Early metabolic trajectory and sedation strategy in V-V ECMO for severe pneumonia: a multicenter cohort study

Prognostic assessment in patients with severe pneumonia requiring veno-venous extracorporeal membrane oxygenation (V-V ECMO) has traditionally relied on static baseline characteristics. Whether the early physiological trajectory and modifiable management strategies, specifically sedation depth, influence survival remains uncertain.

We conducted a multicenter retrospective cohort study. A total of 1484 adult patients with severe pneumonia supported by V-V ECMO were analyzed. Machine learning techniques were employed to objectively identify key predictors of in-hospital mortality among 341 candidate variables. The impact of early sedation strategy (Light vs. Deep) on survival and organ complications was assessed using multivariable logistic regression and propensity score matching (PSM).

Machine learning analysis identified 24-h lactate level as the most important variable, outperforming baseline age and pH. An exploratory threshold of 4.15 mmol/L for 24-h lactate was identified; mortality risk increased substantially beyond this value. Regarding management, the Light Sedation strategy was associated with lower norepinephrine requirements and a reduced incidence of severe AKI requiring CRRT (29.4% vs. 42.9%, p < 0.001). In multivariable analysis adjusted for baseline severity and complications, Light Sedation remained associated with improved survival after adjustment (Adjusted OR 0.612, 95% CI 0.475-0.784, p < 0.001). This survival benefit was confirmed in the PSM cohort (OR 0.54, p < 0.001).

In severe pneumonia patients supported with V-V ECMO, survival associated with the early trajectory of metabolic recovery. A light sedation strategy was consistently associated with improved survival and reduced organ dysfunction; however, causality cannot be established in this observational study, supporting further investigation of physiology-guided and dynamically reassessed sedation strategies in this high-risk population.