Signa Vitae. 2022; 18(5): 75-85. doi: 10.22514/sv.2021.233
Systematic Review

Precision medicine in Acute Respiratory Distress Syndrome

Nanon F.L. Heijnen1,*,, Dennis C.J.J. Bergmans1,2, Marcus J. Schultz3,4,5,6, Lieuwe D.J. Bos3,7

1Department of Intensive Care Medicine, Maastricht University Medical Center+, 6229 HX Maastricht, The Netherlands

2NUTRIM school of Nutrition and Translational Research, Maastricht University, 6229 HX Maastricht, The Netherlands

3Department of Intensive Care Medicine, Amsterdam University Medical Centers, location Academic Medical Center, University of Amsterdam, 1105 AZ Amsterdam, The Netherlands

4Laboratory of Experimental Intensive Care and Anesthesiology (L·E·I·C·A), Academic Medical Centers, location Academic Medical Center, University of Amsterdam, 1105 AZ Amsterdam, The Netherlands

5Mahidol-Oxford Tropical Medicine Research Unit (MORU), Mahidol University, 10330 Bangkok, Thailand

6Nuffield Department of Medicine, University of Oxford, OX3 Oxford, UK

7Department of Respiratory Medicine, Amsterdam University Medical Centers, location Academic Medical Center, University of Amsterdam, 1105 AZ Amsterdam, The Netherlands

*Corresponding Author(s):nanon.heijnen@mumc.nl (Nanon F.L. Heijnen)

History Submitted: 30 August 2021 | Accepted: 14 October 2021 | Published: 08 September 2022
Copyright:  ©2022  The Author(s). Published by MRE Press.
This is an open access article under the CC BY 4.0 license (https://creativecommons.org/licenses/by/4.0/).

Collapse table of contents

Abstract

Many patients with acute respiratory failure fulfill the diagnosis of Acute Respiratory Distress Syndrome (ARDS), forming a very heterogeneous population. Clinical trials have not yielded beneficial treatment effects in ARDS, possibly caused by this heterogeneity. Dividing patients with ARDS into subgroups, each with similar characteristics, may result in improved treatment strategies as part of a precision medicine approach. In this systematic review, we summarize the subphenotypes identified so far, the current state, and future directions for precision medicine in ARDS. Multiple data-driven subphenotypes have been identified based on a wide range of variables. These subphenotypes are associated with differences in clinical outcomes, which could be used for prognostic- and predictive enrichment of future interventional studies. True treatable traits have not been identified yet, deeper phenotyping will hopefully reveal these along with mechanistic differences.

Keywords:Precision medicinePhenotypesARDS
PDF(3.52 MB)|EndNote (RIS)|BibTeX|RefMan|RefWorks

Cite this article

Nanon F.L. Heijnen, Dennis C.J.J. Bergmans, Marcus J. Schultz, Lieuwe D.J. Bos. Precision medicine in Acute Respiratory Distress Syndrome. Signa Vitae. 2022; 18(5): 75-85. doi: 10.22514/sv.2021.233

1. Introduction

Around 10% of critically ill receiving invasive ventilation fulfill the Berlin definition for Acute Respiratory Distress Syndrome (ARDS), approximately 1.5 cases per 100,000 person-years in Europe alone. It is associated with a high mortality and considerable morbidity [1, 2]. The Berlin definition specifies ARDS as acute onset hypoxemia, bilateral opacities on chest radiography, not fully explained by effusion, collapse or nodules, which is not due to cardiac dysfunction or volume overload [1, 3]. It is important to realize that this syndrome comprises a heterogeneous patient population with a multiplicity of underlying pathophysiological processes resulting in alveolar epithelial and lung endothelial injury, increased lung vascular permeability, and protein-rich alveolar oedema [1]. Randomized clinical trials (RCTs) with drugs targeting specific pathways that have been implicated in the pathophysiology of ARDS, like oxidative stress and endothelial injury, failed to improve outcomes. Therefore, supportive therapy remains the cornerstone of care for ARDS [4].

One reason for this failure could be the heterogeneity of the syndrome, which makes a “one-size fits all” approach insufficient [4, 5]. Precision medicine is defined as “treatments targeted to the need of individual patients on the basis of genetic, biomarker, phenotypic, or psychosocial characteristics that distinguish a given patient from other patients with similar clinical presentations” [6]. Human epidermal growth factor receptor 2 (HER-2) targeted therapy in breast cancer and type 2 (eosinophilic) asthma endotype-specific treatment are examples of precision medicine approaches that have revolutionized the treatment of syndrome diagnoses [7, 8, 9]. These examples illustrate that dividing a group of patients with the same syndrome into subgroups, each with similar characteristics, can result in improved treatment strategies. This has led the researchers in critical care to speculate that a precision medicine approach would be appropriate for a heterogeneous syndrome like ARDS as well [10].

In this systematic review, we present the current understanding of precision medicine in ARDS. We provide an overview of the currently identified subgroups in ARDS based on data-driven approaches, evaluate the evidence for heterogeneity of treatment effect in patients with distinct subphenotypes and speculate on the future directions for precision medicine in ARDS.

2. Search strategy and selection criteria

Relevant articles were identified by a search on PubMed for articles up to May 17, 2021, with the terms: “ARDS”, “acute lung injury (ALI)”, “Critical Care”, “Intensive Care”, “Critical illness”, “Phenotype”, “Subphenotype”, “Subgroups”, “Endotypes”, and “Cluster”. Inclusion criteria were (1) original research in (2) adult critically ill patients with ARDS (3) identifying subphenotypes based on patient data (4) using clustering analysis algorithms and (5) providing prognostic or predictive value. Studies using pre-defined not data-driven subgroups or studies on cell phenotypes, animal or preclinical work were excluded. Only articles published in English were considered. See Fig. 1 for flowchart of article selection. After reading, ten original articles remained which fulfilled the selection criteria for this review.

