Introduction

The acute respiratory distress syndrome (ARDS) is a major problem in the intensive care unit (ICU) with a prevalence of 10% and an in-hospital mortality rate of 40% [1, 2]. ARDS pathophysiology is based on a triad of alveolar-capillary membrane injury, high permeability alveolar oedema, and migration of inflammatory cells [3]. This triad is not routinely measured in clinical practice. Therefore, arterial hypoxemia and bilateral opacities on chest imaging following various clinical insults are used as clinical surrogates in the American European Consensus Conference (AECC) definition and the newer Berlin definition of ARDS [4, 5].

Histologically, ARDS is characterized by diffuse alveolar damage (DAD). The correlation between a clinical and histological diagnosis of ARDS is poor [6]. Only half of clinically diagnosed patients with ARDS have histological signs of DAD at autopsy [7,8,9,10]. The number of risk factors for ARDS and consequently the heterogeneous histological substrates found in patients with clinical ARDS have been recognized as a major contributor to the negative randomized controlled trial results among patients with ARDS [11].

It has been suggested that the addition of biomarkers to the clinical definition of ARDS could reduce ARDS heterogeneity by the identification of subgroups [12,13,14,15]. A retrospective latent class analysis of large randomized controlled trials identified two ARDS phenotypes largely based on ARDS biomarkers combined with clinical parameters [16, 17]. These phenotypes responded differently to the randomly assigned intervention arms. Prospective studies are required to validate these ARDS phenotypes and their response to interventions. The set of ARDS biomarkers to prospectively define these phenotypes remains to be established.

Numerous biomarkers and their pathophysiological role in ARDS have been described [12, 18]. In an earlier meta-analysis, biomarkers for ARDS development and mortality were examined in univariate analysis [19]. However, pooling of univariate biomarker data may result in overestimation of the actual effect. For this reason, we conducted a systematic review and included all biomarkers that were multivariately associated with ARDS development or mortality. This study provides a synopsis of ARDS biomarkers that could be used for future research in the identification of ARDS phenotypes.

Methods

This systematic review was prospectively registered in PROSPERO International Prospective Register of Systematic Reviews (PROSPERO identifier CRD42017078957) and performed according to the Transparent Reporting of Systematic Reviews and Meta-analyses (PRISMA) Statement [20]. After the search strategy, two reviewers (PZ, PS, and/or WG) separately performed study eligibility criteria, data extraction, and quality assessment. Any discrepancies were resolved by consensus, and if necessary, a third reviewer was consulted.

We searched for studies that included biomarkers that were associated with ARDS development in critically ill patients at risk for ARDS and mortality in the ARDS population in multivariate analyses adjusted for background characteristics. We did not perform a meta-analysis, because the raw data in all studies was either not transformed or log transformed resulting in varying risk ratios and confidence intervals. In addition, the majority of studies used different biomarker concentration cut-offs, resulting in varying concentration increments for risk ratios. Lastly, the number of days until mortality and variables used in multivariate analysis differed between studies. For these reasons, we limited this study to a systematic review, as the multivariate odds ratios were not comparable and pooling would result in non-informative estimates [21].

Search strategy

We performed a systematic search in Embase, MEDLINE, Web of Science, Cochrane CENTRAL, and Google Scholar from inception until 30 July 2018 with assistance from the Erasmus MC librarian. The search was later updated to 6 March 2020. A detailed description of the systematic search string is presented in Additional file 1. In addition, the reference lists of included studies and recent systematic reviews were screened to identify additional eligible studies.

Study eligibility criteria

All retrieved studies were screened on the basis of title and abstract. Studies that did not contain adult patients at risk for ARDS or with ARDS and any biomarker for ARDS were excluded. The following eligibility criteria were used: human research, adult population, studies in which biomarkers were presented as odds ratios (OR) or risk ratios in multivariate analysis with ARDS development or mortality as outcome of interest, peer-reviewed literature only, and English language. Studies comparing ARDS with healthy control subjects, case series (< 10 patients included in the study), and studies presenting gene expression fold change were excluded.

