Article Text
PDF
Abstract
Background Although acute respiratory distress syndrome (ARDS) is associated with high mortality, its direct causal link with death is unclear. Clarifying this link is important to justify costly research on prevention of ARDS.
Objective To estimate the attributable mortality, if any, of ARDS.
Design First, we performed a systematic review and meta-analysis of observational studies reporting mortality of critically ill patients with and without ARDS matched for underlying risk factor. Next, we conducted a survival analysis of prospectively collected patient-level data from subjects enrolled in three intensive care unit (ICU) cohorts to estimate the attributable mortality of critically ill septic patients with and without ARDS using a novel causal inference method.
Results In the meta-analysis, 44 studies (47 cohorts) involving 56 081 critically ill patients were included. Mortality was higher in patients with versus without ARDS (risk ratio 2.48, 95% CI 1.86 to 3.30; p<0.001) with a numerically stronger association between ARDS and mortality in trauma than sepsis. In the survival analysis of three ICU cohorts enrolling 1203 critically ill patients, 658 septic patients were included. After controlling for confounders, ARDS was found to increase the mortality rate by 15% (95% CI 3% to 26%; p=0.015). Significant increases in mortality were seen for severe (23%, 95% CI 3% to 44%; p=0.028) and moderate (16%, 95% CI 2% to 31%; p=0.031), but not for mild ARDS.
Conclusions ARDS has a direct causal link with mortality. Our findings provide information about the extent to which continued funding of ARDS prevention trials has potential to impart survival benefit.
PROSPERO Registration Number CRD42017078313
- ARDS
- critical care
- clinical epidemiology
Data availability statement
Data are available upon reasonable request. Deidentified subject data are available upon reasonable request from researchers whose proposed use of data has been approved. Inquiries for reuse of data may be made by emailing lkt9003@med.cornell.edu.
Statistics from Altmetric.com
Key messages
What is the key question?
What does the existing data of published work show on attributable mortality, if any, of acute respiratory distress syndrome (ARDS) and how may these frequently unadjusted estimates be improved using contemporary statistical approaches in future observational studies?
What is the bottom line?
There is a need to examine whether ARDS has attributable mortality to provide information on whether continued funding of costly prevention trials may provide survival benefit.
Why read on?
This study systematically appraises and summarises the association between ARDS and mortality for risk factors including sepsis, trauma and others while demonstrating novel statistical methodology using three cohorts that adjusts for confounders, varying lengths of stay and underlying severity of illness.
Introduction
Acute respiratory distress syndrome (ARDS) may be present in up to 10% of patients in the intensive care unit (ICU) and is associated with high mortality.1 Mortality of patients with ARDS ranges from approximately 35% for mild to 46% for severe ARDS according to a recent international, multicenter, prospective cohort study conducted by the Large observational study to UNderstand the Global impact of Severe Acute respiratory FailurE (LUNG SAFE) investigators.1 Several groups of investigators, including the National Heart, Lung, and Blood Institute-funded Prevention and Treatment of Acute Lung Injury Network, have been formed to explore strategies to prevent ARDS assuming that prevention of ARDS would lead to decreased mortality of critically ill patients.2 3 In other words, it seems generally assumed that ARDS is independently associated with mortality.
Previous attempts to determine whether and to which degree ARDS has attributable mortality in critically ill patients might be hindered for a multitude of reasons. First, details on the adjudication of ARDS in relevant observational studies might be unavailable or unreported. Second, it is unclear whether previous studies adequately adjusted for confounders, such as the underlying risk factor and number of non-pulmonary organ failures, which may exert the greatest influence in determining risk of death when ARDS is present.4 Third, previous estimates of the attributable mortality of ARDS might be inaccurate due to underutilisation of statistical methodology which accounts for varying lengths of stay. Novel causal inference methods may address the latter issues and have been employed to answer other questions in the critical illness literature.5–7
Studies evaluating the association between ARDS and mortality have not been systematically appraised. We first performed a systematic review and meta-analysis of relevant evidence to assess the association between ARDS and mortality. We then performed a survival analysis to estimate the effect of ARDS on mortality in three prospectively enrolled ICU cohorts of critically ill patients. A secondary analysis estimating mortality differences of mechanically ventilated patients was also performed. For our final analysis, we used a working marginal structural Cox model (Cox-WMSM), a causal inference method which adjusts for time-dependent covariates, to assess the effect of an abnormal partial pressure of arterial oxygen to fraction of inspired oxygen (PF) ratio on the risk of mortality. Some of the results of our study have been previously reported in the form of an abstract.8 9
Methods
The present study consisted of two components: a ‘meta-analysis’ component and a subsequent original survival analysis of prospectively collected patient-level data from subjects enrolled in three ICU cohorts by using a novel causal inference method (‘ICU cohorts’ component).
