logo CiteULike logo Connotea logo Facebook logo Google logo Mendeley logo
Full Text (PDF)

Jump To

Subjects