Mortality associated with acute respiratory distress syndrome, 2009-2019: a systematic review and meta-analysis

Abstract

Background: Acute respiratory distress syndrome (ARDS) occurs commonly in intensive care units. The reported mortality rates in studies evaluating ARDS are highly variable.

Objective: To investigate mortality rates due to ARDS from before the 2009 H1N1 influenza pandemic began until the start of coronavirus disease 2019 (COVID-19) pandemic.

Design: We performed a systematic search and then ran a proportional meta-analysis for mortality. We ran our analysis in three ways: for randomised controlled trials only, for observational studies only, and for randomised controlled trials and observational studies combined.

Data sources: MEDLINE and Embase, using a highly sensitive criterion and limiting the search to studies published from January 2009 to December 2019.

Review methods: Two of us independently screened titles and abstracts to first identify studies and then complete full text reviews of selected studies. We assessed risk of bias using the Cochrane RoB-2 (a risk-of-bias tool for randomised trials) and the Cochrane ROBINS-1 (a risk-of-bias tool for non-randomised studies of interventions).

Results: We screened 5844 citations, of which 102 fully met our inclusion criteria. These included 34 randomised controlled trials and 68 observational studies, with a total of 24 158 patients. The weighted pooled mortality rate for all 102 studies published from 2009 to 2019 was 39.4% (95% CI, 37.0–41.8%). Mortality was higher in observational studies compared with randomised controlled trials (41.8% [95% CI, 38.9–44.8%] v 34.5% [95% CI, 30.6–38.5%]; P = 0.005).

Conclusions: Over the past decade, mortality rates due to ARDS were high. There is a clear distinction between mortality in observational studies and in randomised controlled trials. Future studies need to report mortality for different ARDS phenotypes and closely adhere to evidence-based medicine.

PROSPERO registration: CRD42020149712 (April 2020).

Acute respiratory distress syndrome (ARDS) accounts for about 10% of intensive care unit (ICU) admissions, with a mortality rate ranging from 35% to 45%.¹ A meta-analysis showed that the mortality rates in observational studies and randomised controlled trials (RCTs) remained static from the implementation of the 1994 American–European Consensus Conference (AECC) definition of ARDS until 2006.² Since the publication of that meta-analysis, an influenza pandemic with a high prevalence of ARDS occurred, a new definition of ARDS was developed³ and several landmark randomised controlled trials evaluating various interventions have been published.4, 5, 6, 7, 8, 9, 10

Over the past two decades, lung-protective ventilation strategies (tidal volume [Vt] 6–8 mL/kg predicted bodyweight and low plateau airway pressure [Pplat] < 30 cmH₂O) have been the mainstay of ARDS management. Adherence to lung-protective ventilation strategies has consistently been shown to improve patient survival.¹¹ Use of prone position ventilation in patients with moderate–severe ARDS is the only other therapeutic intervention that has been associated with improved survival.⁵ Continuous use of neuromuscular blocking agents (NMBAs) has led to disparate results in the two major trials that have evaluated their use.6, 10 Other interventions include the use of higher positive end expiratory pressure (PEEP), inhaled pulmonary vasodilators and diuretics; the benefits of these include improvements in oxygenation and shorter duration of mechanical ventilation, but their use has not been associated with a mortality benefit.12, 13, 14, 15, 16 Based on the current evidence, early and consistent use of extracorporeal membrane oxygenation (ECMO) cannot be justified, and this therapy needs to be considered after failure of conventional mechanical ventilation.7, 17, 18 But more importantly, and independent of the quality or strength of evidence, the application of all these therapies in ARDS is inconsistent¹⁹ and is predominantly influenced by physician comfort and discretion.¹

To study the evolution in ARDS outcomes, we conducted a systematic review and meta-analysis to investigate mortality associated with ARDS from before the 2009 H1N1 influenza pandemic began until and the start of coronavirus disease 2019 (COVID-19) pandemic. We also evaluated temporal changes in mortality rates over the study period.