Data extraction

A standardized form was used for data extraction from all eligible studies. Two clinical endpoints were evaluated in this study: development of ARDS in the at-risk population (patients that did develop ARDS versus critically ill patients that did not) and mortality in the ARDS population (survivors versus non-survivors). The following data were extracted: study design and setting, study population, sample size, the definition of ARDS used in the study, outcome, risk ratio with 95% confidence interval in multivariate analyses, and the variables used in the analyses. In addition, the role of the biomarker in ARDS pathophysiology as reported by the studies was extracted and divided into the following categories: increased endothelial permeability, alveolar epithelial injury, oxidative injury, inflammation, pro-fibrotic, myocardial strain, coagulation, and others. Subsequently, the relative frequency distribution of biomarker roles in ARDS pathophysiology was depicted in a bar chart.

Quality assessment

Methodological quality of the included studies was assessed with the Newcastle-Ottawa Scale (NOS) for assessing the quality of nonrandomized studies in systematic reviews and meta-analyses [22]. Items regarding patient selection, comparability, and outcome were assessed using a descriptive approach, and a risk-of-bias score, varying between 0 (high risk) and 9 (low risk), was assigned to each study.

Results

Literature search and study selection

A total of 8125 articles were identified by the initial search and 972 by the updated search (Fig. 1). After removal of duplicates and reviewing titles and abstracts, we selected 438 articles for full-text review. A total of 86 studies was eligible for data extraction: 35 for ARDS development and 53 for ARDS mortality.

Fig. 1
figure 1

PRISMA flow diagram for a systematic search

Full size image

Study characteristics and quality assessment

The study characteristics of the 35 studies for ARDS development are presented in Table 1. A total of 10,667 critically ill patients was at risk for ARDS, of whom 2419 (24.6%) patients developed ARDS. The majority of studies used the Berlin definition of ARDS (21/35), followed by the AECC criteria of ARDS (13/35). The included biomarkers were measured in plasma, cerebrospinal fluid, and bronchoalveolar lavage fluid. In all studies, the first sample was taken within 72 h following ICU admission.

Table 1 Study characteristics for ARDS development
Full size table

The study characteristics of the 53 studies for ARDS mortality are presented in Table 2. A total of 15,344 patients with ARDS were included with an observed mortality rate of 36.0%. The AECC definition of ARDS was used in the majority of included studies (39/53). The included biomarkers were measured in plasma, bronchoalveolar lavage fluid, and urine. All samples were taken within 72 h following the development of ARDS.

Table 2 Study characteristics for ARDS mortality
Full size table

The median quality of the included publications according to the NOS was 7 (range 4–9) for ARDS development and 8 (range 5–9) for ARDS mortality (Additional file 2).

Biomarkers associated with ARDS development in the at-risk population

A total of 37 biomarkers in plasma, 7 in cerebrospinal fluid, and 1 in bronchoalveolar lavage fluid were assessed in multivariate analyses (Table 3). Five studies examined angiopoeitin-2 (Ang-2) and seven studies examined receptor for advanced glycation end products (RAGE). In all studies, high plasma levels of Ang-2 and RAGE were significantly associated with an increased risk of ARDS development in the at-risk population. Similar results were seen for surfactant protein D (SpD) in plasma in all three studies that assessed SpD. In contrast, biomarkers for inflammation as C-reactive protein (CRP), procalcitonin, interleukin-6, and interleukin-8 were not clearly associated with ARDS development. The majority of biomarkers in plasma are surrogates for inflammation in ARDS pathophysiology (Fig. 2).

