Keywords
acute respiratory distress syndrome - acute lung injury - phenotype - precision medicine
Despite earlier descriptions of noncardiogenic pulmonary edema, the phenotype we now
refer to as the acute respiratory distress syndrome (ARDS) was first described in
a case series published in the Lancet just over 50 years ago.[1] In this study, Ashbaugh and colleagues describe 12 adults with acute onset hypoxic
respiratory failure and poor lung compliance. The patients demonstrated similar pathophysiology
despite distinct insults ranging from trauma to pneumonia to pancreatitis. The authors
also described improvements in oxygenation with the application of positive end expiratory
pressure (PEEP), an observation that revolutionized early ARDS care. Since this initial
phenotypic description, multiple clinical definitions have been proposed and evolved
as our understanding of ARDS has improved.[2]
[3]
[4]
[5] Currently, the 2012 Berlin Definition of ARDS defines the syndrome as the acute
onset of hypoxia and bilateral pulmonary opacities not fully explained by a cardiac
cause.[4] Acute onset is specified to be within 1 week of a precipitating illness and hypoxia
is determined by a PaO2 to FiO2 ratio less than or equal to 300 mm Hg while receiving a minimum of 5 cm H2O of PEEP.
The development of broad consensus definitions of ARDS, including the previous American–European
Consensus Definition,[3]
[4] has allowed for the completion of clinical trials demonstrating a therapeutic benefit
to several supportive care interventions, including lung protective mechanical ventilation
and prone positioning.[6]
[7] Unfortunately, however, the large majority of ARDS clinical trials have failed,
particularly trials of pharmaceutical interventions targeting ARDS biology.[8] The largest challenge in phenotyping ARDS is the lack of a simple diagnostic test,
resulting in the reliance on a consensus definition developed by experts.[4]
[9] The current consensus definition of ARDS remains challenging, as chest radiograph
interpretation has poor inter-rater reliability and clinicians routinely fail to recognize
ARDS when treating patients.[10]
[11]
[12]
[13]
[14] Additionally, given the current intentionally broad definitions of ARDS, the syndrome
has marked clinical, radiologic, pathologic, and biologic heterogeneity. In an autopsy
study of patients who died with ARDS, only 45% demonstrated the histopathologic correlate
of ARDS, diffuse alveolar damage (DAD), with the other 55% percent demonstrating a
variety of other pathologic findings.[15] It is this heterogeneity that is hypothesized to underlie many failures in translation
of promising preclinical therapeutics to patient populations. In this review, we aim
to outline the current approaches to understanding ARDS heterogeneity by identifying
subphenotypes of ARDS with distinct clinical, radiologic, or biologic characteristics.
We will discuss how unpacking heterogeneity has led to some early successes, and the
potential for future success with this approach with an eye toward personalized medicine
in ARDS care.
Evolution of Phenotypes, Subphenotypes, and Endotypes
Evolution of Phenotypes, Subphenotypes, and Endotypes
A phenotype is defined as the set of observable characteristics or traits of an organism
resulting from the interaction of genotype and the environment. The term has often
been used to describe syndromes and disorders in medicine believed to develop from
this interaction. Naturally, the identification of a phenotype in medicine begins
with a description of a group of individuals displaying similar characteristics, not
dissimilar to how Ashbaugh and colleagues initially described 12 cases of ARDS.[1] As a phenotype is investigated further, a natural evolution occurs whereby the phenotype
is better characterized. By better characterizing a phenotype, misclassification of
similar phenotypes is reduced and subtypes are discerned based on unique biology,
clinical characteristics, or response to treatment. In cardiology, the acute coronary
syndrome was initially described as sudden onset chest pain that often resulted in
death.[16] This initial syndrome is now commonly differentiated into unstable angina, non-ST
elevation myocardial infarction, and ST elevation myocardial infarction based on electrocardiogram
and troponin measurements. These three subtypes of the acute coronary syndrome reflect
similar but distinct biology and call for distinct interventions.