Flowchart of article selection.

Fig. 1.Flowchart of article selection.

3. Definitions in precision medicine

Recently, definitions have been proposed to standardize the terminology used in the search for targetable (sub) phenotypes in the critically ill and associated broad defined syndromes, like ARDS.

In this review, we use the following: (1) Phenotype — “A set of clinical features in a group of patients who share a common syndrome or condition”, (2) Subphenotype — “A set of features in a group of patients who share a phenotype, such as shared risk factor, trait, diagnostic feature, expression marker, mortality risk, or outcome in response to treatment, that distinguishes the group from other groups of patients with the same phenotype”, (3) Endotype — “A distinct biological mechanism of disease, often associated with an anticipated response to treatment, that is shared by a subgroup of patients and might be indicated by shared mortality risk, clinical course, or treatment responsiveness”, and (4) Treatable trait — “A subgroup characteristic that can be successfully targeted by an intervention” [10, 11].

It should be noted that a subphenotype does not necessarily comprise an endotype. For a subphenotype to have an endotype there must be a mechanistic difference between the subphenotypes, which can be identified by certain markers. Similarly, it should be noted that an endotype does not mean there is a treatable trait. Only if a mechanistic difference can be successfully targeted by plausible treatment, a treatable trait has been identified. This is the ultimate goal of precision medicine.

In addition to identification of treatable traits, (sub) phenotyping can also be used as a tool for prognostic- and predictive enrichment strategies in RCTs. Enrichment is a core tenet of precision medicine. Using prognostic enrichment, patients with a higher risk at a worse outcome or disease-related endpoint are selected, thereby increasing the absolute effect difference between groups [12]. Predictive enrichment entails selecting patients more likely to respond to a given therapy, increasing both absolute and relative effect, possibly resulting in a smaller required study population [12]. These strategies stimulate development of new drug therapies and tailoring treatments to patients most likely to benefit from them. Combined, prognostic- and predictive enrichment allow for optimal progress towards precision medicine.

4. Identified ARDS subphenotypes

A variety of strategies have been applied in order to identify subphenotypes in ARDS, covering aspects of etiology, physiology and morphology, and biology. Fig. 2 provides an overview of the identified subphenotypes, including the used class-defining variables and predictive variables. Table 1 (Ref. [13, 14, 15, 16, 17, 18, 19, 20, 21, 22]) presents an overview of the subphenotypes with their prevalence and associated clinical outcomes. All described subphenotypes are based on clustering algorithms using a set of variables that did not include clinical outcomes.

Heatmap of all included class-defining variables and identified predictive variables in ARDS subphenotyping models. Only variables significant in at least one study are depicted. Class-defining variables were used for identifying subphenotypes. Predictive variables were used to classify the subphenotypes using fewer variables. The increased class-defining variables: (1) gender implicates a higher percentage of males and (2) ethnicity implicates a higher percentage of white people present in that subphenotype, (3) source of infection is pre-dominantly the thorax, and (4) ARDS risk factor is pre-dominantly pneumonia. Subphenotypes depicted on the left side of the graph were associated with worse clinical outcomes compared to the subphenotypes depicted on the right side of the graph. The numbers above the subphenotypes refer to the original study and correspond to the reference bibliography number. BMI, body mass index; CRP, C-reactive protein; WBC, white blood cell count; ANG-2, angiopoietin-2; ANG2/1, angiopoietin 2 and 1 ratio; ICAM-1, intercellular adhesion molecule-1; IFNy, interferon gamma; IL, interleukin; PAI-1, plasminogen activator inhibitor 1; RAGE, receptor for advanced glycation end products; TNFR1, tumor necrosis factor receptor 1; PF ratio, PaO2/FiO2 ratio.

Fig. 2.Heatmap of all included class-defining variables and identified predictive variables in ARDS subphenotyping models. Only variables significant in at least one study are depicted. Class-defining variables were used for identifying subphenotypes. Predictive variables were used to classify the subphenotypes using fewer variables. The increased class-defining variables: (1) gender implicates a higher percentage of males and (2) ethnicity implicates a higher percentage of white people present in that subphenotype, (3) source of infection is pre-dominantly the thorax, and (4) ARDS risk factor is pre-dominantly pneumonia. Subphenotypes depicted on the left side of the graph were associated with worse clinical outcomes compared to the subphenotypes depicted on the right side of the graph. The numbers above the subphenotypes refer to the original study and correspond to the reference bibliography number. BMI, body mass index; CRP, C-reactive protein; WBC, white blood cell count; ANG-2, angiopoietin-2; ANG2/1, angiopoietin 2 and 1 ratio; ICAM-1, intercellular adhesion molecule-1; IFNy, interferon gamma; IL, interleukin; PAI-1, plasminogen activator inhibitor 1; RAGE, receptor for advanced glycation end products; TNFR1, tumor necrosis factor receptor 1; PF ratio, PaO2/FiO2 ratio.