Methods

Search strategy

We electronically searched MEDLINE and Embase using a highly sensitive strategy to identify the relevant studies published during the period January 2009 to December 2019. Full text versions of potentially relevant articles were obtained for review, and all references cited in these articles were inspected to supplement our search. Details of the search strategy are reported in the Online Appendix. We limited our search to articles published in English.

Study selection

We included RCTs and observational studies in our review. Using standardised criteria, two of us (DS and KS) reviewed titles and abstracts identified by the search strategy independently and in duplicate, and we retrieved studies that DS or KS thought were relevant for full text review. Disagreements between reviewers in study selection and data extraction were resolved by one of us (AD, the senior author). We selected RCTs and observational studies that had enrolled at least 50 adults who had acute lung injury or ARDS and reported mortality. We only included studies where 100% of the sample met any criteria for ARDS. We excluded reports available only in abstract form, duplicate reports and reports on animal studies.

Data extraction

Two of us (DS and KS) independently extracted all the data from the selected studies. Extracted data included: geography (country, continent, institutions); duration of study (months, years); ARDS definition used (Berlin, AECC or other); patient characteristics (age, severity of illness, ratio between arterial partial pressure of oxygen [Pao₂] and fraction of inspired oxygen [Fio₂], Acute Physiology and Chronic Health Evaluation [APACHE] II score, Simplified Acute Physiology Score [SAPS], Sequential Organ Failure Assessment [SOFA] score]); ventilator specific variables (Vt, PEEP, Pplat); adjunctive therapy (inhaled vasodilators, NMBAs, high frequency oscillatory ventilation [HFOV], prone position, ECMO); and all-cause mortality. Driving pressure was calculated using reported PEEP and Pplat values.²⁰ The data from RCTs and observational studies that contained multiple arms were combined for analysis (if data were reported as mean and standard deviation). If a study contained an arm dedicated to investigating an adjunctive therapy versus control, only the control arm from that study was included, to pertain to all ARDS patients. Any missing data are reported in the study table in the Online Appendix. The primary outcome was short term mortality. Short term mortality was defined as hospital mortality where reported, as it was the most often reported. If not reported, then ICU mortality, 90-day mortality, 60-day mortality, or 28- or 30-day mortality was substituted, in this order of preference.

Risk of bias assessment

To assess for risk of bias, two of us (DS and KS) independently reviewed all included studies. The Cochrane RoB-2 tool was used to assess RCTs.²¹ for each RCT, we evaluated five domains and determined an overall risk of bias. The Cochrane ROBINS-1 tool was used to assess observational studies.²² for each observational study, we evaluated seven domains and determined an overall risk of bias. The overall risk of bias for both RCTs and observational studies was determined by the highest risk allotment in any of the evaluated domains.

Data analysis

Baseline characteristics between observational studies and RCTs were compared using the Student t test and χ² test for continuous and categorical variables, respectively. We performed a proportion meta-analysis using random-effects models to obtain pooled estimates of mortality and 95% confidence intervals (CIs) for all RCTs and all observational studies separately.²³ We used the Cochran Q statistic and I² statistic to test for heterogeneity among studies.²⁴ We also ran a cumulative mortality analysis using median year of enrolment to evaluate the mortality trends from 2009 to 2019.²⁵ To explore the heterogeneity in mortality among studies, we performed logistic meta-regression for clinically relevant variables (Pao₂/Fio₂ ratio, Vt, PEEP, Pplat, driving pressure, mean age, APACHE II score, SOFA, use of NMBAs, use of prone position, and use of ECMO) in RCTs and observational studies separately.26, 27, 28 A P value of ≤ 0.05 was considered to be statistically significant. Funnel plot analysis and the Egger's test were used to check for publication bias.29, 30 This was done in subgroups to account for heterogeneity, whereby RCTs were divided by Pao/Fio₂ ratio and observational studies were divided by the number of patients. The statistical analysis was conducted using Stata 15.1 (StataCorp). Two sensitivity analyses were also conducted to address certain aspects of our study and their effects on our findings; in one of these we combined the intervention and control arms of studies which focused on one adjunctive therapy, and in the other we only included studies that reported data on hospital mortality.