Inherent in the process of syndrome evolution is the understanding of endotypes, or
subtypes of a syndrome defined by distinct biology.[17] In pulmonary medicine, the understanding of asthma has evolved from one disorder
to multiple endotypes with different clinical presentations, prognoses, and responses
to therapy. Initially described as a chronic disorder of the lung characterized by
variable airflow obstruction and an underlying inflammatory process, asthma has long
been recognized as heterogeneous.[18] Early descriptions of asthma divided patients into two subgroups, “extrinsic” or
atopic asthma caused by inhalation of allergens and “intrinsic” or nonatopic asthma.[19] As the biologic understanding of asthma improved, two distinct inflammatory endotypes
of asthma based on the presence or absence of eosinophils were described.[20] More recently, unbiased clustering algorithms incorporating clinical and biological
variables have identified multiple clinical phenotypes of asthma.[21]
[22]
[23] Additionally, therapies that target specific inflammatory mediators based on biomarker-defined
endotypes are now in use in asthma, including the anti-immunoglobulin E therapeutic
omalizumab.[24] Similar phenotype evolutions can be described for chronic obstructive pulmonary
disease,[25] pulmonary vasculitis,[26] and interstitial lung disease.[27]
Outside of pulmonary medicine, the field of oncology has had tremendous success in
therapeutically targeting endotypes defined by specific cancerous mutations. For example,
the first step toward precision medicine in lung cancer was the development of chemotherapy
regimens based on cancer histology and stage.[28] With the advent of mutation-targeted therapies, lung cancer therapeutics have now
progressed to precision medicine based on tumor molecular profiles. Specifically,
epidermal growth factor receptor (EGFR) mutations predict response to EGFR tyrosine
kinase inhibitors gefitinib, erlotinib, and afatinib,[29]
[30] and anaplastic lymphoma kinase (ALK) mutations predict response to crizotinib.[31] In earlier trials of unselected non-small cell lung cancer patients, some of these
same therapeutics either failed to demonstrate a benefit or only demonstrated a marginal
benefit.[32]
[33] It was the recognition of biologically defined endotypes that resulted in the identification
of subgroups of patients most likely to therapeutically respond. Similar molecular-targeted
therapeutics have been developed in breast cancer, colorectal cancer, melanoma, and
others.[34]
[35]
[36]
While there is a significant amount to learn from the success of subtyping asthma
and cancer, the ARDS phenotype has unique challenges that make this task difficult.
First, the acute and critical nature of ARDS makes subtyping or endotyping ARDS extremely
time sensitive to provide a targeted therapeutic early in the syndrome. If biomarkers
are used to subtype ARDS, they must be easily and rapidly measured with high reproducibility
and validity. Second, pathology via biopsy is not commonly obtained in ARDS due to
the risks and therefore will not likely routinely be used for ARDS phenotype identification
as it is in cancer. Third, clinicians frequently fail to recognize ARDS, which is
a necessary first step in subtype identification. Despite these challenges, significant
recent progress has been made in phenotyping ARDS ([Table 1]), and researchers are consistently moving toward understanding biologically defined
endotypes of the syndrome. There is substantial overlap between the terms phenotype,
subphenotype, and endotype, and a subphenotype initially defined by a clinical or
radiographic factor may subsequently be found to identify unique biology.
Table 1
Phenotypes of acute respiratory distress syndrome
|
Phenotype
|
Description
|
Differences
|
Potential therapies
|
References
|
|
Hypoxia severity phenotypes
|
Berlin categories:
Mild: 200 < PaO2/FiO2 < 300
Mod: 100 < PaO2/FiO2 < 200
Severe: PaO2/FiO2 < 100
|
• Severity of hypoxia
• DAD more likely pathology in severe
|
• Prone positioning (PaO2/FiO2 < 150)
• Cisatracurium (PaO2/FiO2 < 150)
|
[4]
[7]
[41]
|
|
ARDS by precipitating risk factor
|
Precipitating factors including:
sepsis, trauma, pneumonia, aspiration, transfusion, pancreatitis
|
• Differences in ARDS risk, severity, and mortality
|
|
[53]
[56]
[57]
[58]
[59]
[60]
|
|
Direct versus indirect lung injury
|
Direct: pneumonia, pulmonary contusion, aspiration
Indirect: nonpulmonary sepsis, nonthoracic trauma, transfusions
|
• Epithelial vs. endothelial injury
• Differences in mortality
|
• Epithelial vs. endothelial targeted therapies
• Indirect more likely to respond to PEEP
|
[61]
[64]
[65]
[66]
[67]
[68]
[69]
[70]
[71]
[72]
[73]
[74]
|
|
Timing of onset phenotypes
|
Early onset developing <48 h from admission versus late onset >48 h from admission
|
• Different clinical characteristics
• Elevated RAGE and Ang-2 in early onset
|
|
[75]
[76]
[77]
[78]
|
|
Radiographic phenotypes
|
Nonfocal/diffuse vs. focal/lobar on chest imaging
|
• Differences in morality, lung compliance, indirect lung injury, and plasma RAGE
level
|
• Diffuse more likely to respond to PEEP
|
[81]
[82]
[83]
[84]
[85]
[86]
|
|
Genetic defined endotypes
|
Endotypes of ARDS defined by genetic variability that alters ARDS risk, outcome, or
response to treatment
|
• Distinct ARDS risk, outcome, or response to treatment
|
• Therapies targeting biology implicated by genetic variants
|
[88]
[89]
|
|
Biomarker defined endotypes
|
Endotypes of ARDS defined by biomarker measurements
|
• Distinct ARDS risk, outcome, or response to treatment
|
• Therapies targeting biology implicated by biomarker elevation
|
[109]
[110]
[111]
[112]
[113]
[114]
[115]
[116]
[117]
[118]
[119]
[124]
[130]
[131]
|
|
Hyperinflammatory versus uninflamed
|
Endotypes of ARDS determined from unbiased latent class analysis and cluster analysis
|
• Hyperinflammatory characterized by elevated plasma inflammatory biomarkers, and
higher mortality
|
• Phenotypes responded differently to PEEP and fluid strategy
• Survival benefit observed in response to simvastatin in hyperinflammatory phenotype
|
[142]
[143]
[144]
[147]
[148]
[151]
[152]
|
Abbreviations: Ang-2, angiopoietin-2; ARDS, acute respiratory distress syndrome; DAD,
diffuse alveolar damage; PEEP, positive end expiratory pressure; RAGE, receptor for
advanced glycation end-products.