Table 1.Overview of prevalence and clinical outcome parameters per subphenotype identified in ARDS.
Prevalence (%)ICU-mortality (%)Ventilator-free days (n)28-day mortality (%)30-day mortality (%)60-day mortality (%)90-day mortality (%)
Ranjeva et al. (2021) [13]Phenotype 1193 (73%)--23.3%---
Phenotype 270 (27%)--40.0%---
Garcia et al. (2021) [14]Non-recruitable106 (45%)27 (23%)-----
Recruitable132 (55%)69 (52%)-----
Puybassat et al. (2000) [15]Non-focal45 (63%)24 (53%)-----
Focal26 (37%)11 (42%)-----
Bos et al. (2017) [22]Uninflamed218 (48%)34 (15.6%)21 (11–25)-47 (21.6%)--
Reactive236 (52%)86 (36.4%)9 (0–21)-89 (37.7%)--
Calfee et al. (2014) [21]Hypoinflammatory318 (67%)-17.8---23%
Hyperinflammatory155 (33%)-7.7---44%
Calfee et al. (2014) [21]Hypoinflammatory404 (74%)-18.4---19%
Hyperinflammatory145 (26%)-8.3---51%
Famous et al. (2017) [20]Hypoinflammatory727 (73%)-19--21%22%
Hyperinflammatory273 (27%)-3--44%45%
Sinha et al. (2018) [19]Hypoinflammatory468 (60%)-23 (6–26)--98 (20.9%)100 (21.4%)
Hyperinflammatory277 (40%)-15 (1–23)--101 (36.5%)104 (37.6%)
Calfee et al. (2018) [18]Hypoinflammatory353 (65%)-18 (0–23)59 (17%)--78 (22%)
Hyperinflammatory186 (35%)-2 (0–17)73 (39%)--87 (47%)
Sinha et al. (2021) [16]Hypoinflammatory457 (73%)-20 (11–25)----
Hyperinflammatory167 (27%)-5 (0–20)----
Sinha et al. (2021) [16]Hypoinflammatory211 (63%)-24 (0–28)----
Hyperinflammatory124 (37%)-0 (0–23)----
Kitsios et al. (2019) [17]Hypoinflammatory65 (62%)-nsns--ns
Hyperinflammatory39 (38%)-nsns--ns
All presented data is significant, except for ns (not significant). A dash represents an uninvestigated parameter. ARDS, Acute Respiratory Distress Syndrome; ICU, Intensive Care Unit.

4.1 Clinically-derived subphenotypes

Thus far, two subphenotypes have been identified (1 & 2) using readily available clinical data from a cohort of ARDS patients that had acute respiratory failure related to COVID-19 (Fig. 2). One subphenotype, named ‘Phenotype 2’, showed increased markers of coagulopathy, like D-dimer, prothrombin time (PT), activated partial thromboplastin time (aPTT), and fibrinogen, compared to the other subphenotype, named ‘phenotype 1’. White blood cell count and interleukin-6 (IL-6) were higher in ‘phenotype 2’, but plasma IL-6 concentration was much lower than has been observed in patients with ARDS due to other causes than COVID-19. There was no difference in parameters related to respiratory physiology, such as PaO2:FiO2, driving pressure, minute ventilation, and PaCO2. There was strong evidence for prognostic enrichment as patients with ‘phenotype 2’ had double the odds for 28-day mortality than patients with ‘phenotype 1’ (odds ratio (OR) 2.2; 95% confidence interval (CI) 1.2–3.9; Table 1) [13].

4.2 Physiology and morphology-derived subphenotypes

A ‘non-recruitable’ and ‘recruitable’ subphenotype have been identified in patients with ARDS not related to COVID-19 using latent class analysis on a broad set of parameters related to respiratory mechanics, gas-exchange and Computer Tomography(CT)-derived gas- and tissue volume (Fig. 2). The non-recruitable subphenotype was associated with a non-pulmonary cause of ARDS, fewer moderate-severe ARDS cases, a lower respiratory system elastance, a decreased alveolar dead space, less potentially recruitable lung volume, and less inhomogeneous lungs compared to the recruitable subphenotype. The recruitable subphenotype could be used for prognostic enrichment as it was associated with an increased risk of ICU-mortality (HR 2.9, 95% CI 1.7–2.7) (Table 1) [14].

Three radiological subphenotypes of ARDS were described: lobar attenuations (‘LA’); diffuse attenuations (‘DA’) and patchy attenuations (‘PA’). These were later redefined as ‘focal’ ARDS (LA subphenotype) and ‘non-focal’ ARDS (DA and PA subphenotype) [15, 23]. It is important to note that these subphenotypes were not the result of data-driven evaluation of the CT images, but rather the result of systematic evaluation of these scans by human operators. Non-focal lung morphology is characterized by diffuse and patchy lung aeration loss (increased inhomogeneity) and distinct lung mechanics including decreased total lung gas volume, a lower compliance of the respiratory system and a higher amount of recruitable lung compared with focal ARDS. This subphenotype has also been associated with an increased ICU-mortality in a more recent study [23].

4.3 Biology-derived subphenotypes

Biological data, such as plasma biomarkers, have also been used to identify subphenotypes in ARDS. Two subphenotypes, named “reactive” and “uninflamed”, were identified based on 20 plasma biomarkers of inflammation, coagulation, and endothelial activation (Fig. 2). The “reactive” subphenotype could be characterized by high plasma levels of inflammation, coagulation, and endothelial activation. Patients with the “reactive” subphenotype more frequently had a non-pulmonary cause for ARDS. Patients with the “reactive” subphenotype showed prognostic enrichment as it was associated with a higher ICU- and 30-day mortality and less ventilator-free days (Table 1) [22].