Table 3 Risk ratios for ARDS development in the at-risk population
Full size table
Fig. 2
figure 2

Biomarker role in ARDS pathophysiology

Full size image

Biomarkers associated with mortality in the ARDS population

A total of 49 biomarkers in plasma, 8 in bronchoalveolar lavage fluid, and 3 in urine were included in this study (Table 4). Ang-2, CRP, interleukin-8 (IL-8), RAGE, SpD, and Von Willebrand factor (VWF) in plasma were assessed in four or more studies. However, none of these biomarkers was associated with ARDS mortality in all four studies. Similarly to biomarkers in ARDS development, the majority of biomarkers for ARDS mortality in plasma had a pathophysiological role in inflammation (Fig. 2). The majority of biomarkers measured in bronchoalveolar lavage fluid had a pro-fibrotic role in ARDS pathophysiology.

Table 4 Risk ratios for ARDS mortality in the ARDS population
Full size table

Discussion

In the current systematic review, we present a synopsis of biomarkers for ARDS development and mortality tested in multivariate analyses. We did not perform a meta-analysis because of severe data heterogeneity between studies. Upon qualitative inspection, we found that high levels of Ang-2 and RAGE were associated with ARDS development in the at-risk population. None of the biomarkers assessed in four or more studies was associated with an increased mortality rate in all studies. The majority of plasma biomarkers for both ARDS development and mortality are surrogates for inflammation in ARDS pathophysiology.

Previously, Terpstra et al. [19] calculated univariate ORs from absolute biomarker concentrations and performed a meta-analysis. They found that 12 biomarkers in plasma were associated with mortality in patients with ARDS. However, a major limitation of their meta-analysis is that these biomarkers were tested in univariate analyses without considering confounders as disease severity scores. Given the high univariate ORs as compared to the multivariate ORs found in this systematic review, the performance of these biomarkers is likely to be overestimated. Jabaudon et al. [109] found in an individual patient data meta-analysis that high concentrations of plasma RAGE were associated with 90-day mortality independent of driving pressure or tidal volume. However, they could not correct for disease severity score as these differed between studies. Unfortunately, we were unable to perform a meta-analysis on multivariate data because of heterogeneity of the included studies, as transformation of raw data, biomarker concentration cut-offs, time until outcome, and the variables used in the multivariate analyses varied widely between studies. This could be an incentive to standardize the presentation of ARDS biomarker research in terms of statistics and outcome for future analyses or to make individual patient data accessible.

ARDS biomarkers are presumed to reflect the pathophysiology of ARDS, characterized by alveolar-capillary membrane injury, high permeability alveolar oedema, and migration of inflammatory cells [3]. Previously, Terpstra et al. [19] proposed that biomarkers for ARDS development were correlated with alveolar tissue injury, whereas biomarkers for ARDS mortality correlated more with inflammation. In this systematic review, we found that the majority of biomarkers tested for both ARDS development and mortality were surrogates for inflammation. However, following qualitative inspection, biomarkers for inflammation were not evidently associated with either ARDS development or mortality. In contrast, markers for alveolar epithelial injury (plasma RAGE and SpD) and endothelial permeability (plasma Ang-2) seem to be associated with ARDS development. Therefore, we should consider how we intend to use (a set of) biomarkers in patients with ARDS.

A biomarker for ARDS development should be specific for ARDS, i.e. a biomarker that reflects alveolar injury or alveolar-capillary injury. Half of plasma biomarkers for ARDS development included in this study reflected inflammation. An increase in inflammatory biomarkers is known to correlate with increased disease severity scores [71, 97, 110]. In turn, the majority of studies in this review found significantly higher disease severity scores in the critically ill patients that eventually developed ARDS. Thus, plasma biomarkers for inflammation rather represented an estimation of disease severity and its associated increased risk for the development of ARDS. In addition, biomarkers for inflammation in plasma lack the specificity to diagnose ARDS, as they are unlikely to differentiate sepsis with ARDS from sepsis without ARDS. In contrast, locally sampled biomarkers for inflammation, for example in the alveolar space, could potentially diagnose ARDS [111]. Biomarkers used for ARDS mortality or for the identification of less heterogeneous ARDS phenotypes do not require to be ARDS specific, provided that they adequately predict or stratify patients with ARDS.

