Introduction

The first definition of acute respiratory distress syndrome (ARDS) dates to Ashbaugh and colleagues in 1967 when they described 12 patients with severe acute respiratory failure [1]. These patients had severe hypoxemia that was refractory to supplemental oxygen, but which in some cases was responsive to the application of positive end-expiratory pressure (PEEP). Widespread pulmonary inflammation, edema, and hyaline membranes were observed on autopsy.

Over the next 25 years several definitions were proposed, but there was no single definition for ARDS that was widely accepted and used. In 1994, broad consensus was achieved when the American-European Consensus Conference (AECC) published a definition [2, 3]. This group defined ARDS as the acute onset of hypoxemia (the ratio of partial pressure of arterial oxygen to fraction of inspired oxygen [PaO2/FiO2] ≤200 mmHg), with bilateral infiltrates on frontal chest X-ray, in the absence of left atrial hypertension. They also defined a new over-arching entity termed acute lung injury (ALI), which used the same variables but with a less stringent criterion for hypoxemia (PaO2/FiO2 ≤300 mmHg).

The AECC definition has been widely adopted by clinicians and researchers and has facilitated the acquisition of a great deal of data about ARDS in the ensuing two decades. Over this same time, however, a number of issues with the AECC definition have become apparent [4, 5]. Evaluating the definition as a whole, a modified AECC definition (applied retrospectively using the entire ICU length of stay and requiring four-quadrant air-space disease) was compared to a reference standard of diffuse alveolar damage seen on autopsy. Under these conditions the AECC definition performed reasonably well, with a sensitivity of 75 % and a specificity of 84 % [6]. However, when the AECC definition criteria were strictly applied (bilateral chest X-ray infiltrates) on a daily basis, the sensitivity remained reasonable at 84 %, but the specificity was significantly lower at only 51 % [7]. Moreover, ALI, as defined using the AECC criteria, is under-recognized by clinicians, particularly the subgroup of patients with milder hypoxemia (i.e., with PaO2/FiO2 201–300) [7–9].

The AECC definition requires that onset of respiratory failure be acute, but does not explicitly define the specific timeframe (e.g., hours, days, or weeks). The hypoxemia criterion has generated concerns because PaO2/FiO2 may vary with FiO2, and also in response to other ventilator settings, particularly PEEP [10–15]. The chest X-ray criterion has only moderate inter-observer reliability even when applied by experts, although this can be improved through use of a training set of radiographs [16, 17]. Finally, although the AECC definition includes a pulmonary artery wedge pressure (PAWP) ≤18 mm Hg (when measured), patients with hallmark findings of ARDS often have an elevated PAWP because of elevated pleural pressures and/or vigorous fluid resuscitation [18, 19].

For these reasons and because all definitions should be reviewed and adjusted periodically to reflect new information and experience, we convened a consensus panel to address the limitations of the previous AECC definition and propose revisions. The methodology and the empirical evaluation of the Berlin definition have been published previously [22]; the current publication provides an expanded discussion of the rationale underlying the development of the Berlin definition, and provides further supplementary material (e.g., examples of typical chest X-rays from patients with ARDS) to facilitate its application.

Methods

As described previously [22], the panel revised the definition of ARDS for use in clinical practice and clinical research, using a conceptual model incorporating reliability, validity and a novel iterative approach that included a formal evaluation of the definition using large cohorts of patients with the syndrome (as originally defined using the AECC definition). The Task Force was initiated by the European Society of intensive care medicine (ESICM), which supported the process with an unrestricted grant and all necessary logistics. This grant covered the costs of the face-to-face meeting in Berlin and also a number of subsequent teleconferences. There was no support received from industry for any part of the process.

The ESICM identified three chairs with expertise in ARDS, clinical epidemiology, physiology, and clinical trials (Ranieri, Rubenfeld, and Thompson), who together selected the participants and created the agenda. Panelists (listed in the Appendix) were selected based on their recognized expertise in the field of ARDS, and to ensure intellectual diversity. Specifically, the chairs tried to achieve a balance of opinions by clinicians, trialists, epidemiologists, physiologists, and basic scientists. The chairs also included two junior members with an interest in ARDS.

