key: cord-1030842-f0n27caf
authors: Klitgaard, Thomas Lass; Schjørring, Olav Lilleholt; Nielsen, Frederik Mølgaard; Meyhoff, Christian Sylvest; Barbateskovic, Marija; Wetterslev, Jørn; Perner, Anders; Rasmussen, Bodil Steen
title: Letter the editor: serious methodological concerns about a recently published meta-analysis on oxygen therapy
date: 2021-12-07
journal: J Intensive Care
DOI: 10.1186/s40560-021-00573-5
sha: f777cc012014e8077b901991d4edac601f345fd1
doc_id: 1030842
cord_uid: f0n27caf

In a recent paper, Chen et al. report the findings of a systematic review with meta-analysis concerning conservative versus conventional oxygen therapy for critically ill patients. We wish to commend the authors for their interest in the matter. However, the authors appear to misquote findings, fail to report results for all specified analyses, do not identify all relevant trials, have post hoc changed the eligibility criteria, and have seemingly switched directions of effects in analyses of secondary outcomes. These issues have led to incorrect conclusions concerning the effects of targeted oxygen therapy in critically ill patients.

To the editor,

We have with interest read the systematic review with meta-analysis concerning the effects of conservative versus conventional oxygen therapy for critically ill patients by Chen et al. [1] . However, we have several concerns relating to the methodology and findings. None of the analyses and figures presented in this letter have been published elsewhere. They were specifically constructed for the purpose of this letter.

In the paper by Chen et al. [1] , the mortality rates are erroneously quoted from several trials in the meta-analysis of mortality at longest follow-up. In the paper by Schjørring et al. [2] , a mortality of 514/1447 and 529/1441 in the higher and lower group is incorrectly quoted. The correct mortality was 613/1447 and 618/1441, respectively [2] . Mortality in the liberal group in the study by Barrot et al. was 31/102 [3] , not 39/102 as stated. Twentyeight-day mortality for Asfar et al. is quoted despite 90-day mortality is reported in the trial paper [4] . The ICU-mortality in the modified intention-to-treat population for Girardis et al. is quoted although hospital mortality for the intention-to-treat cohort is reported in the trial paper [5] . A revised meta-analysis is presented in Fig. 1 . Chen et al. reported the RR as 1.01 (95% CI 0.94-1.09), so there is a slight difference in the 95% CI [1] .

The literature search is insufficient as the authors fail to identify four relevant papers focusing on: ICU-patients with acute exacerbation of chronic obstructive pulmonary disease [6]; oxygen therapy after cardiac-arrest [7]; normobaric oxygen in stroke patients [8] ; and hyperoxaemia in stroke patients [9]. The first paper should have been identified and included in the meta-analysis, whilst the latter three should have been identified and excluded as per their stated exclusion criteria [1] . In their PRISMA-diagram, the authors state that six trials

This letter refers to the article available online at https:// doi. org/ 10. 1186/ s40560-021-00563-7.

*Correspondence: tlk@rn.dk 1 Department of Anaesthesia and Intensive Care, Aalborg University Hospital, Aalborg, Denmark Full list of author information is available at the end of the article were excluded after full-text review and present the trials along with reasons for exclusions (Additional file 3: Table S1 ). In the main text and in this table only five trials are quoted. Moreover, the eligibility criteria have been changed post hoc, without justification, now excluding trials with patients at risk of ischaemia or hypoxic encephalopathy. No such criteria are mentioned in the protocol [10].

The authors' choice of subgroup analysis based on baseline ratios of partial pressure of oxygen to fraction of inspired oxygen (PaO 2 /FiO 2 ) as according to mild, moderate, and severe acute respiratory distress syndrome (> 200 mmHg, 100-200 mmHg, and < 100 mmHg, respectively) is problematic, as the results from this analysis, specified in the statistical analysis section, are not presented, except for the results from the sensitivity analysis of trials excluding patients with a PaO 2 /FiO 2 ratio < 100 mmHg (in the abstract). In the main text and their Fig. 2 , the authors pool three trials all excluding patients with baseline PaO 2 /FiO 2 ratios < 100 mmHg [4, 11] Though both trials did not restrict inclusion based on PaO 2 /FiO 2 ratios, most patients included in these two trials clearly satisfy the criteria for inclusion in the subgroup analysis above. As no baseline PaO 2 /FiO 2 ratios were presented by Girardis et al. [5] , no knowledge of severity of respiratory failure can be ascertained. Therefore, this study should be excluded from the subgroup analysis. In the HOT-ICU trial [2] , inclusion was not restricted by PaO 2 /FiO 2 ratio, and the median baseline PaO 2 /FiO 2 ratios were approximately 118 mmHg in both groups. However, a substantial proportion of patients had a ratio ≥ 150 mmHg. We acknowledge that cohortlevel-based separations may seem to provide easy new knowledge when performing a meta-analysis, but with such heterogenous groups of included patients in each trial, the only reliable answer to risks according to baseline degree of respiratory failure would come from individual-based-separations and access to all trials' datasets. Below is provided a revised meta-analysis on mortality at longest follow-up stratified on the specified separation of trials (Fig. 2) . This clearly changes the conclusion of the subgroup analysis, as the subgroup of trials with reported baseline PaO 2 /FiO 2 ratios > 200 mmHg now produces a statistically non-significant result (and non-significant test for subgroup differences), contrary to the results presented by the authors.

Lastly, it appears that the two compared groups have been switched when reporting serious adverse events, despite correct findings are provided in the supplement (Additional file 6) [1] . If inversed, the results are in line with the meta-analysis provided below (Figs. 3, 4, 5) . Conclusions based on these analyses now point in the opposite direction as to what was reported by authors, though still statistically insignificant.

Meta-analyses of high-quality trials are considered the highest level of evidence. Thus, the methodology applied needs to be of similar high quality. If not, inappropriate conclusions may be drawn, potentially misguiding clinical practice. In their review and metaanalysis, Chen et al. fail in several crucial domains, thereby presenting incorrect results and conclusions. 

Ready to submit your research Ready to submit your research ? Choose BMC and benefit from:

? Choose BMC and benefit from:

Conservative oxygen therapy for critically ill patients: a meta-analysis of randomized controlled trials

Lower or higher oxygenation targets for acute hypoxemic respiratory failure

LOCO 2 investigators and REVA research network liberal or conservative oxygen therapy for acute respiratory distress syndrome

Hyperoxia and hypertonic saline in patients with septic shock (HYPERS2S): a two-by-two factorial, multicentre, randomised, clinical trial

None.

Authors' contributions TLK wrote the manuscript. All authors read and approved the final manuscript.

None.

Not applicable.

Ethics approval and consent to participate Not applicable.

Not applicable.

TLK and OLS are coordinating investigators of the Handling Oxygenation Targets in the Intensive Care Unit (HOT-ICU) trial. FMN and TLK are coordinating investigators of the Handling Oxygenation Targets in COVID-19 (HOT-COVID) trial. BSR is the sponsor and primary investigator of both the HOT-ICU and HOT-COVID trials. All authors are part of a working group currently updating the latest Cochrane review concerned with higher or lower oxygen levels in