Inflammatory phenotypes in acute respiratory distress syndrome: external validation and model simplification in multicenter cohorts.

Background: Inflammatory phenotyping represents a promising approach for risk stratification in acute respiratory distress syndrome (ARDS). A previously developed 18-variable model classifies patients into hyperinflammatory and hypoinflammatory phenotypes; however, its generalizability and feasibility for routine implementation remain uncertain. We aimed to externally validate this model and determine whether a simplified 17-variable version, excluding minute ventilation, maintains equivalent prognostic performance. Methods: This retrospective cohort study was conducted using the Medical Information Market for Intensive Care IV (MIMIC-IV) and electronic intensive care unit (eICU) Collaborative Research Database. In the MIMIC-IV cohort, inflammatory phenotypes were assigned using the original 18-variable model, and their association with 30-day mortality was evaluated using logistic regression. A simplified 17-variable model excluding minute ventilation was also assessed. Model discrimination was compared using the area under the receiver operating characteristic curve (AUC) and DeLong's test. External validation was performed in the eICU cohort. Sensitivity analyses included multiple imputation in MIMIC-IV and complete-case analysis in the eICU cohort. Results: In the MIMIC-IV cohort ( Conclusion: The inflammatory phenotype model demonstrated robust prognostic value across independent ARDS cohorts. A simplified 17-variable model, excluding minute ventilation, maintained an equivalent performance, supporting its potential as a practical and generalizable tool for clinical implementation.