Background: Acute respiratory distress syndrome (ARDS) poses challenges in early identification. Exhaled breath contains metabolites reflective of pulmonary inflammation. Aim: To evaluate the diagnostic accuracy of breath metabolites for ARDS in invasively ventilated intensive care unit (ICU) patients. Methods: This two-center observational study included critically ill patients receiving invasive ventilation. Gas chromatography and mass spectrometry (GC–MS) was used to quantify the exhaled metabolites. The Berlin definition of ARDS was assessed by three experts to categorize all patients into “certain ARDS”, “certain no ARDS” and “uncertain ARDS” groups. The patients with “certain” labels from one hospital formed the derivation cohort used to train a classifier built based on the five most significant breath metabolites. The diagnostic accuracy of the classifier was assessed in all patients from the second hospital and combined with the lung injury prediction score (LIPS). Results: A total of 499 patients were included in this study. Three hundred fifty-seven patients were included in the derivation cohort (60 with certain ARDS; 17%), and 142 patients in the validation cohort (47 with certain ARDS; 33%). The metabolites 1-methylpyrrole, 1,3,5-trifluorobenzene, methoxyacetic acid, 2-methylfuran and 2-methyl-1-propanol were included in the classifier. The classifier had an area under the receiver operating characteristics curve (AUROCC) of 0.71 (CI 0.63–0.78) in the derivation cohort and 0.63 (CI 0.52–0.74) in the validation cohort. Combining the breath test with the LIPS does not significantly enhance the diagnostic performance. Conclusion: An exhaled breath metabolomics-based classifier has moderate diagnostic accuracy for ARDS but was not sufficiently accurate for clinical use, even after combination with a clinical prediction score.
Original languageEnglish
Article number96
JournalCritical Care
Issue number1
Publication statusPublished - 1 Dec 2024


  • ARDS
  • Breath analysis
  • Prediction model
  • VOCs

Cite this