Meta-analysis
The present systematic review and meta-analysis were reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement and were registered with PROSPERO.10
We searched for observational studies assessing mortality between patients with ARDS and patients without ARDS in MEDLINE, EMBASE and SCOPUS using the terms and synonyms (“acute lung injury” OR “acute respiratory distress syndrome” OR “ARDS”) AND (“regression” OR “matched” OR “attributable” OR “excess” OR “extra” OR “case-control”) AND (“mortality” OR “survival” OR “outcome”) up to 12 December 2016. There were no limitations on time or language of publications. Additional information regarding inclusion and exclusion criteria as well as risk factor assessment are in the online supplemental appendix.
Supplemental material
Data extraction
Data were extracted independently from each publication by two experienced reviewers (LKT and JSH). Any discrepancies were resolved through consensus with a third author (IIS). Studies were classified based on whether the risk factor was ‘sepsis’, ‘trauma’ or ‘other’. Mortality data were extracted, and in some cases calculated, based on values or percentages provided in the publications. Both hospital and ICU length of stay were also recorded. For additional detail, please see the online supplemental appendix.
Risk of bias assessment
Eligible studies were assessed for their risk of bias using the ‘Tool to Assess Risk of Bias in Cohort Studies’ by Cochrane.11 For additional detail regarding risk of bias assessment, please see the online supplemental appendix.
Outcomes in the meta-analysis
The primary outcomes of the meta-analysis were in-hospital all-cause mortality and short-term all-cause mortality (which combines 28-day mortality and ICU mortality). Secondary outcomes included long-term all-cause mortality (which combines mortality up to 3 months and 1 year), length of stay in the ICU and length of stay in the hospital.
ICU cohorts
This was an observational cohort study of prospectively enrolled, critically ill patients in three ICUs in the USA or South Korea, matched by the risk factor sepsis. Findings were reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology statement.12 Patients were enrolled in the NewYork-Presbyterian Hospital/Weill Cornell Medicine (WCM) cohort from 2014 to 2018, in the Brigham and Women’s Hospital (BWH) cohort from 2008 to 2018 and in the Asan Medical Center (ASAN) cohort from 2011 to 2016. Patients were screened daily by research coordinators for suspicion of sepsis and patient or his or her surrogate were approached to obtain written informed consent for enrolment into the biobank. All three of the cohorts are medical ICUs. The Sepsis-3 definition was used to define sepsis; for cases prior to 2016, this definition was applied retrospectively.13 The American-European Consensus Conference definition was used to define ARDS cases prior to 2012.14 The Berlin definition of ARDS was used for patients enrolled in their respective cohorts from 2012 onward.15 ARDS was adjudicated independently by at least two physicians based on clinical context, chest X-rays and arterial blood gases of all patients enrolled in the three ICU cohorts. Additional details on subject recruitment and data adjudication have been published previously16 17 and are provided in the online supplemental appendix.
Data analysis
Subjects with and without ARDS were pooled for further analysis and baseline characteristics were obtained. Severity of illness as measured by the Sequential Organ Failure Assessment (SOFA) scores were calculated for each patient on admission to the ICU.18 A modified SOFA score was calculated for each subject by removing the respiratory component of the score.
Severity of ARDS as determined by PF ratio was calculated for each patient according to the Berlin classification. For the primary analysis, the values were taken from data obtained within the first 24 hours of ICU admission for subjects without ARDS and within 24 hours of ARDS identification for subjects with ARDS. The rationale for this approach was to compare baseline severity of illness at the beginning of ARDS versus the beginning of sepsis without ARDS. For additional detail on determination of severity of ARDS, please see the online supplemental appendix.