Severity Phenotypes of ARDS
Severity Phenotypes of ARDS
The Berlin Definition of ARDS proposed three severity phenotypes; mild, moderate,
and severe, defined by the ratio of PaO2 to FiO2.[4] When applied to multicenter clinical datasets, the higher severity stages were associated
with a higher mortality, ranging from 27% for mild, 32% for moderate, and 45% for
severe ARDS. While targeting the PaO2/FiO2 ratio as a surrogate outcome for mortality has largely failed in clinical trials,
the Berlin severity categories have the potential to serve for prognostic enrichment.[37]
[38] Prognostic enrichment is a strategy whereby a clinical trial enrolls patients more
likely to experience an outcome of interest, reducing the sample size needed to detect
a treatment effect of the intervention.[39] In addition to predicting mortality, severity categories based on the PaO2/FiO2 ratio may also help identify distinct subtypes of disease. In the previously discussed
autopsy study of 356 patients with ARDS, DAD was only present in 45%; however, those
patients with moderate and severe ARDS were significantly more likely to have DAD
on pathology.[15] This finding suggests that the moderate and severe subtypes identify a more homogeneous
population with ARDS. Therefore, the moderate–severe Berlin severity categories may
also serve for predictive enrichment. Predictive enrichment is a strategy by which
patients more likely to respond to a specific therapy are enrolled in a clinical trial.[40] If an intervention in ARDS is aimed at treating the biology that results in DAD,
it may be more likely to have a demonstrable benefit in the more homogeneous moderate
to severe subgroups of ARDS.
An approach targeting a severe phenotype of ARDS based on the PaO2/FiO2 ratio has been successful in two randomized controlled trials, the ARDS et Curarisation
Systematique (ACURASYS) study and the Proning Severe ARDS Patients (PROSEVA) study.[7]
[41] In ACURASYS, 340 ARDS patients with a PaO2/FiO2 ≤ 150 mm Hg were randomized to 48 hours of cisatracurium or placebo while heavily
sedated early in ARDS.[41] Given the potential adverse effects of systemic paralysis and heavy sedation,[42]
[43] the study only enrolled patients with severe gas-exchange impairments based on a
PaO2/FiO2 < 150 mm Hg.[44] The early use of systemic cisatracurium significantly improved adjusted 90-day survival
(hazard ratio [HR]: 0.68; 95% confidence interval [CI]: 0.48–0.98) and increased time
off the ventilator without increasing muscle weakness. In PROSEVA, 466 ARDS patients
also with a PaO2/FiO2 less than 150 mm Hg were randomized to undergo prone positioning for at least 16
hours a day or to be left supine. The PROSEVA investigators selected the 150 mm Hg
cutoff based on prior meta-analyses of negative clinical trials that suggested a benefit
to prone positioning only among those subjects severely hypoxemic at the time of enrollment.[45]
[46] Early prone positioning resulted in a reduction in 90-day mortality from 41.0 to
23.6% (HR: 0.44; 95% CI: 0.29–0.67). These two trials demonstrate the success of using
a PaO2/FiO2 ratio to identify severe ARDS phenotypes for clinical trial enrollment, but this
potential expands beyond the PaO2/FiO2 ratio. Several severity-scoring systems are used in clinical research, predict mortality,
and may be useful to identify severity phenotypes of ARDS.[47]
[48]
[49] Additionally, while they are not included in the Berlin Definition, severity phenotypes
based on respiratory parameters (e.g., lung compliance, dead space, or oxygenation
ratio) also have the potential to provide useful information for future prognostic
or predictive enrichment.[50]
[51]
[52]
Impact of Clinical Variables on ARDS Phenotypes
Impact of Clinical Variables on ARDS Phenotypes
In the original description of ARDS, Ashbaugh and colleagues described a series of
ARDS cases developing after the onset of heterogeneous precipitating illnesses.[1] The authors highlighted the similarity in the respiratory syndrome these patients
developed despite distinct clinical characteristics; however, subsequent authors have
recognized significant heterogeneity in ARDS based on several clinical variables.
