LAUSR

In 1967, Ashbaugh et al first described acute respiratory distress syndrome (ARDS)—an acute illness, characterised by tachypnoea, hypoxaemia and loss of lung compliance occurring after a variety of pulmonary and nonpulmonary insults (including trauma, acute pancreatitis, viral pneumonitis). This concept is retained as the ARDS illness model within the current consensus definitions, with acute defined as within 7 days of insult, and hypoxaemia categorised using partial pressure of oxygen/fraction of inspired oxygen concentration (PaO2/FiO2 ratio) into mild (<40 Kpa), moderate (13.3–26.6 Kpa) and severe (≤13.3 Kpa) ARDS on a positive end expiratory pressure of >5 cm water. Fifty years on, ARDS remains a clinical challenge. Globally, ARDS remains clinically underrecognised, with an acute hospital mortality of 46% in patients with severe ARDS. Further, after more than 150 randomised controlled trials (RCTs), we do not have a single drug proven to benefit patients with ARDS. Notably, the histopathological hallmark of ARDS, diffuse alveolar damage (DAD), is only found in half of the patients, and is difficult to ascertain during acute illness. This clinical challenge led to the hypothesis that the heterogeneity of ARDS will manifest as subpopulations with similar clinical, biological, outcome and/or treatment response characteristics. Further, these subpopulations may be unique to ARDS or shared with other critical illness syndromes. If we could identify ARDS subpopulations based on clinical and/or biological characteristics, this may highlight molecular mechanisms to target in RCTs, subpopulations with a higher risk of adverse outcomes or greater treatment responses. Calfee et al have led the field of determining such ARDS subpopulations, primarily with data from patients enrolled into RCTs, using latent class analyses (LCA) of clinical and biomarker data. They consistently report a two class model (two ARDS subpopulations or subphenotypes) as the best fit for the clinical and biomarker data analysed. The hyperinflammatory ARDS subpopulation (Phenotype-2) is less common, characterised by higher plasma concentrations of cytokines, greater vasopressor use, lower serum bicarbonate concentrations, and a higher prevalence of sepsis, when compared with the more common hypoinflammatory ARDS subpopulation (Phenotype-1). Importantly, hyperinflammatory ARDS has higher mortality and differential treatment response to PEEP, simvastatin and fluid management. Further, these phenotypes are stable over the first 3 days, giving an enrolment window for RCTs and they can be identified with limited biomarker information. A similar two ARDS subpopulation model primarily in patients enrolled into observational cohort study, using only biomarker data and with clustering analysis, has also been reported by Bos et al. For such ARDS subpopulations to be useful, they must have feasible diagnostic standards to enable categorisation at the bedside, while remaining reproducible and biologically informative. We have actively avoided the term ‘endotype’ as it denotes subpopulations with specific biological mechanisms, whereas these ARDS subpopulations represent a cluster of visible properties and are best referred to as phenotypes. In this context, let us consider the work by Sinha et al . The authors tested whether the hyperinflammatory and hypoinflammatory subpopulations reported in RCTs are identifiable in two observational cohorts—a singlecentre cohort study of Validating Acute Lung Injury markers for Diagnosis (VALID) and a twocentre cohort study of Early Assessment of Renal and Lung Injury (EARLI). First, they performed LCA and showed that a two class model provided best fit for the data from VALID and EARLI cohorts. Second, they report that three or four marker classifiers (consisting of interleukin 8, bicarbonate,