Results

Study selection

Our search strategy yielded 5844 citations published from 2009 to 2019 after duplicates were excluded. We completed detailed evaluations of the full text versions of 1660 articles and included 102 articles in our qualitative assessment (Figure 1). The final selection of 102 articles included 34 RCTs (33.3%) and 68 observational studies (66.7%). In total, 24 158 patients were enrolled in the included studies. Twelve of the included articles were reports of studies that investigated an adjunctive therapy versus control, so we only included the control arms from these studies in our primary analysis. However, in one of our sensitivity analyses, we included both arms from these 12 studies, which increased the total number of patients to 26 142. We also performed a sensitivity analysis of studies reporting hospital mortality; this consisted of 17 RCTs (50.0% of all included RCTs) and 39 observational studies (57.4% of all included observational studies), totalling 14 052 patients.

Figure 1.

Study selection process

ARDS = acute respiratory distress syndrome. * More than one reason was possible for study exclusion; further detail on included studies is available in the Online Appendix.

Study characteristics

To diagnose ARDS, 47 studies (46.1%) used the AECC criteria, 52 studies (51.0%) used the Berlin criteria, two studies (1.9%) used either criteria and one study (1.0%) used the Chinese critical care medicine definition. Mean age was reported in 83 studies (81.4%). The mean baseline Pao₂/Fio₂ ratio was reported in 84 studies (82.4%), mean APACHE II score was reported in 45 studies (44.1%) and mean SOFA score was reported in 41 studies (40.2%). Tidal volume in mL/kg was reported in 68 studies (66.7%), PEEP was reported in 70 studies (68.6%), Pplat was reported in 55 studies (53.9%) and we were able to calculate driving pressure for 48 studies (47.1%). The use of inhaled vasodilators, NMBAs, HFOV, prone position and ECMO was reported in 27 (26.5%), 32 (31.4%), 13 (12.7%), 35 (34.3%) and 27 (26.5%) studies, respectively (Table 1). There were no differences in baseline patient characteristics, severity of initial illness and use of ventilatory strategies between the RCTs and observational studies (Table 1). The lack of asymmetry on the funnel plot and the result of the Egger's test imply that publication bias did not alter the results in any of the subgroup analyses (Online Appendix, e-Figures 7–14).

Table 1.

Study characteristics^*

Characteristics	All studies (N = 102)	Randomised controlled trials (n = 34)	Observational studies (n = 68)	P*
ARDS criteria†
Berlin	52 (51.0%)	12 (35.3%)	40 (58.8%)
AECC	47 (46.1%)	21 (61.8%)	26 (38.2%)
Berlin or AECC	2 (1.9%)	0	2 (3.0%)
Other	1 (1.0%)	1 (2.9%)	0
Mortality rate	39.4%	34.5%	41.8%	0.005
Age, years‡	55.6 ± 7.3 (81.4%)	56.2 ± 4.3 (97.1%)	55.3 ± 8.7 (73.5%)	0.56
Pao2/Fio2 ratio, mmHg‡	132.0 ± 33.8.0 (82.4%)	134 ± 30.5 (94.1 %)	130.8 ± 35.9 (76.5%)	0.67
APACHE II score*	22.5 ± 3.7 (44.1%)	21.5 ± 3.2 (47.1%)	23.0 ± 3.8 (42.6%)	0.19
SOFA score‡	9.6 ± 1.9 (40.2%)	9.3 ± 1.7 (44.1%)	9.8 ± 2.0 (38.2%)	0.37
Tidal volume, mL/kg‡	7.17 ± 0.88 (66.7%)	7.04 ± 0.87 (91.2%)	7.29 ± 0.88 (54.4%)	0.24
PEEP, cmH20‡	10.4 ± 2.0 (68.6%)	11.0 ± 2.2 (82.4%)	10.0 ± 1.7 (61.8%)	0.036
Plateau pressure, cmH20‡	25.6 ± 2.7 (53.9%)	25.7 ± 2.6 (73.5%)	25.5 ± 2.8 (44.1%)	0.79
Driving pressure, cmH20‡	15.3 ± 2.2 (47.1%)	15.1 ± 2.3 (67.7%)	15.5 ± 2.2 (36.8%)	0.57
Inhaled vasodilators†	27 (26.5%)	9 (26.5%)	18 (26.5%)	0.12
NMBAs†	32 (31.4%)	14 (41.2%)	18 (26.5%)	0.57
HFOV†	12 (11.8%)	4 (11.8%)	8 (11.8%)	0.50
Prone positioning†	35 (34 .3%)	13 (38.2%)	22 (32.3%)	0.64
ECMO†	25 (24.5%)	7 (20.6%)	18 (26.5%)	0.06
Continent†
Europe	38 (37.3%)	15 (44.1%)	23 (33.8%)
North America	23 (22.5%)	8 (23.5%)	15 (22.1%)
Asia	31 (30.4%)	6 (17.6%)	25 (36.8%)
Australia	1 (1.0%)	0	1 (1.5%)
South America	2 (2.0%)	0	2 (2.9%)
Global	7 (6.8%)	5 (14.7%)	2 (2.9%)