4.4 Subphenotypes based on combined variables

The majority of publications report on analyses based on combinations of clinical and biological variables. Two subphenotypes, named the “hypoinflammatory” and “hyperinflammatory”, have been consistently identified throughout multiple datasets using latent class analysis. The hyperinflammatory subphenotype has been characterized by higher plasma concentrations of IL-6, IL-8, soluble tumor necrosis factor receptor-1 (sTNFR1), and plasminogen activator inhibitor-1 (PAI-1), higher heart rate and total minute ventilation. This subphenotype also had a lower systolic blood pressure, bicarbonate, and protein C compared to the hypoinflammatory subphenotype. In other words, the hyperinflammatory subphenotype reflects a more severe inflammation, shock, and metabolic acidosis. There was prognostic enrichment for mortality and duration of mechanical ventilation [21]. In several subsequent secondary analyses of RCTs in ARDS, similar subphenotype profiles were identified, which validated the original finding [18, 19, 20]. Even the use of a less comprehensive dataset revealed two subphenotypes with comparable characteristics and clinical outcomes [18]. This is indicative of the robustness of these subphenotypes in a highly selected patient population of ARDS. Importantly, the “hypo-” and “hyperinflammatory” subphenotypes were also identified in prospective observational cohort studies using a more comprehensive set of variables. These studies confirmed the potential for prognostic enrichment of the hyperinflammatory subphenotype in an unselected population of consecutive ARDS patients [16, 17].

5. Evidence for heterogeneity of treatment effect

Each of the above described subphenotype approaches revealed a subgroup with an increased risk of mortality and selection of this subgroup could be used for prognostic enrichment. Differences in baseline risk of death could introduce non-random variation in treatment effect (heterogeneity of treatment effect, HTE), which might explain some indeterminate results of previous RCTs [24, 25, 26]. However, predictive enrichment of future intervention studies could provide more considerable HTE and this is most important for the design of future precision medicine studies.

Secondary analyses of three RCTs in ARDS patients showed potential HTE when using identified subphenotypes for risk stratification. Firstly, the multicenter Assessment of Low Tidal Volume and Elevated End-Expiratory Pressure to Obviate Lung Injury (ALVEOLI) trial compared the effect of mechanical ventilation with higher versus lower positive end-expiratory pressure (PEEP) within 36h of ARDS onset on mortality. The original analysis showed similar clinical outcomes regardless of the PEEP levels used [27]. A secondary analysis of this trial showed a subphenotype-dependent treatment effect. Patients with the hyperinflammatory subphenotype who received the high PEEP strategy had improved clinical outcomes (reduced mortality, more ventilator-free days and organ failure free-days) compared to the low PEEP strategy. Patients with the “hypoinflammatory” subphenotype showed strikingly opposite results with improved clinical outcomes using a low PEEP strategy compared to a high PEEP strategy [21]. Secondly, the Fluid and Catheter Treatment (FACTT) trial compared the effect of conservative versus liberal fluid management within 48h of ARDS onset on mortality. Conservative fluid management shortened the duration of mechanical ventilation, without showing a difference in 60-day mortality [28]. In a secondary analysis, hyperinflammatory patients had improved clinical outcomes (reduced 60- and 90-day mortality) when randomized to the liberal fluid strategy as compared to the conservative fluid strategy, while the hypoinflammatory patients showed the inverse association. However, no subphenotype-dependent significant difference in ventilator-free days was observed [20]. Thirdly, the multicenter Hydroxymethylglutaryl-CoA reductase inhibition with simvastatin in Acute lung injury to Reduce Pulmonary dysfunction (HARP-2) trial compared the effect of simvastatin versus placebo within 48h of ARDS onset on ventilator-free days. No differences in clinical outcomes were found (ventilator-free days, non-pulmonary organ failure, and 28-day mortality) [29]. However, differences were observed across patients stratified by treatment and subphenotype in a secondary analysis. Specifically, patients with the hyperinflammatory subphenotype had a higher 28-day survival using simvastatin compared to placebo [19]. In addition, potential HTE for simvastatin in ARDS was also observed in another secondary analysis using the APACHE II score as risk modifier [25]. Combined, these secondary analyses support the idea that indeterminate trial results can be the result of heterogeneity in trial populations. Subphenotyping could play a role in predictive enrichment trial strategies by reducing some of the heterogeneity within the larger ARDS population.

The first and currently only prospective evaluation of a precision medicine by subphenotypes in a RCT is the LIVE-trial: Lung Imaging for Ventilator Settings in ARDS [30]. They tested whether personalized mechanical ventilation strategies based on morphology subphenotypes (non-focal and focal) improved the overall survival of ARDS patients compared to standard care. Personalized mechanical ventilation strategies entailed tailored tidal volumes, PEEP levels, recruitment manoeuvres, and prone positioning per group (PP). The primary analysis of the LIVE-trial did not show survival benefit in favour of the precision medicine approach (HR: 1.01; 95% CI 0.61–1.66, p = 0.98). Further analysis showed that in 21% (85 out of 400) of all included patients the lung morphology was misclassified based on chest imaging. For the classification of non-focal and focal ARDS both CT-scan and chest radiography was allowed, but CT scans were performed only in 29% (56 patients) of the patients randomized to the precision medicine approach. Despite the high agreement about lung morphology classification between experts (k = 0.94), only moderate agreement was found between local investigators who allocated patients to the precision medicine approach (k = 0.52). The high likelihood of misclassification can be explained by the limited availability of CT-scans and misinterpretation of chest radiography. Interestingly, subgroup analyses revealed that: (1) correctly classified patients receiving personalized mechanical ventilation had lower 90-day mortality compared to the control group and (2) that expert classification revealed the same beneficial effect, but (3) misclassified patients had higher 90-day mortality when receiving personalized mechanical ventilation compared to the control group. So, due to the possible influence of misclassification, the contribution of using morphology subphenotypes for precision mechanical ventilation remains uncertain. These trial results emphasize (1) the requirement of subphenotypes to be robust and not subject to individual interpretation and (2) that misclassification can harm patients.

