The current study has several strengths, including the use of two distinct cohorts to replicate findings, a diverse patient population drawn from multiple centers around the United States, the use of a broad panel of biomarkers designed to capture several different aspects of ARDS pathogenesis, and rigorous approaches to biomarker measurement using the same validated assays for both cohorts. The study also has some limitations. First, not all biomarkers previously associated with ARDS pathogenesis or prognosis were measured. Other biomarkers associated with lung epithelial cell injury (club cell 16, KL-6),34,35 endothelial injury (soluble intercellular adhesion molecule-1, circulating glycosaminoglycans),24,30 disordered coagulation and fibrinolysis (plasminogen activator inhibitor-1, protein C, thrombomodulin),36,37 and inflammation (IL-1/IL-1-receptor antagonist, soluble tumor necrosis factor receptor 1),38 to name a few, might further enhance the molecular phenotypes of epithelial and endothelial injury in direct vs indirect ARDS. Second, the larger of the two patient samples used in these analyses came from a secondary analysis of a randomized controlled trial, which excluded many patients at highest risk for mortality from ARDS and was not designed to test the hypothesis under study in this analysis. This could potentially limit generalizability and dampen the strength of the prognostic associations. This limitation is mitigated by the inclusion of the single-center sample from a cohort that had many fewer exclusions and in which findings were largely similar to the multicenter sample. Third, the etiology of ARDS was identified in the multicenter cohort by the investigative team at the local site in contrast to the single-center study in which a single investigator (L. B. W.) classified all ARDS risk factors. This inherent variability introduced by the multicenter design may partly explain why some of the biomarker differences are slightly less robust in the multicenter cohort. Finally, biomarker data were not available on all patients enrolled in the multicenter cohort. Biomarker data were missing for some patients, as further detailed in e-Appendix 1, because of the lack of plasma availability; however, no substantive differences were found between patients with and without plasma samples (data not shown).