Data were harmonised among the cohorts using comorbidity measures as detailed by Elixhauser et al.19 Five comorbid conditions were selected including congestive heart failure, chronic pulmonary disease, renal failure, liver disease and malignancy (a pooled category of lymphoma, metastatic cancer and solid tumour without metastasis).
Outcomes in the ICU cohorts
Our primary outcome in the ICU cohorts was 28-day mortality. Potential confounders for which adjustment was made included age, comorbidities, modified SOFA score and cohort. We also performed a sensitivity analysis to compare 28-day mortality for patients with and without pneumonia. For our secondary analyses, we then stratified by mechanical ventilation status at the time of ARDS diagnosis or ICU admission to examine whether the overall effect of ARDS on 28-day mortality was modified in patients who were invasively mechanically ventilated. All patients were censored at hospital discharge. Lastly, we investigated whether changes in daily PF ratio, adjusting for potential time-varying confounders such as modified SOFA score and mechanical ventilation, affect 28-day mortality in critically ill patients.
Patient and public involvement
This research contained no direct patient or public involvement, however the research question was formed based on continued international efforts to fund research for the prevention and treatment of ARDS.
Statistical analyses
For the meta-analysis, we assessed the attributable mortality of ARDS after calculating the risk ratio (RR) of mortality of patients who developed ARDS compared with patients who did not. We pooled dichotomous (expressed as RR with 95% CIs) and continuous (expressed as mean difference with 95% CI) effect measures using a random effects model. We assessed length of stay in ICU and hospital by converting median values with interquartile ranges to means with variance where applicable.20 We quantified statistical heterogeneity using the I2 statistic. Values of I2 <40%, <60% and <90% and ≥90% were considered to denote non-important, moderate, substantial and considerable heterogeneity, respectively.21 To address heterogeneity, we performed prespecified subgroup analyses to dissect the effect of sepsis, trauma and other risk factor-associated ARDS on mortality. We also carried out another prespecified subgroup analysis including studies that adjusted for baseline severity of illness using multivariable regression. We used Review Manager V.5.3 software (Nordic Cochrane Center, Cochrane Collaboration, Copenhagen, 2014).
For the ICU cohorts, all descriptive p values are two-sided and were calculated using analysis of variance (ANOVA), Kruskal-Wallis or χ2 tests for normal continuous variables, non-normal continuous variables and categorical variables, respectively. We first calculated the unadjusted mortality incidences using Kaplan-Meier survival curves. We then used survival analysis with targeted minimum loss-based estimation (TMLE) to assess the effect of ARDS on mortality.22 23 TMLE is an estimation method that adjusts for patient characteristics that are informative of censoring (hospital discharge) or confound the relation between ARDS and mortality. It uses estimators of two models: the probability of ARDS and the probability of death, conditional on patient characteristics.24 25 These models were estimated with a model ensembling technique known as super learning.23 26 We included five candidate estimators in our super learner: polynomial multivariate regression splines, random forest, gradient boosting, least absolute shrinkage and selection operator (LASSO) penalised regression and main-term logistic regression. Estimates were for the difference in mortality by 28 or 60 days for each category of ARDS severity (any, mild, moderate or severe) compared with no ARDS diagnosis. Further, in order to investigate whether changes in daily PF ratio adjusting for potential time-varying confounders affect 28-day mortality, we used daily longitudinal data available from the WCM cohort. We used TMLE to estimate a Cox-WMSM,27 adjusting for the time-dependent confounders of mechanical ventilation and modified SOFA scores, in addition to the baseline confounders of our primary analysis. Directed acyclic graphs were created to illustrate the assumed causal systems of the aforementioned models (online supplemental appendix 8).28 Statistical analyses were performed in R V.3.6.1 (Vienna, Austria, 2019).29
Results
Meta-analysis
Figure 1 shows the flow diagram for study selection. Of the 3119 initially retrieved articles, 44 studies involving 47 cohorts were included.30–73 Of the 56 081 patients pooled, 5897 had ARDS and 50 184 did not have ARDS. Online supplemental appendix 9 summarises the characteristics of included studies, in which the underlying risk factor was sepsis (16 cohorts), trauma (19 cohorts) and other (12 cohorts), respectively.