5.1 Subphenotype classification using parsimonious models and time-related changes

Most of the cluster and latent class analyses algorithms are not suitable for clinical classification of patients at the bedside due to the number of variables required as input. Therefore, predictive models containing fewer variables have been identified to classify patients with high accuracy (Fig. 2; Table 2, Ref. [14, 16, 17, 19, 20, 21, 31]). This also provides guidance in developing classifying tests suitable for clinical practice, like IL-6 and TNFR1 point-of-care tests (ClinicalTrials.gov Identifier: NCT04009330). Awaiting these point-of-care tests for specific plasma markers, researchers were recently able to classify patients in hypo- and hyperinflammatory subphenotypes using readily available clinical data including demographics variables (e.g., age, sex, ARDS risk factor), respiratory variables (e.g., PaO2/FiO2 ratio, PaCO2), vital signs (e.g., temperature, heart rate, respiratory rate), and laboratory variables (e.g., hematocrit, white cell count, platelets, sodium) with high accuracy (AUC: 0.95; 95% CI 0.94–0.96; Table 2) [31]. Although this classification was performed in highly selected study populations, this finding is very promising. Together with previous results, this provides multiple opportunities to enable classification in clinical practice.

Table 2.Predictive models classifying ARDS subphenotypes.
SubphenotypesPredictive model variablesAUC
Calfee et al. (2014) [21]Hypoinflammatory vs. Hyperinflammatory (ARMA)IL-6, sTNFR1, vasopressor use0.94
Calfee et al. (2014) [21]Hypoinflammatory vs. Hyperinflammatory (ALVEOLI)IL-6, sTNFR1, vasopressor use0.93
Famous et al. (2017) [20]Hypoinflammatory vs. Hyperinflammatory (FACTT)IL-8, sTNFR1, bicarbonate0.95
Sinha et al. (2018) [19]Hypoinflammatory vs. Hyperinflammatory (SAILS)IL-8, sTNFR1, bicarbonate0.95
Kitsios et al. (2019) [17]Hypoinflammatory vs. HyperinflammatoryNot defined0.93
Sinha et al. (2020) [31]Hypoinflammatory vs. HyperinflammatoryIL-8, protein C, bicarbonate, vasopressor use.0.96
(ARMA, ALVEOLI, SAILS, FACTT)
Sinha et al. (2020) [31]Hypoinflammatory vs. HyperinflammatoryClinical classifier model (demographic, respiratory, vital signs, laboratory data)0.95
(ARMA, ALVEOLI, SAILS, FACTT)
Sinha et al. (2021) [16]Hypoinflammatory vs. HyperinflammatoryIL-8, protein C, bicarbonate, vasopressor use0.94
(VALID, EARLI)
Bos et al. (2017) [22]Uninflamed vs. reactiveIl-6, IFN-gamma, ANG2/1, PAI-10.98
Garcia et al. (2021) [14]Non-recruitable vs. RecruitableDead space, respiratory system elastance, lung inhomogeneity, proportion of non-aerated lung tissue0.99
ARMA, ALVEOLI, SAILS, and FACTT are different randomized controlled trial cohorts in patients with ARDS. VALID and EARLI are prospective observational cohort studies in patients with ARDS. Abbreviations: IL, Interleukin; IFN-gamma, interferon gamma; ANG, angiopoietin; PAI-1, plasminogen activator inhibitor 1; sTNFR1, soluble tumor necrosis factor receptor 1.

All above-described subphenotypes have been identified using data obtained at ICU admission or at enrollment in clinical trials. The hypo- and hyperinflammatory subphenotype have shown to be largely stable over the first 3 days [32]. As it remains uncertain whether subphenotypes reflect different temporal stages in ARDS, it is important for the usability of subphenotype classification in clinical trials to evaluate the subphenotype stability over the evolution of ARDS. Baseline levels of innate immunity biomarkers (TNFR1, fractalkine, and ST-2) and procalcitonin were higher in the hyperinflammatory patients and showed similar trajectory overtime compared to hypoinflammatory patients. However, angiopoietin-2 (ANG-2) (endothelial injury) and receptor for advanced glycation end products (RAGE; marker of epithelial injury) attenuated over time [17]. Hypothetically, if this host-response trajectory also occurs in the reactive subphenotype (which is plausible as the reactive and hyperinflammatory subphenotype have similar characteristics), this could influence the classification, since ANG-2 is used in the prediction model for the reactive and uninflamed subphenotype [22].

In a secondary analysis of the Evaluating Health Outcomes and QOL After ALI Among Participants of the ALTA, OMEGA, EDEN, and SAILS ARDS Network Trials (SAILS-ALTOS) with a long term follow-up (up to 12 months), the physical, mental health, and cognitive outcomes were not different between patients who were classified as having the hypo- or hyperinflammatory subphenotype at study enrollment [33]. This might suggest that these subphenotypes reflect an acute phase of critical illness, resolve at some point and that other factors attribute to long term dysfunction.

5.2 Underlying processes captured by subphenotypes

The identified subphenotypes have not been linked directly to pathophysiological mechanisms leading to ARDS. It is noteworthy that in studies (which included biological data) the most important contributing class-defining variables are linked to the innate immune response (i.e., TNFR1, IL-6, IL-8) [17, 18, 20, 21]. It could be speculated that these subphenotypes reflect a more general underlying inflammatory reaction, as these markers are not ARDS-specific. This is supported by the identification and validation of resembling subphenotypes (hypo-/hyperinflammatory and unreactive/reactive) in both patients at risk for ARDS and mechanically ventilated patients without ARDS with similar characteristics, blood leukocyte gene expression profiles, and clinical outcomes [17, 34, 35, 36, 37].