AECC = American–European Consensus Conference. APACHE = Acute Physiology and Chronic Health Evaluation. ARDS = acute respiratory distress syndrome. ECMO = extracorporeal membrane oxygenation. HFOV = high frequency oscillatory ventilation. NMBAs = neuromuscular blocking agents. Pao₂/Fio₂ = ratio between arterial partial pressure of oxygen and fraction of inspired oxygen. PEEP = positive end expiratory pressure. SOFA = Sequential Organ Failure Assessment.

^*By Student t test.

^†Data are number (percentage) of studies reporting characteristic.

^‡Data are mean ± standard deviation for characteristic (percentage of studies reporting characteristic).

Quality assessment

The risk of bias was low in 18 RCTs, moderate in 13 RCTs and high in three RCTs (Online Appendix, e-Table 3). High risk of bias in RCTs was driven mainly by deviation from the intended intervention. The risk of bias was low in two observational studies, moderate in 53 observational studies and high in 13 observational studies (Online Appendix, e-Table 4). High and moderate risk of bias in observational studies was driven mainly by confounding owing to study design.

Mortality

Hospital mortality (56 studies) was the most reported variable relating to short term mortality in the included studies. In the remaining 46 studies, it was substituted with ICU mortality (ten studies), 90-day mortality (six studies), 60-day mortality (ten studies), and 28- or 30-day mortality (20 studies). The weighted pooled mortality for all 102 studies included in our study was 39.4% (95% CI, 37.0–41.8°%). Mortality was higher in observational studies compared with RCTs (41.8% [95% CI, 38.9–44.8%] v 34.5% [95% CI, 30.6–38.5%]; P = 0.005; Figure 2 and Table 1). There was significant heterogeneity among the included studies (I² = 93.2%; P < 0.01). This heterogeneity persisted across both observational studies (I² = 92.3%; P < 0.01) and RCTs (I² = 93.4%; P < 0.01) (Figure 2). The reported mortality rates from the included studies are listed in Figure 2, in chronological order according to publication year. Mortality did not differ significantly by continent where the study was performed when analysed by one-way analysis of variance (ANOVA; F[5,96] = 1.62; P =0.16).

Figure 2.

Forest plot of mortality for all studies included, grouped by study design (randomised controlled trial or observational study)*

* Studies are listed in chronological order by year of publication (shown in parentheses). The effect size and 95% CI for each study are represented by a closed diamond and horizontal line, respectively; the pooled estimates for all observational studies, all randomised controlled trials and all studies are represented by an open diamond.

The reported mortality rate did not change based on the sensitivity analysis. The first sensitivity analysis, in which we included the interventional and control arms of the 12 studies that evaluated a particular adjunctive therapy, resulted in a pooled mortality rate of 39% (95% CI, 37–41%) (Online Appendix, e-Figure 15). Consistent with the primary analysis, the weighted pooled mortality was higher in observational studies compared with RCTs (42% [95% CI, 39–44%] v 34% [95% CI, 30–38%]; P < 0.01). The second sensitivity analysis, in which we only included the 56 studies that reported hospital mortality, resulted in a pooled mortality rate of 40% (95% CI, 37–43%) (Online Appendix, e-Figure 16). Again, mortality was higher in observational studies compared with RCTs (43% [95% CI, 39–46%] v 33% [95% CI, 28–38%]; P < 0.01). The cumulative mortality analysis, which we conducted using the median year of enrolment (which ranged from 2000 to 2018), shows the evolution of mortality in studies published from 2009 to 2019, excluding two studies which did not mention enrolment dates31, 32 (Figure 3); on visual inspection of these data, there appeared to be an initial decrease in mortality rate, after which the rate stabilised around the mean.