The first source of clinical heterogeneity described was the underlying precipitating
factor for ARDS. The incidence of ARDS varies significantly across patients with different
precipitating factors, ranging from 30 to 40% of sepsis patients to 10 to 25% of severe
trauma patients.[53]
[54]
[55]
[56]
[57] In addition, the severity, biology, and outcomes vary as well. Sepsis-related ARDS
is associated with a higher disease severity and higher mortality than nonsepsis-related
ARDS, partially explained by the severity of illness as well as comorbidities.[58]
[59] Conversely, trauma-associated ARDS has been reported to be associated with a lower
mortality than nontrauma-associated ARDS even after adjusting for baseline clinical
factors and severity of illness.[60] In one study, patients with trauma-associated ARDS also had significantly lower
plasma levels of markers of epithelial and endothelial injury but not markers of acute
inflammation or disordered coagulation, suggesting biologic differences by ARDS precipitating
risk factor.[60]
Another traditional way to subphenotype ARDS is by grouping patients into direct (or
pulmonary) versus indirect (or extrapulmonary) ARDS.[3]
[61] The majority of direct lung injury results from pneumonia, aspiration, and pulmonary
contusion, while the majority of indirect lung injury results from nonpulmonary sepsis,
nonthoracic trauma, and transfusions.[62] This categorization is imperfect because many patients have mixed etiologies for
lung injury and an overlap of direct and indirect insults[63]; however, several physiologic, pathologic, and biologic differences between direct
and indirect ARDS have been reported. Pathologically, direct ARDS has been associated
with significantly more alveolar collapse, fibrin deposition, and alveolar wall edema
than indirect ARDS in an autopsy study.[64] Radiographically, direct ARDS has been characterized as a combination of ground
glass opacities and consolidation with asymmetry of the consolidated areas.[65]
[66]
[67] Conversely, indirect ARDS has been characterized as predominantly ground glass opacities
without significant consolidation and a more central distribution. Physiologically,
patients with direct ARDS have been reported to have a higher lung elastance at baseline
relative to indirect ARDS, but may be less likely to respond to PEEP.[67]
[68] There is also evidence that direct and indirect ARDS may differ biologically as
well. Direct ARDS is associated with higher plasma concentrations of biomarkers of
epithelial injury, including surfactant protein D (SP-D) and the receptor for advanced
glycation end-products (RAGE).[69]
[70] In contrast, indirect ARDS is associated with higher plasma concentrations of biomarkers
of endothelial injury, including angiopoietin-2 (Ang-2) and von Willebrand factor
(VWF).[69]
[71]
[72]
[73] Additionally, several genetic polymorphisms have been reported to be associated
with only direct or indirect ARDS but not the other subphenotype.[74] Despite the extensive research that has gone into characterizing direct and indirect
ARDS, the two phenotypes overlap substantially, and their clinical utility is thus
far limited.
The Berlin Definition requires ARDS to develop within 1 week of the onset of the underlying
precipitating illness[4]; however, several researchers have identified significant heterogeneity based on
the timing of onset of ARDS within that week.[75]
[76]
[77]
[78] Croce and colleagues first described early- and late-onset posttraumatic ARDS as
distinct.[76] These authors defined early-onset ARDS as occurring within 48 hours after presentation
and late-onset occurring beyond 48 hours after admission. Early-onset ARDS was associated
with more profound hemorrhagic shock while pneumonia was significantly associated
with late-onset ARDS. More recently, latent class analysis (LCA) was used to identify
subphenotypes of posttraumatic ARDS based on the certainty of the ARDS diagnosis over
time.[77] Similar to the earlier study, the 48-hour time point was identified to separate
early-onset and late-onset phenotypes. The early-onset phenotype was again associated
with hemorrhagic shock and the severity of thoracic trauma compared with the late-onset
phenotype. Additionally, early-onset ARDS was also associated with higher plasma levels
of RAGE and Ang-2, suggesting a biological distinction characterized by an alveolar
capillary barrier injury pattern in this phenotype. Zhang and colleagues studied patients
with moderate to severe ARDS precipitated by a variety of ARDS risk factors, again
dividing patients between early- and late-onset ARDS at 48 hours after admission.[78] In this study, late-onset ARDS was associated with a higher mortality than early-onset
ARDS despite lower severity of illness scores at admission.