COVID-19 has added another frequent cause for ARDS. Patients with COVID-19-associated ARDS did not show the extensive systemic inflammatory response seen in non-COVID-19 related ARDS. Patients with COVID-19 also much more frequently had single organ failure [13]. Remarkably, an exploratory analysis revealed a lower prevalence of the hyperinflammatory subphenotype in COVID-19-associated ARDS compared to the other ARDS cohorts, and surprisingly higher 28-day mortality rates for both subphenotypes in COVID-19-associated ARDS [38]. This highlights that clustering algorithms might not be sufficient when leaving fundamental differences, like etiology and risk factors, out of the scope even when a wide range of variables were used in the derivation phase.

Given the multiple subphenotypes described in this review, it is possible that we end up with a multi-layered system just like in asthma, where stratification is based on age of symptom onset, lung function (FEV1), allergic status and type of airway inflammation [9]. In ARDS, the following layers could be considered: etiology, lung morphology, abnormalities in gas exchange, and biology. Integrating these aspects into intervention studies and clinical care is one of the key challenges for future research.

6. Towards precision medicine in ARDS

While there are promising results with regards to identified subphenotypes, the goal of precision medicine -identifying treatable traits- has not yet been reached. To generate treatable traits, it is pivotal to increase our knowledge about the underlying pathophysiological mechanisms reflected by the identified subphenotypes, allowing us to link biological differences and determine whether a marker doesn’t only differentiate but also acts as a mediator. The current subphenotypes are mainly derived from clinical and blood biomarker data, omitting the pulmonary biology. The clinical pulmonary parameters included provide superficial insight into the pulmonary status but do not show a consistent difference between both subphenotypes [18, 1921]. The majority of the subphenotypes are identified in secondary analyses of datasets from RCTs with ARDS patients, which could be an explanation for the lack of consistent difference and advocates for studies with an unselected population. Furthermore, preliminary results with a small sample size showed no profound differences in a selected set of alveolar inflammatory mediators, emphasizing the importance of elucidating the pulmonary biology within the identified subphenotypes [39]. Despite the challenges associated with mapping the lung compartment, the link between the biological progression or resolution of the identified subphenotypes, and the phases in the pathogenesis of ARDS should be explored. Increasing our understanding of these subphenotypes in several areas is pivotal in order to understand the beneficial and harmful aspects of the host response within each subphenotype which could reveal the next steps towards precision medicine in ARDS.

As shown in Fig. 2, there is a broad range of class-defining variables that differ between subphenotypes resulting in unique sets of predictive variables. However, there is also considerable overlap and this allows us to integrate the available evidence into a bigger picture (Fig. 3). For example: the hyperinflammatory subphenotype is associated with worse clinical outcomes, more likely to have a non-pulmonary primary risk factor, and patients have increased levels of circulating RAGE compared to the hypoinflammatory subphenotype [18, 20, 21]. Subphenotypes based on lung morphology showed that the non-focal subphenotype was associated with worse clinical outcomes, alveolar fluid clearance (AFC) impairment, and increased RAGE levels [40]. Strikingly, RAGE itself seems to be inversely correlated to AFC rates [41]. One could therefore postulate that the non-focal subphenotype and hyperinflammatory subphenotype overlap to a large extent. If so, the RAGE pathway could be of interest as possible treatable trait. Since these subphenotypes were derived from completely disjointed variables, overlap might be a key indicator of possible pathways to target for researching treatable traits.

Venn diagram depicting an example of overlapping established associations between selected variables and the hyperinflammatory, reactive, recruitable, and non-focal subphenotype. RAGE, receptor for advanced glycation end products.

Fig. 3.Venn diagram depicting an example of overlapping established associations between selected variables and the hyperinflammatory, reactive, recruitable, and non-focal subphenotype. RAGE, receptor for advanced glycation end products.

7. Conclusions

The recognition of ARDS heterogeneity has created an opportunity to identify various subphenotypes, associated with different clinical outcomes. Key challenges will be (1) the characterization of the lung compartment and (2) integrating our subphenotypes related to clinical variables, lung morphology, gas-exchange abnormalities and biology in pre-clinical models and clinical trials. Deeper phenotyping, with parallel use of prognostic- and predictive enrichment strategies, will hopefully reveal mechanistic differences and treatable traits, marking the beginning of precision medicine in ARDS.

Author contributions

NFLH, DCJJB, MJS, LDJB contributed to the study concept and design. NFLH performed the data collection and wrote the first draft of the manuscript. DCJJB, MJS, LDJB commented on previous versions of the manuscript. All authors read and approved the final manuscript.

Ethics approval and consent to participate

Not applicable.

Acknowledgment

The authors would like to thank Professor Antonio Artigas for editing the special issue on New Insights in Acute Respiratory Distress Syndrome and for inviting us to contribute.

Funding

This research was funded by the Amsterdam UMC fellowship, the Innovative Medicine Initiative, the Dutch lung foundation, and Health Holland, grant number 10.2.17.181 PPS.

Conflict of interest

The authors declare no conflict of interest.

References

Thompson BT, Chambers RC, Liu KD. Acute Respiratory Distress Syndrome. New England Journal of Medicine. 2017; 377: 562–572.

[Google Scholar]

Confalonieri M, Salton F, Fabiano F. Acute respiratory distress syndrome. European Respiratory Review : an Official Journal of the European Respiratory Society. 2017; 26: 160116.