Figure 3.

Cumulative mortality analysis from 2000 to 2018*

* Studies are listed in chronological order by median year of enrolment (shown in parentheses). The effect size and 95% CI for each study are represented by a closed diamond and horizontal line, respectively

In our meta-regression analysis of RCTs, the initial Pao₂/Fio₂ ratio and SOFA scores were associated with mortality, but mean age, APACHE II score, ventilator-specific variables and adjunctive therapies did not have any impact on mortality (Online/Appendix, e-Table 5). In our meta-regression analysis of observational studies, no variables displayed any association with mortality (Online Appendix, e-Table 6).

Discussion

Our meta-analysis demonstrates a very small reduction in ARDS-associated mortality since 2009. Comparing data from our analysis with previously reported results, the cumulative reported mortality rate has dropped 5 percentage points from 44.3% to 39.3% over the past decade.² As has been reported in other studies, mortality rates are consistently higher in observational studies compared with RCTs.2, 33

Temporal trends in mortality should display advancements in ARDS management and their impact on the care of patients with ARDS. It has been over a decade since the rate of 44.3% was reported,² and since then significant changes in the management of ARDS have been studied and instituted. These include, but are not limited to, prone positioning (the Proning Severe ARDS Patients [PROSEVA] study showed a reduction in mortality among patients with moderate–severe ARDS), use of NMBAs (the ARDS et Curarisation Systematique [ACURASYS] showed a reduction in mortality among patients with moderate–severe ARDS), and use of ECMO (the ECMO to Rescue Lung Injury in Severe ARDS [EOLIA] study showed increased survival with ECMO). However, despite these landmark studies, they have not necessarily translated into a mortality reduction in published reports. This highlights the importance of accumulating evidence showing that ARDS is not a homogenous process and that there are distinct phenotypes with varying disease trajectories, heterogeneity in treatment effect and distinct differences in outcomes.34, 35 Recent work by various groups has shown distinct differences in the amount of inflammation among patients, and highlighted the presence of two phenotypes that are associated with different clinical and biological characteristics, mortality rates, responses to treatment, and numbers of ventilator-free days.34, 36, 37 These two phenotypes also differ in terms of treatment effect related to PEEP, fluid status and response to statins, which underscores the importance of finding patient populations which may benefit from different interventions.35, 37, 38 In addition, heterogeneity in treatment effect emphasises the need to study ARDS differently and may be a key factor behind the lack of a significant decrease in mortality over the past decade.

ARDS is an under-recognised disease process with significant geographical and centre- specific variability.1, 19 The Large Observational Study to Understand the Global Impact of Severe Acute Respiratory Failure (LUNG SAFE) evaluated the incidence and outcomes of ARDS across 459 ICUs in 2016.¹ The study found that ARDS was a highly under-recognised disease process, with 40% of patients meeting the criteria not being diagnosed. This under-recognition is problematic given that it suggests that the diagnosis of ARDS is often delayed, with a concomitant high likelihood of delay in treatment. This is especially concerning given that many of the ARDS treatments with proven benefit have only demonstrated a benefit when instituted early in ARDS.4, 9, 29 Another key factor in the lack of decrease in mortality could be insufficient standardisation of ARDS management. A recently published study evaluated variation in early management of ARDS across 29 academic and community centres in the United States.¹⁹ It found substantial variation in the use of initial lung protective ventilation (0–65%), use of adjunctive therapies (27.1–96.4%), and 28-day mortality (16.7–73.3%). With underdiagnosis of ARDS leading to a delay in use of proven treatments, along with high variability in instituting those treatments, we have seen stagnant mortality rates over the past decade.