One of the challenges in subphenotyping ARDS based on timing of ARDS onset is identifying
the exact time of onset of the precipitating illness. In all of these studies, the
onset time was the time of emergency department presentation or intensive care unit
admission. In trauma, this time point is likely within several hours of traumatic
injury, but in other precipitating syndromes such as sepsis, the exact beginning of
the sepsis syndrome is not always clear. Additionally, variability may exist if the
ARDS onset time is based on the time when all ARDS criteria are met (oxygenation and
radiograph criteria) versus the time point when the first criterion is met.[79]
[80] Despite a wealth of studies describing subphenotypes of ARDS based on clinical factors,
it is important to remember that no therapies specific to a clinical subphenotype
have been proven effective. Conversely, data from the landmark ARDS Network ARMA trial
suggests lung protective mechanical ventilation is effective in multiple clinical
ARDS subtypes.[6]
Radiographic Phenotypes of ARDS
Radiographic Phenotypes of ARDS
Radiographically, ARDS has largely been described as two phenotypes, nonfocal/diffuse
ARDS and focal/lobar ARDS, based on morphologic characteristics on computed tomography
(CT).[81] While there is significant correlation between these two categories and the clinical
categories of direct and indirect lung injury, they do not fully overlap. ARDS categorized
by diffuse rather than focal infiltrates has been associated with a higher mortality,
lower lung compliance, more commonly indirect lung injury, and a lower inflection
point on the pressure–volume curve of the lung.[82]
[83] Additionally, authors have reported distinct responses to the application of PEEP
and recruitment maneuvers based on radiographic phenotype.[84]
[85] Patients with nonfocal or diffuse ARDS appear to have significant alveolar recruitment
without significant overdistension with the application of increasing PEEP, while
patients with focal ARDS tend to develop significant overdistension without recruitment.
Supporting a biologic basis to radiographic phenotypes, another study reported a strong
association of plasma concentrations of the epithelial biomarker RAGE and nonfocal
CT based lung-imaging patterns in patients with ARDS.[83] A clinical trial is currently pending, which randomized patients to receive a traditional
low tidal volume ventilation protocol versus a mechanical ventilation protocol tailored
to radiographic lung morphology.[86]
Genetic and Biomarker-Defined Endotypes of ARDS
Genetic and Biomarker-Defined Endotypes of ARDS
Approaches to phenotype ARDS based on clinical, radiologic, or severity variables
are intuitive, particularly to clinicians who regularly treat ARDS; however, these
approaches have only been modestly successful and fail to fully differentiate biologic
differences that have the potential to be pharmacologic targets. Multiple biologic
pathways have been implicated in ARDS, including endothelial and epithelial dysfunction,
innate immune activation with immune cell recruitment, intravascular coagulation,
and intraalveolar fibrosis.[87] However, the degree to which each pathogenic pathway is dominant in each individual
patient is likely variable and based on a patient's clinical characteristics and genetics,
as well as time from insult.[88] One approach to identify endotypes of ARDS is to classify patients based on genetic
variability or based on the concentrations of biomarkers measured during critical
illness that represent targetable biologic processes.
The heritability of ARDS is difficult to measure as the requirement for a major environmental
insult precludes family pedigree studies; however, there is significant evidence that
ARDS risk is altered by genetic variability.[74]
[88]
[89] Evolutionary pressures such as hemorrhagic shock, host–pathogen interactions, dehydration,
and starvation select on mechanisms important in critical illness, including ARDS.[90] Therefore, it is reasonable to assume that evolutionary pressures which shape modern
human genetic diversity impact ARDS risk and may be useful in identifying ARDS endotypes
more likely to respond to novel therapeutics. Many discrete genetic polymorphisms
in genes related to innate immunity, alveolar–capillary barrier function, surfactant
function, oxidative stress, and other pathogenic pathways have been implicated in
ARDS.[88]
[89] An extensive review of these genetic associations is beyond the scope of this article,
but a few examples illustrate the potential impact of genetically defined endotypes
on ARDS therapeutics.
One of the earliest reported genetic associations with ARDS is with a common deletion
located in the human ACE gene that results in a higher plasma and tissue angiotensin converting enzyme (ACE)
activity and a higher risk of ARDS.[91]
[92] The renin–angiotensin system has long been implicated in ARDS pathogenesis.[93] ACE is responsible for the conversion of angiotensin I to angiotensin II in the
pulmonary vasculature, which results in vasoconstriction among other effects.[94] Therapeutics that negatively regulate the ACE axis, including ACE2, are currently
being developed for ARDS.[95] A population that has a baseline higher ACE activity, such as one predicted by genetic
variability in the ACE gene, might be an ideal population in which to test such a therapeutic.
