To identify discrete acute respiratory distress syndrome subphenotypes among VV-ECMO patients using high-resolution electronic health record clustering and assess differences in clinical outcomes.
Approach:
Study Design: Single-centre analysis of 598 adult patients with ARDS treated with VV-ECMO, using K-means clustering and Shapley Additive Explanations models.
Key Findings:
Distinct subphenotypes identified, primarily driven by inflammation (procalcitonin and C-reactive protein), kidney/liver function (creatinine and urea), coagulation (fibrinogen and D-dimer), and mechanical ventilation (positive end-expiratory pressure).
Survival rates varied significantly between clusters, particularly within kidney/liver function (32%) and combined parameter categories (21%).
Clusters with multi-organ dysfunction had longer ICU lengths of stay.
Interpretation:
Early data-driven clustering of EHR parameters identifies clinically meaningful ARDS subphenotypes in VV-ECMO patients.
Limitations:
Study conducted at a single centre, which may limit generalizability.
Exclusion of patients managed without cannulation at referring hospitals.
Conclusion:
Subphenotype-based stratification may refine risk stratification and management in severe ARDS treated with VV-ECMO.