[Google Scholar]

Ranieri VM, Rubenfeld GD, Thompson BT, Ferguson ND, Caldwell E, Fan E, et al. Acute respiratory distress syndrome: the Berlin Definition. Journal of the American Medical Association. 2012; 307: 2526–2533.

[Google Scholar]

Matthay MA, McAuley DF, Ware LB. Clinical trials in acute respiratory distress syndrome: challenges and opportunities. The Lancet Respiratory Medicine. 2017; 5: 524–534.

[Google Scholar]

Bos LDJ, Artigas A, Constantin J, Hagens LA, Heijnen N, Laffey JG, et al. Precision medicine in acute respiratory distress syndrome: workshop report and recommendations for future research. European Respiratory Review. 2021; 30: 200317.

[Google Scholar]

Jameson JL, Longo DL. Precision medicine–personalized, problematic, and promising. The New England Journal of Medicine. 2015; 372: 2229–2234.

[Google Scholar]

Loibl S, Gianni L. Her2-positive breast cancer. Lancet. 2017; 389: 2415–2429.

[Google Scholar]

Wenzel SE, Schwartz LB, Langmack EL, Halliday JL, Trudeau JB, Gibbs RL, et al. Evidence that severe asthma can be divided pathologically into two inflammatory subtypes with distinct physiologic and clinical characteristics. American Journal of Respiratory and Critical Care Medicine. 1999; 160: 1001–1008.

[Google Scholar]

Wenzel SE. Asthma phenotypes: the evolution from clinical to molecular approaches. Nature Medicine. 2012; 18: 716–725.

[Google Scholar]

Prescott HC, Calfee CS, Thompson BT, Angus DC, Liu VX. Toward Smarter Lumping and Smarter Splitting: Rethinking Strategies for Sepsis and Acute Respiratory Distress Syndrome Clinical Trial Design. American Journal of Respiratory and Critical Care Medicine. 2016; 194: 147–155.

[Google Scholar]

Reddy K, Sinha P, O’Kane CM, Gordon AC, Calfee CS, McAuley DF. Subphenotypes in critical care: translation into clinical practice. The Lancet Respiratory Medicine. 2020; 8: 631–643.

[Google Scholar]

FDA. Enrichment Strategies for Clinical Trials to Support Determination of Effectiveness of Human Drugs and Biological Products. Guidance for Industry. 2019. Available at: https://www.fda.gov/BiologicsBloodVaccines/GuidanceComplianceRegulatoryInformation/Guidances/default.htm (Accessed: 20 May 2021).

[Google Scholar]

Ranjeva S, Pinciroli R, Hodell E, Mueller A, Hardin CC, Thompson BT, et al. Identifying clinical and biochemical phenotypes in acute respiratory distress syndrome secondary to coronavirus disease-2019. EClinicalMedicine. 2021; 34: 100829.

[Google Scholar]

Wendel Garcia PD, Caccioppola A, Coppola S, Pozzi T, Ciabattoni A, Cenci S, et al. Latent class analysis to predict intensive care outcomes in Acute Respiratory Distress Syndrome: a proposal of two pulmonary phenotypes. Critical Care. 2021; 25: 154.

[Google Scholar]

Puybasset L, Cluzel P, Gusman P, Grenier P, Preteux F, Rouby J-. Regional distribution of gas and tissue in acute respiratory distress syndrome. i. Consequences for lung morphology. Intensive Care Medicine. 2000; 26: 857–869.

[Google Scholar]

Sinha P, Delucchi KL, Chen Y, Zhuo H, Abbott J, Wang C, et al. Latent class analysis-derived subphenotypes are generalisable to observational cohorts of acute respiratory distress syndrome: a prospective study. Thorax. 2021. (in press)

[Google Scholar]

Kitsios GD, Yang L, Manatakis DV, Nouraie M, Evankovich J, Bain W, et al. Host-Response Subphenotypes Offer Prognostic Enrichment in Patients with or at Risk for Acute Respiratory Distress Syndrome. Critical Care Medicine. 2019; 47: 1724–1734.

[Google Scholar]

Calfee CS, Delucchi KL, Sinha P, Matthay MA, Hackett J, Shankar-Hari M, et al. Acute respiratory distress syndrome subphenotypes and differential response to simvastatin: secondary analysis of a randomised controlled trial. The Lancet Respiratory Medicine. 2018; 6: 691–698.

[Google Scholar]

Sinha P, Delucchi KL, Thompson BT, McAuley DF, Matthay MA, Calfee CS. Latent class analysis of ARDS subphenotypes: a secondary analysis of the statins for acutely injured lungs from sepsis (SAILS) study. Intensive Care Medicine. 2018; 44: 1859–1869.

[Google Scholar]

Famous KR, Delucchi K, Ware LB, Kangelaris KN, Liu KD, Thompson BT, et al. Acute Respiratory Distress Syndrome Subphenotypes Respond Differently to Randomized Fluid Management Strategy. American Journal of Respiratory and Critical Care Medicine. 2017; 195: 331–338.

[Google Scholar]

Calfee CS, Delucchi K, Parsons PE, Thompson BT, Ware LB, Matthay MA. Subphenotypes in acute respiratory distress syndrome: latent class analysis of data from two randomised controlled trials. The Lancet Respiratory Medicine. 2014; 2: 611–620.

[Google Scholar]

Bos LD, Schouten LR, van Vught LA, Wiewel MA, Ong DSY, Cremer O, et al. Identification and validation of distinct biological phenotypes in patients with acute respiratory distress syndrome by cluster analysis. Thorax. 2017; 72: 876–883.

