Pre-ECMO characteristics, organ dysfunction and exploratory mortality prediction in COVID-19 versus non-COVID-19 ARDS treated with veno-venous ECMO: a retrospective cohort study

Abstract Objective The transferability of pre-ECMO characteristics and established mortality prediction tools across COVID-19 (C-19) and non-COVID-19 (nC-19) acute respiratory distress syndrome (ARDS) remains uncertain. We compared clinical characteristics, organ dysfunction, respiratory support and outcome in a large cohort of patients treated with veno-venous extracorporeal membrane oxygenation (vvECMO), and explored mortality prediction in C-19 ARDS. Methods This retrospective single-center cohort study included n = 600 vvECMO patients (C-19: n = 245, nC-19: n = 355) hospitalized at the Charité ARDS ECMO Center from 01/2007 to 12/2018 (nC-19) and 03/2020–01/2022 (C-19). Pre-ECMO clinical and laboratory variables were compared between cohorts and by ICU survival. RESP score performance was assessed, and an exploratory C-19 ECMO mortality prediction (CEMP) score was derived from pre-ECMO variables using an explainable machine-learning workflow with SHapley Additive exPlanations (SHAP). Results C-19 patients were older (median 57.0 vs. 51.0 years, p < 0.0001), more often male (74.7% vs. 66.8%, p = 0.0337), and had a higher BMI (29.3 vs. 26.0 kg/m², p < 0.0001). nC-19 patients had a higher CCI (3 vs. 2, p = 0.0137) and SOFA score (12 vs. 11, p = 0.0004). Septic shock before vvECMO initiation was more prevalent in nC-19 patients (65.4% vs. 13.9%, p < 0.0001), and nC-19 patients had higher median baseline bilirubin, creatinine, and lactate levels (all p < 0.0001). nC-19 patients had higher median PEEP, pPeak, and plateau pressure than C-19 patients, whereas their PaO₂/FiO₂ ratio was lower ( p < 0.0001). Duration of vvECMO support was longer in C-19 patients (26.0 vs. 12.6 days, p < 0.0001); ICU mortality did not differ significantly (55.1% vs. 47.9%, p = 0.0966). RESP showed limited discrimination in C-19 patients, and the CEMP derivation analysis yielded an apparent AUC of 0.757 (95% CI 0.69–0.82). Conclusion This large single-center comparative vvECMO cohort highlights differences between C-19 and nC-19 ARDS in organ dysfunction burden, respiratory characteristics, and treatment duration. ICU mortality did not differ significantly between groups. RESP showed limited discrimination in C-19; CEMP remains exploratory and requires calibration and independent external validation before clinical application.