Association between respiratory phenotypes and successful ventilator liberation in early-onset and late-onset ARDS: a multicenter cohort study

Early-onset and late-onset acute respiratory distress syndrome (ARDS) differ in clinical course and outcomes, but their longitudinal respiratory trajectories remain unclear. We aimed to characterize respiratory trajectories in early- and late-onset ARDS and identify trajectory-based phenotypes associated with ventilator liberation and mortality. In this multicenter retrospective cohort study, 907 patients with ARDS (519 early-onset, 388 late-onset) were analyzed. Partitioning around medoids clustering was applied to longitudinal trajectories of the PaO 2 /FiO 2 ratio, tidal volume, and driving pressure in the overall cohort and onset-specific subgroups. Associations with outcomes were evaluated using multivariable regression, propensity score matching, and inverse probability weighting. Early-onset and late-onset ARDS showed distinct respiratory trajectories and clinical outcomes. Three reproducible respiratory trajectory phenotypes were identified across the overall cohort and onset-specific subgroups. The high-risk phenotype was characterized by persistent oxygenation impairment, elevated driving pressure, and higher tidal volume. Compared with the reference phenotype, the high-risk phenotype was associated with lower odds of successful ventilator liberation within 7 days (OR 0.39, 95% CI 0.16–0.74) and higher 28-day mortality (OR 10.84, 95% CI 7.41–16.07). Exploratory pathway association analyses identified APACHE II score, SOFA score, and work of breathing as variables with correlational pathways linking high-risk respiratory trajectory phenotype to failure of early ventilator liberation, while APACHE II score showed the strongest correlational pathway linking the phenotype to 28-day mortality. Landmark sensitivity clustering restricted to longitudinal respiratory measurements within the first 48 h after ARDS onset identified three consistent respiratory subphenotypes. The high‑risk cluster (Cluster 3) exhibited higher 28 day all-cause mortality and lower 7 day successful ventilator liberation rate. These group level trends recapitulated the directional pattern from primary analysis. Early-onset and late-onset ARDS exhibit distinct longitudinal respiratory trajectories. Trajectory-based respiratory phenotyping identifies a high-risk subgroup associated with ventilator liberation failure and mortality, supporting the use of dynamic respiratory phenotyping for risk stratification in ARDS. The predictive and clinical guiding value of dynamic respiratory phenotyping needs validation via prospective cohorts.