Clinical heterogeneity and biological differences make it difficult to reflect the advancements in ARDS treatment with meta-analytical techniques. As such, the overall mortality rate associated with ARDS is much higher than the suggested benchmark for trials a decade ago.³⁹

In our meta-analysis, we found a significant difference in reported mortality when comparing RCTs and observational studies. This difference is not surprising as RCTs usually ensure strict adherence to protocols, with the probable exclusion of patients with a poor prognosis.40, 41 In a systematic review of 14 studies that was published in 2020, it was found that standardised care protocols reduce mortality when compared with usual care.⁴² However, there is a paucity of evidence on improving implementation of proven treatments and protocols. In a systematic review published in 2021, in which more than two decades literature had been searched, fewer than ten studies on interventions aimed to increase adherence to the use of lung-protective ventilation and prone positioning were identified.⁴³ Given the availability of evidence-based interventions conferring a survival benefit, we had expected to see a larger focus on implementation and subsequent reduction in mortality. This has become of greater importance given the ongoing COVID-19 pandemic. With hospitals heavily burdened by ARDS worldwide, the absence of a marked improvement in the past decade is even more glaring and consequential.

To explore the high degree of heterogeneity in treatment, we searched for treatment modifiers using meta-regression techniques. In observational studies we did not find any statistically significant associations between the variables we tested and mortality. In RCTs, however, where stricter protocols are followed, we found that a lower initial Pao₂/Fio₂ ratio and a higher initial SOFA score were associated with mortality. This reflects the fact that trials which included patients who were more unwell had higher mortality rates. Although observational studies have an inherent lack of strict protocols, it is important that they be designed with this variability and heterogeneity kept in mind.⁴⁴ Designing studies with more attention to inclusion criteria may help stratify similar patients and account for the heterogeneity in mortality rates in published studies.

Our systematic review and meta-analysis has several strengths. We conducted a comprehensive literature search using broad search terms. We included studies from institutions across five continents and report an international evaluation on the trend in ARDS mortality, whereas a study published in 2019 reported on the trend in RCT mortality in the United States alone.⁴⁵ Our inclusion criteria were carefully predefined and implemented in a methodological fashion. In addition, we only included studies which exclusively evaluated ARDS patients. The reporting of ARDS mortality is a complex multitiered dynamic and is affected by lack of diagnosis, centre-specific variability, different phenotypes, and lack of treatment standardisation. In this article, we have highlighted these intricacies so that future studies can present outcomes more scrupulously.

Despite the strengths of our meta-analysis, there are potential limitations. First, we used hospital mortality as the primary mortality type in our analysis as it was most frequently reported (56 of 102 studies). With a reported incidence of mortality of 3–15% in ARDS after discharge from the ICU,46, 47 use of hospital mortality as the primary mortality type may have affected our overall reported ARDS-related mortality rate. When hospital mortality was not available, we substituted it with ICU, 90-day, 60-day, or 28- or 30-day mortality. However, to tackle this potential inadequacy, we conducted a sensitivity analysis that only included studies reporting hospital mortality, and this produced almost identical results. Finally, we included all criteria for ARDS diagnosis in our systematic review and meta-analysis, and we did not stratify results based on ARDS severity as it was seldom reported.

Conclusions

The mortality rate due to ARDS has changed very little over the past decade. Over this period, significantly lower mortality rates were found in RCTs than in observational studies. The lack of a significant change in overall mortality rates highlights the need to address factors driving the heterogenous nature of ARDS during study design for more accurate reporting of mortality. Future studies should focus on exploring different ARDS phenotypes and adhering to evidence-based medicine when reporting mortality.

Acknowledgements: Abhijit Duggal has received grants from the National Institutes of Health (No. 5U01HL123023) and the Centers for Disease Control and Prevention (No. 1.200-2016-91801), none related to this publication.

Competing interests

Abhijit Duggal is a member of the Steering Committee for ALung Technologies, not related to this publication. All other authors declare that they do not have any potential conflict of interest in relation to this manuscript.