Dynamic oxygenation subgroup bringing new insights in ARDS: more predictive of outcomes and response to PEEP than static PaO2/FiO2

Background Acute respiratory distress syndrome (ARDS) is a rapidly evolving condition. Dynamic assessments using patient trajectories may provide novel insights into disease heterogeneity. The primary objective of this study was to develop and validate dynamic oxygenation subgroups of ARDS based on longitudinal arterial oxygen tension/fractional inspired oxygen (PaO 2 /FiO 2 ) ratios from the day of ARDS diagnosis to day 3. The secondary objective was to determine whether these dynamic physiological subgroups are more effective in predicting patient outcomes and treatment response than the Berlin criteria-defined static oxygenation subgroups. Methods We used group-based trajectory modelling to construct longitudinal oxygenation subgroups over the first 3 days following ARDS diagnosis, based on five ARDS databases. Additionally, we compared these longitudinal subgroups with static Berlin criteria-defined mild, moderate and severe subgroups in terms of clinical characteristics, outcomes and positive end-expiratory pressure (PEEP) responses. Results A total of 814 and 2505 patients with ARDS were included in the training cohort (Chinese ARDS Database) and validation cohorts (FACTT, SAILS, ALVEOLI and MIMIC-IV), respectively. We derived three longitudinal oxygenation subgroups: group 1 (n=406, 49.88%), group 2 (n=302, 37.10%) and group 3 (n=106, 13.02%). The PaO 2 /FiO 2 of the three subgroups exhibited the following trends over time: persistently low, gradually increasing and rapidly improving. The mortality differences across the three longitudinal subgroups (62.59% vs 35.79% vs 17.39%) were significantly greater than those in Berlin criteria-defined subgroups (57.53% vs 41.34% vs 32.18%; χ²=55.65 vs 13.45, p=0.001 for heterogeneity comparison). Additionally, treatment heterogeneity (high vs low PEEP) was more pronounced in longitudinal subgroups (interaction p=0.001) compared with Berlin subgroups (interaction p=0.72). The longitudinal subgroup exhibited variations in patient demographics, PaO 2 /FiO 2 development trajectory and PEEP treatment response, compared with the subgroup defined by the Berlin criteria. Conclusions We identified three longitudinal oxygenation subgroups of ARDS that were more predictive of prognosis and response to PEEP than the subgroups defined by static PaO 2 /FiO 2 ratios in the ARDS Berlin criteria.