The heterogeneity of ARDS has been recognized as a major contributor to the negative randomized controlled trial results among patients with ARDS [11]. Therefore, it is necessary to identify homogeneous ARDS phenotypes that are more likely to respond to an intervention. This is known as predictive enrichment [112]. Previously, patients with ARDS have been successfully stratified based on clinical parameters, such as ARDS risk factor (pulmonary or extra-pulmonary) or PaO2/FiO2 ratio [113]. ARDS biomarkers could be used to stratify patients with ARDS based on biological or pathophysiological phenotype. For example, trials of novel therapies designed to influence vascular permeability may benefit from preferentially enrolling patients with high Ang-2 concentrations. Recently, clinical parameters have been combined with a set of biomarkers in a retrospective latent class analysis. In three trials, two distinct phenotypes were found: hyperinflammatory and hypoinflammatory ARDS [16, 17]. Patients with the hyperinflammatory phenotype had reduced mortality rate with higher positive end-expiratory pressures and with liberal fluid treatment, whereas the trials themselves found no difference between the entire intervention groups. The next step is to validate the identification of ARDS phenotypes based on latent class analysis in prospective studies. An adequate combination of biomarkers and clinical parameters remains to be established. Until now, there is no list of biomarkers that are associated with ARDS development or mortality independently of clinical parameters. This systematic review may guide the selection of ARDS biomarkers used for predictive enrichment.

This systematic review has limitations. First, the intent of this systematic review was to perform a meta-analysis. However, we decided not to perform a meta-analysis, as the biomarker data handling and outcomes varied widely among studies, and pooling would have resulted in a non-informative estimate [21]. Arguably, this is a positive result, as it refrains us from focusing on the few biomarkers that could be pooled in a meta-analysis and guides us into a direction were multiple biomarkers combined with other parameters are of interest. In a heterogeneous syndrome as ARDS, the one biomarker probably does not exist. Second, the first sampling moment varied between sampling at ICU admission until 72 h following ICU admission. Initially, ARDS is characterized by an exudative phase followed by a second proliferative phase and late fibrotic phase [3]. The moment of sampling likely influences biomarker concentrations, as both alveolar membrane injury and inflammation increase during the exudative phase. This is also seen in six biomarkers that have been measured at separate days, resulting in a significant change in adjusted OR for four biomarkers (Table 4) [61, 98, 104, 105]. Third, the aim of this systematic review was to assess the independent risk effects of biomarkers measured in various bodily fluid compartments. However, the majority of studies assessed biomarkers in plasma. It remains to be answered whether other bodily fluid compartments, for example from the airways and alveolar space themselves, might outperform ARDS biomarkers in plasma, especially for ARDS development. Fourth, all studies found in this systematic review used a clinical definition of ARDS as standard for ARDS diagnosis. Given the poor correlation between a clinical diagnosis and a histopathological diagnosis of ARDS, these studies are diagnosing a very heterogeneous disease syndrome [7,8,9,10]. In order to actually evaluate ARDS development, biomarkers should be compared to a histopathological image of DAD, although acquiring histology poses great challenges by itself. Fifth, as only biomarkers assessed in multivariate analyses were included in this study, new promising biomarkers evaluated in univariate analyses were excluded from this study. Lastly, non-significant biomarkers in multivariate analyses were more likely not to be reported, although some studies report non-significant results nonetheless.

Conclusion

In here, we present a list of biomarkers for ARDS mortality and ARDS development tested in multivariate analyses. In multiple studies that assessed Ang-2 and RAGE, high plasma levels were associated with an increased risk of ARDS development. We did not find a biomarker that independently predicted mortality in all studies that assessed the biomarker. Furthermore, biomarker data reporting and variables used in multivariate analyses differed greatly between studies. Taken together, we should look for a combination of biomarkers and clinical parameters in a structured approach in order to find more homogeneous ARDS phenotypes. This systematic review may guide the selection of ARDS biomarkers for ARDS phenotyping.