All modifications to the ARDS definition were based on the principle that syndrome definitions must fulfill three criteria: feasibility, reliability and validity [20]. Feasible definitions should rely on diagnostic tests and/or clinical data that are routinely used by intensivists and can be performed in a short enough time frame to facilitate clinical trial enrolment. Since it is essential for researchers to identify patients with similar characteristics across studies, and for clinicians to apply the results of research at the bedside, syndrome definitions must be reliable as measured by inter-observer agreement. The most common technique to establish the validity of diagnostic criteria requires a gold standard, which is not available in ARDS. However, many syndromes in medicine are defined without a reference standard including depression, sepsis, and community-acquired pneumonia. Various indirect approaches exist to evaluate validity (e.g., face, construct, predictive, and concurrent validity) for syndromes that do not have reference standards (Table 1). In addition to feasibility and reliability, we emphasized face validity (the “conceptual model” of how clinicians recognize the syndrome) and predictive validity (the ability of the syndrome definition to predict outcome and/or response to therapy). Finally, the panel felt that any revision of the definition should be compatible with prior definitions to facilitate interpretation of older studies.

Table 1 Glossary of terms and their application in the Berlin definition
Full size table

Each presenter was asked to evaluate potential defining criteria using a framework focusing on reliability, feasibility, and validity. We used an informal consensus technique involving whole group discussions moderated by one of the chairs, conducted in-person and by teleconference. Consensus was demonstrated by unanimous consent around presented options. When issues were contentious, individual yes/no votes were counted to ensure that all panelists had an opportunity to voice their opinion.

After 2 days of in-person discussions (September 30–October 1, 2011) in Berlin, Germany a draft definition was proposed [21]. This draft definition was then evaluated using existing ARDS databases to determine characteristics of patients in each category of ARDS and examine predictive validity for mortality. Finally, the panel reconvened by multiple teleconferences in February 2012 to conduct further discussions and produce the final Berlin Definition of ARDS [22]. Prior to submission for publication the consensus definition underwent a process of independent peer review and subsequent endorsement by each of the ESICM, the American Thoracic Society, and the Society of Critical Care Medicine.

The ARDS conceptual model

Face validity derives from an understanding of how clinicians recognize patients with the syndrome, therefore, considerable discussion focused on developing a conceptual model of ARDS. The panel agreed that ARDS is a type of acute diffuse lung injury associated with a predisposing risk factor, characterized by inflammation leading to increased pulmonary vascular permeability and loss of aerated lung tissue. The hallmarks of the clinical syndrome are hypoxemia and bilateral radiographic opacities [23] (using standard chest X-ray or computed tomography [CT] scan), associated with several physiological derangements including: increased pulmonary venous admixture, increased physiological dead space, and decreased respiratory system compliance. The morphological hallmarks in the acute phase are lung edema, inflammation, hyaline membranes, and alveolar hemorrhage (i.e., diffuse alveolar damage) [24].

There are many common etiologic risk factors for ARDS, which the AECC definition classified into direct and indirect lung injury categories. Although some experimental and clinical studies show modest overall differences in the inflammatory responses and radiographic patterns as well as physiologic responses to ventilatory treatment, the direct and indirect categories overlap to such a large degree that the committee decided not to include direct and indirect ARDS as distinct entities in the Berlin definition (Table 2). Identification of the risk factor leading to ARDS in an individual patient, regardless of its direct or indirect nature, rather serves to guide therapy for the underlying disease leading to ARDS.

Table 2 Common risk factors for ARDS (adapted with permission [22])
Full size table

The Berlin ARDS definition

The resultant “Berlin Definition” has been published previously and is outlined in Table 3 [22]. Three mutually exclusive categories of mild, moderate, and severe ARDS were created to provide better separation of prognosis and treatment selection. The rationale, controversies, and recommendations for each diagnostic criterion are discussed in the following sections.