[Google Scholar]

Rouby J-, Puybasset L, Cluzel P, Richecoeur J, Lu Q, Grenier P. Regional distribution of gas and tissue in acute respiratory distress syndrome. II. Physiological correlations and definition of an ARDS Severity Score. Intensive Care Medicine. 2000; 26: 1046–1056.

[Google Scholar]

Senn S. Mastering variation: variance components and personalised medicine. Statistics in Medicine. 2016; 35: 966–977.

[Google Scholar]

Santhakumaran S, Gordon A, Prevost AT, O’Kane C, McAuley DF, Shankar-Hari M. Heterogeneity of treatment effect by baseline risk of mortality in critically ill patients: re-analysis of three recent sepsis and ARDS randomised controlled trials. Critical Care. 2019; 23: 156.

[Google Scholar]

Kent DM, Rothwell PM, Ioannidis JPA, Altman DG, Hayward RA. Assessing and reporting heterogeneity in treatment effects in clinical trials: a proposal. Trials. 2010; 11: 85.

[Google Scholar]

Brower RG, Lanken PN, MacIntyre N, Matthay MA, Morris A, Ancukiewicz M, et al. Higher versus lower positive end-expiratory pressures in patients with the acute respiratory distress syndrome. The New England Journal of Medicine. 2004; 351: 327–336.

[Google Scholar]

Wiedemann HP, Wheeler AP, Bernard GR, Thompson BT, Hayden D, deBoisblanc B, et al. Comparison of two fluid-management strategies in acute lung injury. The New England Journal of Medicine. 2006; 354: 2564–2575.

[Google Scholar]

McAuley DF, Laffey JG, O’Kane CM, Perkins GD, Mullan B, Trinder TJ, et al. Simvastatin in the acute respiratory distress syndrome. The New England Journal of Medicine. 2014; 371: 1695–1703.

[Google Scholar]

Constantin J, Jabaudon M, Lefrant J, Jaber S, Quenot J, Langeron O, et al. Personalised mechanical ventilation tailored to lung morphology versus low positive end-expiratory pressure for patients with acute respiratory distress syndrome in France (the LIVE study): a multicentre, single-blind, randomised controlled trial. The Lancet Respiratory Medicine. 2019; 7: 870–880.

[Google Scholar]

Sinha P, Churpek MM, Calfee CS. Machine Learning Classifier Models can Identify Acute Respiratory Distress Syndrome Phenotypes Using Readily Available Clinical Data. American Journal of Respiratory and Critical Care Medicine. 2020; 202: 996–1004.

[Google Scholar]

Delucchi K, Famous KR, Ware LB, Parsons PE, Thompson BT, Calfee CS. Stability of ARDS subphenotypes over time in two randomised controlled trials. Thorax. 2018; 73: 439–445.

[Google Scholar]

Hashem MD, Hopkins RO, Colantuoni E, Dinglas VD, Sinha P, Aronson Friedman L, et al. Six-month and 12-month patient outcomes based on inflammatory subphenotypes in sepsis-associated ARDS: secondary analysis of SAILS-ALTOS trial. Thorax. 2021. (in press)

[Google Scholar]

Bos LDJ, Scicluna BP, Ong DSY, Cremer O, van der Poll T, Schultz MJ. Understanding Heterogeneity in Biologic Phenotypes of Acute Respiratory Distress Syndrome by Leukocyte Expression Profiles. American Journal of Respiratory and Critical Care Medicine. 2019; 200: 42–50.

[Google Scholar]

Heijnen NFL, Hagens LA, Smit MR, Cremer OL, Ong DSY, van der Poll T, et al. Biological Subphenotypes of Acute Respiratory Distress Syndrome Show Prognostic Enrichment in Mechanically Ventilated Patients without Acute Respiratory Distress Syndrome. American Journal of Respiratory and Critical Care Medicine. 2021; 203: 1503–1511.

[Google Scholar]

Drohan CM, Nouraie SM, Bain W, Shah FA, Evankovich J, Zhang Y, et al. Biomarker-Based Classification of Patients with Acute Respiratory Failure into Inflammatory Subphenotypes: a Single-Center Exploratory Study. Critical Care Explorations. 2021; 3: e0518.

[Google Scholar]

Sathe NA, Zelnick LR, Mikacenic C, Morrell ED, Bhatraju PK, McNeil JB, et al. Identification of persistent and resolving subphenotypes of acute hypoxemic respiratory failure in two independent cohorts. Critical Care. 2021; 25: 336.

[Google Scholar]

Sinha P, Calfee CS, Cherian S, Brealey D, Cutler S, King C, et al. Prevalence of phenotypes of acute respiratory distress syndrome in critically ill patients with COVID-19: a prospective observational study. The Lancet Respiratory Medicine. 2020; 8: 1209–1218.

[Google Scholar]

Heijnen NFL, Hagens LA, Smit MR, Schultz MJ, Poll T, Schnabel RM, et al. Biological subphenotypes of acute respiratory distress syndrome may not reflect differences in alveolar inflammation. Physiological Reports. 2021; 9: e14693.

[Google Scholar]

Jabaudon M, Blondonnet R, Lutz J, Roszyk L, Bouvier D, Guérin R, et al. Net alveolar fluid clearance is associated with lung morphology phenotypes in acute respiratory distress syndrome. Anaesthesia, Critical Care & Pain Medicine. 2016; 35: 81–86.

[Google Scholar]

Jabaudon M, Blondonnet R, Roszyk L, Bouvier D, Audard J, Clairefond G, et al. Soluble Receptor for Advanced Glycation End-Products Predicts Impaired Alveolar Fluid Clearance in Acute Respiratory Distress Syndrome. American Journal of Respiratory and Critical Care Medicine. 2015; 192: 191–199.

[Google Scholar]