Flow diagram of studies included in the systematic review and meta-analysis. ARDS, acute respiratory distress syndrome. LUNG SAFE, Large observational study to UNderstand the Global impact of Severe Acute respiratory FailurE.
Risk of bias assessment
Online supplemental appendix 10 shows the assessment for risk of bias among included studies. Biases in reliability of ARDS identification were common and most studies did not adjust for confounders, such as baseline severity of illness using a modified (ie, after exclusion of its respiratory component) severity score.
Primary outcomes in the meta-analysis
Data from 33 cohorts and 51 366 patients revealed that in-hospital all-cause mortality was higher in patients with versus without ARDS (26.1% vs 4.9%; RR 2.48, 95% CI 1.86 to 3.30; p<0.001; figure 2).30 32–34 37 38 42–44 46–48 50–55 57–61 64 66 68–73 Statistical heterogeneity (I2=95%) was considerable. A sensitivity analysis was performed of studies with low or moderate risk of bias in ARDS adjudication (online supplemental appendix 12).
Forest plot of in-hospital all-cause mortality among septic, trauma and other patients. ARDS, acute respiratory distress syndrome.
Data from 14 cohorts and 5053 patients revealed that short-term all-cause mortality was higher in patients with versus without ARDS (39.2% vs 30.6%; RR 1.91, 95% CI 1.22 to 3.01; p=0.005; online supplemental appendix 13).31 35 36 40 41 45 49 54 56 62 63 65 67 71 Statistical heterogeneity (I2=93%) was considerable.
Secondary outcomes in the meta-analysis
There was no difference in long-term all-cause mortality in patients with versus without ARDS (nine cohorts; 2260 patients; 26.7% vs 19.6%; RR 1.52, 95% CI 0.92 to 2.52; p=0.10; online supplemental appendix 14).32 38 39 47 49 54 Statistical heterogeneity (I2=87%) was substantial. Stay in the ICU (10 cohorts; 10 259 patients; mean difference 9.19 days, 95% CI 6.16 to 11.77; p<0.001; online supplemental appendix 15)34 36 38 43 53 57 64 68 70 73 and the hospital (10 cohorts; 10 234 patients; mean difference 11.20 days, 95% CI 7.22 to 15.18; p<0.001; online supplemental appendix 16)34 39 49 52 53 57 64 68 70 73 were more prolonged for patients with versus without ARDS. Statistical heterogeneity was considerable for both stay in the ICU (I2=91%) and stay in the hospital (I2=90%).
Subgroup analyses in the meta-analysis
A prespecified subgroup analysis according to underlying risk factor was performed. In the subgroup of seven cohorts enrolling septic patients, in-hospital all-cause mortality was higher in patients with versus without ARDS (50.3% vs 23.6%; RR 1.85, 95% CI 1.18 to 2.90; p=0.007; online supplemental appendix 17).32 42 50 54 58 71 72 In the subgroup of 17 cohorts enrolling trauma patients, in-hospital all-cause mortality was higher in patients with versus without ARDS (21.9% vs 4.3%; RR 2.86, 95% CI 1.90 to 4.30; p<0.001; online supplemental appendix 18).30 32 33 46–48 50 53 55 57 59–61 64 68 70 73 In the subgroup of nine cohorts enrolling patients with ARDS risk factor other than sepsis or trauma, in-hospital all-cause mortality was higher in patients with versus without ARDS (29.4% vs 4.2%; RR 2.42, 95% CI 1.46 to 4.01; p<0.001; online supplemental appendix 19).34 37 38 43 44 51 52 66 69
A total of 12 cohorts controlled for baseline severity of illness using multivariable regression analysis.31 34 41 44 47 53 57–59 62–64 Approximately half of these (5 out of 12 or 41.6%) showed that ARDS was not an independent predictor of mortality.41 44 53 58 64 One (trauma) study controlled for modified (ie, after excluding the respiratory component) baseline severity of illness.59
ICU cohorts
Of the 1203 critically ill patients enrolled in the three ICU cohorts, a total of 658 subjects with sepsis were included in our analysis. Characteristics of included patients by cohort are presented in table 1. Subjects from ASAN had a higher mean age and total and modified SOFA scores compared with BWH and WCM, respectively. ASAN had smaller proportions of each comorbidity compared with BWH and WCM. The distribution of modified SOFA scores and PF ratios by cohort is available in online supplemental appendices 20 and 21, respectively.