Table 3 The Berlin definition of ARDS (with permission from [22])
Full size table

The original AECC definition stated that the onset of ARDS was acute, to exclude chronic pulmonary conditions that can cause hypoxemic respiratory failure, however, it did not state an explicit time frame. Observational data suggest that the majority of patients with ARDS are identified within 72 h of the recognition of the underlying risk factor, with nearly all patients identified within 7 days [25, 26]. The panel therefore defined “acute onset” as ARDS developing within 1 week of a known clinical insult or new/worsening respiratory symptoms.

The AECC definition required bilateral infiltrates consistent with pulmonary edema on frontal chest radiograph, but there is poor inter-observer reliability in interpreting chest radiographs using this definition among intensivists and radiologists [16, 17]. To help address this issue the panel attempted to make the chest radiograph criterion more explicit specifying that it should include bilateral opacities consistent with pulmonary edema that are not fully explained by effusions, lobar/lung collapse, or nodules/masses on chest radiograph as a defining criterion for ARDS, and also recognized that opacities seen on CT scan could be substituted if available. The CT scan abnormalities were not included in the definition as a core part of the definition because they were not considered feasible at the current time due to concerns regarding safety, cost, and lack of widespread availability. In addition, to enhance inter-observer reliability, we have included a set of chest radiographs judged by the panel to be illustrative of the spectrum of images that are consistent, inconsistent, or equivocal for the diagnosis of ARDS (see Online Supplement for images and accompanying commentary). The committee considered a requirement for more extensive opacities (i.e., at least three quadrants) to define severe ARDS; however, this did not improve predictive validity for mortality and was dropped after further consensus discussion [22].

As with the AECC definition, the panel recognized that hydrostatic edema (i.e., cardiac failure or fluid overload) is one of the most common alternative diagnoses in patients presenting with ARDS. However, given the declining use of pulmonary artery catheters worldwide and the recognition that hydrostatic edema and ARDS may coexist [18, 19], the pulmonary artery wedge pressure (PAWP) criterion was removed. The panel therefore decided that patients whose respiratory failure is not fully explained by cardiac failure or fluid overload as judged by the treating physician using all available data may qualify as having ARDS. Nevertheless, if no known etiologic risk factor for ARDS is apparent (Table 2), objective evaluation of cardiac function (e.g., echocardiography or cardiac output measurement) is required to help rule out hydrostatic edema secondary to heart failure. In order to improve reliability and clarity in judging this criterion, the panel also developed a number of clinical vignettes illustrating cases that would and would not qualify as ARDS based on ruling out hydrostatic edema (Online Supplement).

The AECC definition classified ARDS by PaO2/FiO2 ratio regardless of the level of PEEP [2]. Since PEEP can affect the reliability and specificity of PaO2/FiO2, to classify the severity of ARDS [10, 13], a minimum level of 5 cm H2O PEEP (or non-invasive CPAP for Mild ARDS) has been included in the updated definition. The panel had originally proposed a requirement of PEEP ≥10 cm H2O as a criterion for the severe ARDS group, but this was subsequently dropped as it did not improve predictive validity for mortality within the group of patients with PaO2/FiO2 ≤100 [22].

Recent data suggest that for a given PaO2/FiO2 ratio, higher FiO2 is associated with a higher mortality [15]; however, since clinicians normally titrate FiO2 to maintain a PaO2 between 60–80 mm Hg, most patients with a PaO2/FiO2 ≤100 mm Hg already have a FiO2 of 0.7 or higher. Therefore, to avoid additional complexity that may have decreased feasibility; the committee did not add a minimum FiO2 requirement to the definition.

Some physicians rely on oxyhemoglobin saturation by pulse oximetry (SpO2) instead of arterial blood gas analysis to monitor arterial oxygenation, especially in patients with less severe ARDS. The SpO2/FiO2 ratio correlates with the PaO2/FiO2 ratio [27], so feasibility of the definition could be enhanced by the use of SpO2 in place of PaO2. However, in some patients SpO2 is not concordant with PaO2 [28]. In particular, with SpO2/FiO2 patients receiving FiO2 of 1.0 and with SpO2 of 100 % could be classified as Severe ARDS, regardless of their PaO2, and therefore the SpO2/FiO2 ratio may misclassify patients. Therefore, the committee did not include SpO2/FiO2 as an alternative to PaO2/FiO2 in the revised definition.