Another relationship that has the potential to identify a genetically defined endotype
is the association between ABO histo-blood type A and increased risk of ARDS.[96]
[97] The ABO gene encodes an enzyme responsible for placing terminal carbohydrate modifications
on red blood cells as well as endothelial cells, epithelial cells, and platelets.[98] Variation in the ABO gene that determines ABO blood type is thought to have evolved via historic human–pathogen
interactions, including with malaria.[99] The ABO gene is now recognized as among the most pleiotropic (one gene affecting multiple
phenotypes) genes in the genome, and has been associated with multiple diseases and
quantitative traits.[100]
[101]
[102]
[103]
[104] With regard to ARDS, ABO variation that determines blood type is strongly associated with plasma levels of
multiple endothelial-derived glycoproteins important in ARDS, including VWF, intercellular
adhesion molecule 1 (ICAM-1), and Ang-2.[104]
[105]
[106]
[107]
[108] It is possible that a therapeutic targeting these proteins or the endothelium in
general could have a distinct effect in different blood types, a genetic trait that
is easily accessible clinically.
Similar to genetics, another approach to endotyping ARDS is to measure biomarkers
representative of the activation of particular biologic pathways. These biomarkers
may then be used to enroll patients in a clinical trial targeting their biology. Multiple
plasma biomarkers have been reported to predict ARDS or ARDS outcome, including markers
of inflammation (e.g., interleukin (IL)-6, IL-8),[109]
[110] endothelial activation and/or injury (e.g., Ang-2, ICAM-1, VWF),[111]
[112]
[113] epithelial injury (e.g., RAGE, SP-D),[114]
[115] and impaired coagulation (e.g., protein C).[116]
[117] One of the most promising plasma biomarkers for endotype identification in ARDS
is Ang-2. Plasma Ang-2 is an established biomarker and mediator of endothelial activation
and permeability and is strongly associated with ARDS risk and outcome.[72]
[111] Genetic variation in the Ang-2 gene (ANGPT2) is associated with ARDS risk,[118]
[119] and exogenous administration of Ang-2 potentiates lung injury in rodent models.[120] These data strongly suggest that Ang-2 causally contributes to ARDS pathogenesis;
however, it is often difficult to distinguish biomarkers that causally contribute
to disease and should be therapeutically targeted from biomarkers that only correlate
with disease. For example, in cardiology the plasma low-density lipoprotein concentration
causally contributes to heart disease and is therapeutically targeted while C-reactive
protein predicts disease risk but does not causally contribute to heart disease.[121]
[122] One method to distinguish potentially causal from correlative biomarkers with observational
data is mendelian randomization (MR) analysis.[123] In MR, an individual's genetic background is used as an instrumental variable to
infer causality of a measured biomarker. In the case of Ang-2, MR analysis strongly
suggests that plasma Ang-2 causally contributes to ARDS and should be prioritized
for drug development.[124] It is possible that an endothelial endotype of ARDS defined by elevated plasma Ang-2
may be the group most likely to respond to Ang-2-targeted therapies.
Another plasma biomarker that may identify an endotype of ARDS is the IL-1 receptor
antagonist (IL1RA). IL1RA is an inhibitory anti-inflammatory cytokine that competes
with proinflammatory cytokines IL-1α and IL-1β to bind the IL-1 receptor without triggering
receptor signaling.[125] In a large-scale genetic study, a coding genetic variant in the IL1RA gene (IL1RN) was associated with decreased risk of ARDS in multiple critically ill populations,
as well as increased plasma IL1RA in the setting of sepsis and trauma.[126] These findings suggest that the coding variant in IL1RN confers protection from ARDS by increasing plasma IL1RA. Therefore, IL1RA has high
potential for therapeutic benefit in ARDS. Three randomized placebo-controlled trials
tested human recombinant IL1RA in sepsis, including patients with ARDS, but failed
to demonstrate a benefit.[127]
[128]
[129] In a retrospective secondary analysis of one of these trials, Meyer et al demonstrated
significant heterogeneity of treatment effect based on plasma concentrations of IL1RA
measured at study enrollment.[130] Patients with a higher endogenous IL1RA had a statistically significant survival
benefit to recombinant IL1RA therapy, while those with lower IL1RA did not. The test
for interaction was also statistically significant, suggesting a true difference in
the treatment effect between groups stratified by measured endogenous IL1RA. While
it may be counterintuitive that patients with high endogenous IL1RA also had a benefit
to the administration of recombinant IL1RA, endogenous IL1RA may be serving as a biomarker
of activation of the entire IL-1 axis. In the study conducted by Meyer et al, IL1RA
was more easily measured than the proinflammatory marker of the IL-1 axis, IL-1β.