Characteristics of patients included in the three ICU cohorts of the original analysis
Of the subjects with sepsis in our analysis, 141 (21.4%) were adjudicated to have ARDS. Characteristics of included patients by presence or absence of ARDS are presented in table 2. There were no significant differences in presence of comorbidities or proportion of subjects with sepsis or septic shock between subjects with ARDS and those without ARDS. Subjects with ARDS had a lower mean age in years compared with subjects without ARDS (60.8 vs 65.1; p=0.003). Subjects with ARDS had higher mean total SOFA (9.95 vs 7.95; p<0.001) and modified SOFA (7.00 vs 6.08; p=0.004) scores compared with subjects without ARDS.
Characteristics of patients with versus without ARDS included in the three ICU cohorts of the original analysis
Primary outcome in the ICU cohorts
In the ICU cohorts, the primary outcome of 28-day mortality was reached in 38.3% of patients with ARDS and 19.1% of patients without ARDS (p<0.001). Estimates (unadjusted, adjusted and adjusted with stratification by ARDS severity) of the difference in 28-day mortality for patients with ARDS compared with patients without ARDS were calculated. The unadjusted estimate for increased mortality of ARDS was 15% (95% CI 4% to 26%; p=0.01). The adjusted estimate for increased mortality of ARDS was 15% (95% CI 3% to 26%; p=0.015). The adjusted estimate for increased mortality of severe ARDS was 23% (95% CI 3% to 44%; p=0.028), moderate ARDS was 16% (95% CI 2% to 31%; p=0.031) and mild ARDS was 1% (95% CI −27% to 29%; p=0.952) (figure 3). This was also the case for estimates of 60-day mortality of ARDS (online supplemental appendix 22). The difference in 28-day mortality attributable to ARDS was significant for patients with pneumonia but not significant for septic patients without pneumonia (online supplemental appendix 23).
Forest plot of mortality incidence of ARDS among patients included in the three ICU cohorts of the original analysis. Confounders for which adjustment was made included age, comorbidities, modified sequential organ failure assessment (ie, after excluding its respiratory component) score and cohort. The adjusted estimate for increased mortality of any ARDS was 15% (95% CIs 3% to 26%), mild ARDS was 1% (95% CI −27% to 29%), moderate ARDS was 16% (95% CI 2% to 31%) and severe ARDS was 23% (95% CI 3% to 44%). ARDS was categorised as mild, moderate and severe according to the Berlin definition. ARDS, acute respiratory distress syndrome; ICU: intensive care unit.
Secondary analyses in the ICU cohorts
The estimate for the effect of ARDS on 28-day mortality for non-invasive mechanical ventilation was 22% (95% CI −5% to 49%; p=0.116). The estimate for the effect of ARDS on 28-day mortality for invasive mechanical ventilation was 7% (95% CI −6% to 19%; p=0.278) (online supplemental appendix 24).
After adjustment for daily modified SOFA score, non-invasive mechanical ventilation and invasive mechanical ventilation, an abnormal PF ratio was found to increase the HR of mortality by a factor of 1.40 (95% CI 0.88 to 2.24; p=0.16) (online supplemental appendix 29). Online supplemental appendices 25–28 show PF ratios and outcome status of each patient over the 28-day time period used for this analysis.
Discussion
By accumulating evidence from 47 cohorts involving 56 081 critically ill patients, we found that ARDS was associated with mortality after controlling for underlying risk factor. Then, by taking advantage of three large, well-phenotyped ICU cohorts enrolling 1023 patients, we found that ARDS may have a direct causal link with mortality—with significant differences seen for severe and moderate, but not mild ARDS.