The term acute lung injury (ALI) (i.e., all patients with PaO2/FiO2 ≤300) was removed from the ARDS definition, due to the perception that many clinicians and researchers viewed ALI as a category of patients (i.e., PaO2/FiO2 201–300) that is separate from ARDS rather than an umbrella term for all patients (leading to the frequent use of the term ALI/ARDS that has no interpretation in the AECC definition). Under this new framework, each subcategory of ARDS (mild, moderate, severe) is defined by mutually exclusive ranges of PaO2/FiO2. The creation of the mild ARDS category formalizes what was previously perceived as patients with a less severe form of the syndrome but by applying the term ARDS, it recognizes the severity of their illness (mortality 27 % [22]) and response to lung protective ventilation [29]. For severe ARDS, two possible thresholds were considered, PaO2/FiO2 ≤150 or ≤100. The panel decided that the lower threshold may better represent the subgroup in which one could consider therapies that, in some studies appear to be beneficial in severe ARDS, such as prone positioning [30]. The committee felt that the new PaO2/FiO2 thresholds chosen for the different levels of ARDS severity could be helpful in categorizing patients with respect to different therapeutic approaches (Fig. 1).

Fig. 1
figure 1

Aligning Therapeutic Options with The Berlin Definition (adapted from [48] with permission). This figure depicts potential therapeutic options according to the severity of ARDS. Boxes in yellow represent therapies that in the opinion of the panel still require confirmation in prospective clinical trials. This figure is just a model based on currently available information. In the coming years, various aspects of the figure will likely change; proposed cut-offs may move, and some therapies may be found to not be useful, while others may be added

Full size image

Additional physiological measurements

Plateau pressure (as measured after an end-inspiratory pause) reflects the combined effects of tidal volume, PEEP, and the compliance of the respiratory system and is associated with mortality [31]. Theoretically, routine measurement of the plateau pressure may help in classifying patients based on their ARDS severity, aid in setting the ventilatory strategy, and in following the evolution of disease over time. However, plateau pressure is not routinely measured in some centers, and the use of certain ventilator modes (e.g., pressure-controlled ventilation) or supported ventilation with spontaneous breathing (e.g., pressure-support ventilation) render the measurement of plateau pressure impractical.

Static respiratory system compliance (i.e., the change in lung volume for a given driving pressure defined as tidal volume divided by plateau pressure minus PEEP) reflects the degree of lung volume loss. Although this measurement has the same concerns regarding feasibility as plateau pressure, the panel initially suggested including it (compliance <40 mL/cmH2O) as part of the definition of Severe ARDS, based on a prior definition using a scoring system for ARDS [32]. Increased dead space is common in patients with ARDS and is associated with increased mortality [33, 34]. Measuring dead space, however, is challenging and was felt not to be feasible in routine clinical practice. As a rough surrogate of dead space, a minute ventilation corrected for PaCO2 (calculated as minute ventilation × PaCO2/40 mmHg) threshold of >10 L/min was initially chosen by the panel for the definition [35], as it corresponds approximately to a dead space fraction of 50 %, which was associated with increased mortality in a previous study [34]. However, static compliance and corrected minute ventilation were dropped from the final Berlin definition as they increased complexity without improving predictive validity. The panel was careful to emphasize, however, that the exclusion of these variables from the definition in no way decreases their importance in the daily evaluation and management of patients with ARDS.

Other criteria

We considered a number of additional potential diagnostic criteria for inclusion in the Berlin definition; a complete list of these and the reasons for their exclusion are shown in Table 4. Extravascular lung water (EVLW) can be measured using transpulmonary thermodilution techniques and high levels are associated with increased mortality in ARDS patients [36]. At the present time, technology to measure EVLW is relatively costly, invasive, not widely available, and has significant methodological limitations [37], so we did not include it in the definition. The panel also considered the potential inclusion of biomarkers (e.g., IL-6, TNF-α) or genetic markers (e.g., ACE gene polymorphism) to aid in the identification of patients with ARDS. Despite a large number of candidate biomarkers and genetic markers studied [38, 39], none has currently demonstrated adequate sensitivity and specificity for use in the diagnosis of ARDS [38–40].

