Subphenotypes in acute respiratory distress syndrome: A scoping review across clinical, biological, computational, imaging, omics, and artificial intelligence approaches.

Summary

Scoping review of 60 adult studies (2013–2025) across modalities showing reproducible ARDS subphenotypes, particularly hyperinflammatory vs hypoinflammatory classes associated with differences in mortality, ventilator-free days, organ failure, and heterogeneity of treatment effect. Clinical and computational parsimonious classifiers may be closest to bedside translation, whereas imaging/omics/AI need more external validation.

Key Findings

• Across 60 studies, the most robust and externally validated distinction was between hyperinflammatory and hypoinflammatory ARDS subphenotypes, which differ in mortality, ventilator-free days, and organ failure.
• Biological and computational phenotyping approaches showed stronger reproducibility and validation than imaging and omics; parsimonious classifiers (few variables) are feasible for near real-time assignment.
• Secondary analyses of randomized trials indicate heterogeneity of treatment effect by phenotype for interventions such as fluid management, statins, corticosteroids, and recruitment maneuvers.

Clinical Implications

Supports use of hyperinflammatory/hypoinflammatory classification for prognostic enrichment and hypothesis generation in trials; suggests prioritizing parsimonious classifiers (clinical or computational) for real-time bedside application and prospective phenotype-stratified/adaptive trials.

Limitations

• As a scoping review, formal quantitative synthesis (meta-analysis) and bias assessment are limited; heterogeneity across studies complicates direct comparison.
• Variable external validation and inconsistent reporting standards across included studies limit certainty about classifier performance in diverse settings.

Future Directions

Prospective, phenotype-stratified or adaptive RCTs using standardized, transparent algorithms; external validation of parsimonious classifiers in international cohorts; integration of multimodal data with explainable AI to improve interpretability.