Future precision trials of recombinant IL1RA for an IL-1 axis endotype of ARDS warrant
further consideration. Other promising endotype-defining biomarkers that could be
targeted therapeutically include soluble RAGE (sRAGE), thrombomodulin, and tumor necrosis
factor receptor-1 (TNFR-1), among others.[114]
[131]
[132]
[133]
Unbiased Approaches to Endotype ARDS
Unbiased Approaches to Endotype ARDS
As our understanding of the complex biology of ARDS has improved, mathematical and
statistical methods to understand heterogeneity have also advanced. Cluster analysis
techniques, such as hierarchical or k-means clustering, have been used to identify
groups within a larger population with similar characteristics. As discussed previously,
these methods have been used successfully in asthma to identify subphenotypes with
distinguishable clinical characteristics and response to asthma therapies.[21]
[22] In critical care, Wong and colleagues used cluster-based techniques to identify
two subphenotypes of pediatric sepsis with distinct outcomes using whole blood gene
expression data.[134]
[135] The two subphenotypes were subsequently found to have distinct responses to systemic
corticosteroids.[136] Similarly, in adult sepsis populations, researchers have identified two sepsis response
signatures in peripheral blood leukocyte gene expression data using hierarchical clustering.[137]
[138] The first sepsis response signature was characterized by an immunosuppressed phenotype
and was associated with a higher mortality than the second sepsis response signature.
Another group of investigators used similar methods to identify four phenotypes of
sepsis with significant overlap with the two sepsis response signatures.[139] Cluster analyses have advantages including an unbiased mathematical approach to
the data based on characteristics introduced to the model rather than preexisting
assumptions. This approach allows for the discovery of biologically significant endotypes
that may not be readily apparent. However, cluster-based approaches are limited in
that they only identify heterogeneity based on the variables considered and can suffer
from problems with overfit data and challenges with replication.[140]
Another method to endotype syndromes, used with some success in ARDS and similar to
cluster analysis, is LCA. While cluster analysis is a mathematical method to identify
clusters, LCA is a statistical modeling method that identifies unobserved (latent)
groups within a heterogeneous population.[141] In ARDS, LCA has been used to identify two distinct subphenotypes characterized
by distinct biology, outcomes, and response to therapy.[142]
[143]
[144] Initially using data from patients enrolled in the NHLBI ARDSNet ARMA and ALVEOLI
trials,[6]
[145] LCA was applied to clinical and plasma biomarker data and identified a hyperinflammatory
phenotype of ARDS.[142] This phenotype was characterized by higher plasma concentrations of inflammatory
biomarkers, worse shock, and higher mortality. Additionally, the effect of treatment
with PEEP differed based on phenotype whereby the high PEEP strategy appeared to be
more effective in the hyperinflamed phenotype. The identified subphenotypes were subsequently
reported to be largely stable over the first 3 days of trial enrollment,[143] and were validated in a post-hoc analysis of the FACTT trial.[144]
[146] Additionally, the ARDS subphenotypes responded differently to the fluid liberal
versus conservative strategy tested in the FACTT trial. Specifically, the hyperinflamed
subtype had a higher mortality with a conservative fluid strategy while the other
subtype had a lower mortality with a conservative fluid strategy. These findings again
suggest heterogeneity in the response to therapies that have previously been applied
in a “one size fits all” paradigm. In this study, the authors also developed a simple
model based on three biomarkers (IL-8, TNFR1, and bicarbonate) to categorize patients
accurately in the two subphenotypes. If validated prospectively, this simpler model
may remove the need to re-perform the complicated LCA analysis to phenotype ARDS patients
making endotype identification potentially possible clinically.
LCA has also been applied to data from pharmacologic trials, specifically those studying
the efficacy of statins for ARDS.[147]
[148] In secondary analyses of both the HARP-2 trial investigating simvastatin and the
SAILS trial investigating rosuvastatin,[149]
[150] LCA identified two subphenotypes of ARDS with similar biological characteristics,
outcomes, and distributions as the prior trials.[147]
[148] While the HARP-2 and the SAILS trials were negative, the hyperinflammatory subphenotype
of ARDS had improved survival with simvastatin compared with placebo. This survival
benefit was not seen with rosuvastatin in SAILS, for reasons that may be related to
differences in the two statins' biology, dosing, clinical trial design, or patient
populations. Despite identifying benefits to PEEP, fluid strategy, and simvastatin
specific to LCA-defined endotypes, these analyses are post hoc and should not be implemented
without prospective trials.
Other investigators have also applied clustering techniques to an observational cohort
study of patients with ARDS.[151] In one study, cluster analysis was performed using 20 plasma biomarkers of inflammation,
coagulation, and endothelial activation, without any clinical data. Again, two subphenotypes
of ARDS were identified and described as uninflamed and reactive. The reactive phenotype
was associated with a higher mortality and could be accurately identified using five
biomarkers (IL-6, interferon gamma, angiopoietin 1/2, and plasminogen activator inhibitor-1).
