The optimal arterial oxygenation range in critical care

To the Editor:

Oxygen therapy remains a crucial intervention in critical care. One of the core aspects in oxygen therapy concerns the optimal oxygenation level, which is commonly measured in the form of partial pressure of arterial oxygen (P_aO2). Multiple randomised controlled trials (RCTs) have been performed to investigate different oxygenation goals in critical care [1, 2]. However, one important challenge, as demonstrated by a recent network meta-analysis [2], is that most RCTs only compared two oxygenation levels that differed across studies, making it difficult to determine an optimal oxygenation goal [2]. While several meta-analyses have been performed, most have dichotomised oxygenation goals into liberal (i.e. targeting a higher oxygenation level) versus conservative (i.e. targeting a lower oxygenation level): a relatively qualitative oversimplification of the problem [3, 4]. Therefore, further investigations are needed to determine a more quantitative optimal oxygenation goal.

A recent network meta-analysis proposed a system classifying arterial oxygenation into three ranges: conservative (P_aO2 55–90 mmHg), moderate (P_aO2 90–150 mmHg) and liberal (P_aO2 >150 mmHg) [2]. To allow for further discrimination in the conservative range, a system with four ranges was also considered, comprising far-conservative (P_aO2 55–70 mmHg), conservative (P_aO2 70–90 mmHg), moderate (P_aO2 90–150 mmHg), and liberal (P_aO2 >150 mmHg) [2]. Although this network meta-analysis did not find differences in mortality among different oxygenation goals in critically ill patients, the survival analysis and the analysis based on the surface under the cumulative ranking curve (SUCRA) score suggested that maintaining P_aO2 in the range of 70–150 mmHg might be superior [2].

Following the publication of the above mentioned network meta-analysis [2], two new studies have been published [5, 6]. Adding these trials, which meet the inclusion and exclusion criteria, to the original network meta-analysis would expand the number of eligible patients from 2532 to 5859. We, therefore, conducted an updated network meta-analysis by including these two new studies to enhance our understanding of this topic. Like the original network meta-analysis, the classification of oxygenation ranges was based on the actual, not the targeted, P_aO2. The new studies compared two oxygenation goals in each study, with the goals classified as conservative and moderate per the trinary and quadruple classification systems. We used 90-day mortality as the primary outcome in this updated analysis, as this is the common outcome measure reported in the eligible studies.

The results based on the trinary classification are presented in figure 1a–c. There were no differences among the conservative, moderate and liberal goals based on network meta-analysis (figure 1a). The survival analysis showed that the moderate goal led to improved survival compared to the liberal goal (hazard ratio (HR) 0.77, 95% CI 0.60–0.99), while there were no differences among the other comparisons (figure 1c). Moreover, the SUCRA score suggested that the moderate goal was likely the most effective (SUCRA 84.2%), followed by the conservative goal (SUCRA 41.9%) (figure 1b). The results based on the quadruple classification are presented in figure 1d–f. There were no differences among the far conservative, conservative, moderate and liberal goals based on network meta-analysis (figure 1d). The survival analysis showed that both the moderate goal (HR 0.76, 95% CI 0.59–0.98) and the conservative goal (HR 0.76, 95% CI 0.58–0.99) led to improved survival compared with the liberal goal, while there were no differences among the other comparisons (figure 1f). Again, the SUCRA score suggested that the moderate goal was likely the most effective (SUCRA 82.6%), followed by the conservative goal (SUCRA 66.9%) (figure 1e).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1

Results of different analyses. a–c) Results of the analyses based on the trinary classification. d–f) Results of the analyses based on the quadruple classification. a, d) Results of network meta-analysis. b, e) Results of SUCRA score analysis. c, f) Results of the survival analysis. The Kaplan–Meier curves were constructed using the pooled patient-level data extracted from six randomised controlled trials [1]. We used the survival data up to 90 days when available from the following studies: one study up to 28 days [2], one study up to 60 days [3], one study up to 80 days [4], and five studies up to 90 days [5–9]. HR: hazard ratio; CI: confidence interval; CrI: credible interval; P_aO2: partial pressure of arterial oxygen; SUCRA: surface under the cumulative ranking curve.

We also performed a trial sequential analysis based on the data comparing conservative and moderate goals per the quadruple classification system. This analysis based on three studies indicates that the required information size (n=1158) is met (the pooled number of patients, n=4147) and there is no difference between conservative and moderate goals [5–7]. The inclusion of these additional studies has strengthened our confidence in this result, as the same analysis in the original network meta-analysis was only based on one study (n=959) [7].

In summary, the integrated data based on 10 RCTs [2, 5–12] suggest that maintaining P_aO2 in the 70–150 mmHg range might lead to improved 90-day survival compared to a P_aO2 range below 70 mmHg or above 150 mmHg in critically ill patients requiring respiratory support. A narrower optimal range within 70–150 mmHg might be possible, but we need more data: maybe individual patient data meta-analysis. At present, the available evidence suggests that different P_aO2 targets in the 70–150 mmHg range might not lead to different outcomes.

Shareable PDF