Table 4 Criteria considered but not included in the Berlin definition (with permission from [21])
Full size table

Although the pathologic correlate of ARDS is diffuse alveolar damage (DAD), studies have demonstrated only moderate agreement between the clinical diagnosis of ARDS and DAD at autopsy [6]. Members of the panel were not in complete agreement that DAD is the sole pathologic correlate of ARDS, and some considered pneumonia and non-cardiogenic edema as compatible with ARDS when clinical criteria are met. Because of this, and since making a pathological diagnosis of ARDS using lung biopsy may be associated with increased risk of complications, the committee did not include this in the definition. At the present time, lung biopsy may be considered in patients with persistent ARDS of unknown etiology to rule out an underlying etiology that may respond to a specific treatment [41, 42].

Continuous phenotype definition for ARDS

Previous investigators have hypothesized that differences in mortality rates reported in observational studies and variable responses to therapy in clinical trials may be due to arbitrary thresholds for hypoxemia in the definition, as well as variability in the interpretation of the other diagnostic criteria [43]. The panel considered the possibility of defining a continuous phenotype (e.g., probability and/or severity of ARDS defined with a numeric score) rather using a simple yes/no definition, which may better represent the clinical spectrum of ARDS [44]. Given the increase in complexity this would entail, along with uncertainty regarding the meaning and importance of an intermediate phenotype (e.g., meeting oxygenation criteria but with equivocal chest X-ray findings), the panel felt that further research was required before considering a continuous phenotype in the definition of ARDS.

The Berlin definition and future ARDS clinical trials

The panel recommends that future trials be designed using one or more of the ARDS subgroups as a base study population, which may be further refined using physiologic and/or other criteria specific to the putative mechanism of action of the study intervention (e.g., IL-6 levels for a trial of an IL-6 antagonist; more stringent hypoxemia criteria for a study on extracorporeal membrane oxygenation) (Fig. 1). The different categories of ARDS (mild, moderate, severe) may also be useful for stratification in clinical trials.

Future research on the Berlin ARDS definition

Some important first steps have already been taken in determining the frequency distribution of patients within each subgroup of ARDS using this new definition [22]. Moreover, this work provides investigators with an idea of the proportion of patients in each risk category who may be eligible for inclusion in future clinical trials of therapeutic interventions and observational studies.

Implementation of the Berlin definition will over time allow for a formal evaluation of its feasibility (including barriers/facilitators to its uptake in clinical practice and research settings) and reliability of case identification based on each of the major diagnostic criteria (e.g., opacities on chest radiography, exclusion of hydrostatic pulmonary edema). The face and content validity of the Berlin definition may be assessed through feedback from a wider panel of clinicians and researchers. Finally, the criterion validity of the definition may be assessed by comparison to autopsy findings, both from pre-existing data and in future studies.

Ongoing research into the identification of accurate diagnostic and/or prognostic genetic polymorphisms or biomarkers for ARDS may help to further improve the specificity of the definition in the future. Similarly, reproducible and valid methods for the direct measurement of pulmonary vascular permeability or extravascular lung water will be important advances over current methods of assessing the presence and origin of lung edema, and could be incorporated into the future definition of ARDS.

Conclusions

The Berlin definition was developed to achieve a more reliable definition that will facilitate case recognition and better match treatment options and clinical outcomes to severity of illness categories. Important incremental advances in this ARDS definition include: the focus on feasibility, reliability, and validity during definition development; the incorporation of an empiric evaluation process in refining the definition [22]; and the creation of explicit examples to aid in application of the radiographic and origin of edema criteria (Online Supplement). The Berlin Definition will need to evolve as new information and experience is gained from its widespread implementation in clinical practice and research, as well as from the development of new diagnostic tools (e.g., imaging techniques, biomarkers, extravascular lung water measurement) which may be considered for inclusion in future ARDS definitions.