These authors subsequently demonstrated that a third of genes are differentially expressed
in whole blood between the two subphenotypes, providing further evidence of significant
biological heterogeneity.[152] These identified subphenotypes share similar characteristics to those identified
in the LCA analyses conducted by Calfee and colleagues, though the precise degree
of overlap between the two approaches remains unknown. [Table 2] includes a comparison of the subphenotypes of ARDS identified via LCA or cluster
analysis.
Table 2
Unbiased approaches to identify ARDS endotypes applied to populations from five clinical
trials and one cohort study
|
Cohort
|
Phenotype
|
% of study population
|
90-day mortality
|
Heterogeneity of treatment response
|
Reference
|
|
ARMA
|
• Hypoinflammatory
• Hyperinflammatory
|
67%
33%
|
23%
44%
|
|
[142]
|
|
ALVEOLI
|
• Hypoinflammatory
• Hyperinflammatory
|
74%
26%
|
19%
51%
|
Differences in response to a high PEEP strategy
|
[142]
|
|
FACTT
|
• Hypoinflammatory
• Hyperinflammatory
|
73%
27%
|
22%
45%
|
Differences in response to a liberal vs. conservative fluid strategy
|
[144]
|
|
MARS
|
• Uninflamed
• Reactive
|
48%
52%
|
22%[a]
38%[a]
|
|
[151]
|
|
HARP-2
|
• Hypoinflammatory
• Hyperinflammatory
|
65%
35%
|
17%
39%
|
Differences in response to simvastatin therapy
|
[148]
|
|
SAILS
|
• Hypoinflammatory
• Hyperinflammatory
|
60%
40%
|
21%
38%
|
No differences based on rosuvastatin
|
[147]
|
Abbreviations: ALVEOLI, Assessment of Low Tidal Volume and Elevated End-expiratory
volume to Obviate Lung Injury Study; ARDS, acute respiratory distress syndrome; ARMA,
Prospective, Randomized, Multicenter Trial of 12 vs. 6 mL/kg Tidal Volume Positive
Pressure Ventilation for Treatment of Acute Lung Injury and Acute Respiratory Distress
Syndrome; FACTT, Fluid and Catheter Treatment Trial; HARP-2, HMG-CoA Reductase Inhibition
in Acute Lung Injury to Reduce Pulmonary Dysfunction Trial; MARS, Molecular Diagnosis
and Risk Stratification for Sepsis Cohort Study; PEEP, positive end expiratory pressure;
SAILS, Statins for Acutely Injured Lungs from Sepsis Study.
a 30-day mortality.
Implications for Clinical Trials
Implications for Clinical Trials
Given the number of failed past trials, it is clear that our design of clinical trials
in ARDS needs to evolve, and the preceding evidence suggests that incorporating endotypes
and/or subtypes of disease into clinical trials in ARDS may offer an improved approach.
Prognostic enrichment based on the severity of oxygenation impairment has demonstrated
some success[7]
[41]; however, the true potential of precision medicine rests on predictive enrichment
based on biologically defined phenotypes or endotypes. Endotypes defined by one or
multiple biomarkers and/or clinical characteristics have the potential to select patients
most likely to benefit from new pharmacologic therapies targeting specific biologic
pathways. Clinical trial enrollment could start by measuring a specific plasma marker
and only including those subjects over a prespecified threshold concentration. Alternatively,
innovative adaptive trial designs could enroll all patients regardless of biomarker
concentration, but subsequently adjust randomization so patients not benefiting from
a therapy based on their biomarker level are less likely to be randomized to the therapy
within a clinical trial.[153]
Several challenges exist before biomarker-driven clinical trials for ARDS can be initiated.
First, biomarkers must be rapidly measurable to be of utility in critical care. The
technology to measure protein biomarkers or gene expression rapidly is available;
however, few tests are currently available clinically. Second, biomarker-driven trials
require more knowledge of the performance, stability, and responsiveness of the biomarker
over time as patients progress to ARDS. Third, thresholds of biomarker levels by which
patients may benefit from a therapy are largely unclear and may require further prospective
testing or initial enrollment of all patients in an adaptive clinical trial. Fourth,
we have limited mechanistic understanding of ARDS endotypes and must have strong evidence
that a target drug works only in one subtype prior to initiating a targeted clinical
trial.
Conclusions
In conclusion, ARDS is a heterogeneous syndrome with a lack of therapies directed
at syndrome biology. In the last several years, distinct subphenotypes of ARDS have
been described with distinct clinical, pathologic, radiographic, and/or biologic characteristics.
Early evidence of differential response to therapies based on subtypes defined by
plasma biomarkers and clinical characteristics has been reported. Future work should
focus on furthering our mechanistic understanding of ARDS endotypes, identifying more
important ARDS phenotypes, and developing targeted therapies with the ultimate goal
of applying these therapies to the patients most likely to respond to them.
