Sinha P, et al. - Given hypoinflammatory and hyperinflammatory phenotypes (two distinct phenotypes) of acute respiratory distress syndrome (ARDS) have been identified by applying latent class analysis (LCA) in five randomised controlled trial (RCT) cohorts, and differential outcomes as well as treatment response related to the phenotypes have been reported, researchers undertook this retrospective analysis to create parsimonious models that could be accurate as well as feasible for clinical use for phenotype identification. They used derivation dataset (n = 2,022), including three RCT cohorts from the National Lung, Heart, and Blood Institute ARDS Network (ARMA, ALVEOLI, and FACTT), to derive machine learning and logistic regression classifer models. A fourth cohort (SAILS; n=715) was utilized from the same network to obtain validation. In all of these cohorts, LCA-derived phenotypes were considered as the reference standard. Interleukin (IL)-8, IL-6, protein C, soluble tumour necrosis factor receptor 1, bicarbonate, and vasopressor use were identified as the 6 most important classifier variables. Three-variable (IL-8, bicarbonate, and protein C) and four-variable (3-variable plus vasopressor use) models, from the nested models, were adjudicated to be the best performing. The good accuracy of both models against LCA classifications was evident in the validation test set. Overall, it was inferred that parsimonious classifier models using three or four variables can afford accurate identification of ARDS phenotypes. Pending the development of real-time testing for important biomarkers and prospective validation, these models could aid the recognition of ARDS phenotypes to allow their use in clinical trials and practice.