Our prospective study from three ICU cohorts employed innovative statistical approaches from the field of causal inference to estimate the attributable mortality of ARDS. Our adjusted survival analysis accounted for variables which predict both ARDS diagnosis and censoring—namely, age, modified SOFA, comorbidities and cohort. In modifying the total SOFA score to remove the respiratory component, we more accurately quantified the contribution of organ failures other than pulmonary to mortality. The potential for significant bias due to immortal time is low in our estimate, given that the median time from ICU admission to ARDS onset was 0 days for all cohorts. When adjusted results were stratified by ARDS severity, only severe and moderate ARDS had significant estimates for increased mortality (23% and 16%, respectively). There was no significant effect of mild ARDS on mortality (1%). This finding is possibly due to the fact that several mild ARDS cases may improve rapidly, potentially complicating study findings and conclusions.1 74
The results of our secondary analysis of the effect of invasive mechanical ventilation on ARDS suggest decreased 28-day mortality among patients who were treated with invasive mechanical ventilation compared with non-invasive mechanical ventilation at the time of ARDS identification (7% vs 22%); however, this finding did not reach statistical significance. On the other hand, the use of a Cox-WMSM allowed us to incorporate important time-varying confounders of ARDS and mortality, namely, organ failure scores and mechanical ventilation status, while quantifying the effect of an abnormal PF ratio on the risk of mortality. Modelling showed that, after adjustment for modified SOFA score and type of mechanical ventilation, an abnormal PF ratio non-significantly increased the risk of mortality over 28 days by 40%. We hope our techniques using granular data in patients with complex courses might serve as a framework for future study design.
There are several strengths of our prospective study analysis. We used data from three large, well-phenotyped ICU cohorts that differ in their composition of patients with respect to baseline characteristics, permitting a diverse population for analysis. We used Sepsis-3 definitions, which incorporate the most updated understanding of sepsis pathobiology and organ failure.13 The degree of organ failure among patients was quantified using the SOFA score, a model that reflects factors that may influence patient outcome over time and represents an improvement in previously available outcome prediction models.75 In our estimate of the attributable mortality of ARDS, we accounted for potential confounders that predict development of ARDS as well as death (showing the importance of adjusting for sepsis risk factor), while applying statistical approaches that overcome the limitations of logistic regression seen in our meta-analysis. Finally, we demonstrated an approach toward addressing the inherent bias of time-varying confounding of ARDS evolution and mortality. Notably, in both of our analyses, the timing of SOFA as an adjustment variable was carefully considered. In our primary analysis examining the severity of ARDS onset with mortality, the SOFA score used for adjustment was comprised only of data up to the day of ARDS onset. In our secondary analysis examining an exposure of time-varying PF ratios, we used marginal structural models, which were developed to appropriately handle the dual nature of time-varying SOFA scores as mediators and confounders of the relation between time-varying PF ratio and mortality.
Our work has several limitations. In regards to the meta-analysis, multiple definitions of ARDS were employed among included studies, reflecting evolving approaches to create a method of valid and reliable ARDS identification.14 15 To address these differences, we used the appropriate Cochrane tool to assess risk of bias related to adjudication of the diagnosis of ARDS and we carried out a sensitivity analysis of studies with low or moderate bias to find similar results (online supplemental appendix 12). Second, studies included in the systematic review widely varied in terms of ARDS risk factors, while the category of ‘other’ risk factors is heterogeneous itself as it also included ARDS without identifiable risk factors. Such patients without identifiable risk factors may have better outcomes, including a higher proportion of rapidly improving ARDS but not lower mortality, than those with risk factors.74 76 77 Third, not all studies included in the systematic review provided details on the aetiology of sepsis, the age of septic patients or the status of mechanical ventilation.78 A limitation of our prospective study is that the three ICU cohorts differed slightly in their calculation of SOFA score at ARDS identification. A large proportion of subjects had malignancy or non-malignant immunosuppression, perhaps due to the cohorts being large academic centres treating advanced malignancies, organ transplant recipients and other subjects on immune-suppressing therapies. We and other authors have previously shown that immune-compromising conditions are associated with development of ARDS as well as higher mortality.79 80
In conclusion, our comprehensive meta-analysis of 47 cohorts involving over 56 000 critically ill patients suggests that ARDS is associated with mortality after controlling for underlying risk factor, with a stronger association for trauma than for septic patients. Our subsequent analysis of three ICU cohorts using causal inference methods shows that ARDS is a significant risk factor for mortality. When results were stratified by ARDS severity, both severe and moderate ARDS had significant estimates for mortality. In addition to estimating the effect of ARDS on mortality, we demonstrated how longitudinal components of ARDS can be used to estimate the impact a changing ARDS severity level has on mortality. Our study is probably the most thorough attempt so far to establish whether ARDS is associated with attributable mortality and our findings provide information about the extent to which continued funding of ARDS prevention trials has potential to provide survival benefit.
Data availability statement
Data are available upon reasonable request. Deidentified subject data are available upon reasonable request from researchers whose proposed use of data has been approved. Inquiries for reuse of data may be made by emailing lkt9003@med.cornell.edu.
Ethics statements
Patient consent for publication
Ethics approval
The cohorts of critically ill patients were approved by Institutional Review Board of Weill Cornell Medicine (1405015116A005), the Partners Human Research Committee (2008-P-000495) and Asan Medical Center (2011-0001).
Acknowledgments
The authors would like to thank Dr Kuei-Pin Chung (National Taiwan University Hospital, Taipei, Taiwan) for translating Chinese language publications included in this systematic review and meta-analysis. The authors also appreciate Imaani Easthausen who provided the forest plot in figure 2, as well as Dr David Benkeser and Nima Hejazi who assisted with their R package, survtmle. The authors thank Dr Bernard Park, Dr Gordon Rubenfeld, Dr Jesus Villar, Dr Lorraine Ware, Dr Brian Fuller, Dr Marc Moss, Dr Ognjen Gajic, Dr Miriam Treggiari, Dr Matthew Martin, Dr Martin Croce, Dr Olof Brattström, Dr Jason Hoth, Dr Jordan Weinberg, Dr Martin Zielinski, Dr Joseph DuBose, Dr Jeremy Cannon, Dr Ali Salim, Dr Laura Eberhard, Dr Kevin Chung, Dr Gavin Perkins, Dr Thomas Stewart and John Milberg for showing praiseworthy academic attitude by responding to our request for additional information. The authors also thank Dr Gordon Rubenfeld for his valuable scientific insight into this work.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
Contributors IIS conceived the study. LKT, KH and IIS had full access to all data in the study. LKT, KH, ID and IIS contributed to the study concept and design. LKT performed the literature search. LKT, KH, CO, ID, JSH, EJS, DRP, LG-E, AH, MPV, RMB, LEF, J-WH and IIS collected, analysed and interpreted the data. LKT and KH wrote the draft manuscript. LKT, KH, JSH, AMKC and IIS revised the manuscript for important intellectual content. IIS had final responsibility for the decision to submit for publication.
Funding This work was supported by the National Institutes of Health (NIH) grants R01 HL055330 and P01 HL108801 (to AMKC). LKT was supported in part by NIH T32 HL134629 (to FJ Martinez), Weill Cornell CTSC UL1-TR-002384 and KL2-TR-002385 and the Stony Wold-Herbert Fund Fellowship. JSH was supported by the Stony Wold-Herbert Fund Fellowship. IIS was supported by grants from the Hellenic Thoracic Society (2019) and the Hellenic Foundation for Research and Innovation (80-1/15.10.2020).
Competing interests AMKC is a cofounder and stock holder of and serves on the Scientific Advisory Board for Proterris, which develops therapeutic uses for carbon monoxide. He also has a use patent on carbon monoxide. He served as a consultant for an advisory board meeting of Teva Pharmaceutical Industries, July 2018. RMB serves on the Advisory Board for Merck. LEF reports clinical trials support from Asahi Kasei Pharma America. None declared: LKT, KH, CO, ID, JSH, EJS, DRP, LG-E, AH, MPV, JWH and IIS.
Provenance and peer review Not commissioned; externally peer reviewed.
Linked Articles
- Airwaves