Extracorporeal membrane oxygenation (ECMO) can partially substitute for cardiopulmonary function when conventional therapies fail, thereby providing critically ill patients with an opportunity for further definitive management [1]. With continuous advances in membrane oxygenators, centrifugal pumps, and anticoagulation strategies, the indications for ECMO have expanded, and its utilization has become increasingly frequent in emergency and critical care settings [2, 3]. The two major configurations are venovenous ECMO (VV-ECMO), for severe respiratory failure, and venoarterial ECMO (VA-ECMO), for cardiogenic shock or circulatory collapse. Accumulating evidence suggests that ECMO can improve short-term survival in selected patients and has become an important modality in the management of life-threatening critical illness [4–6].

Despite substantial advances, ECMO continues to carry high in-hospital mortality and complex complications, and further investigation is needed into short- and long-term outcomes and their determinants. Most existing mortality risk models have been derived from single-center, small-sample cohorts predominantly conducted outside China, characterized by heterogeneous variable selection and limited external validity [7, 8]. Several studies have proposed multivariable prognostic models and identified predictors (e.g., APACHE II, lactate, albumin); however, variability in model specification and performance assessment has limited comparability and hindered a unified, widely applicable framework [4, 9]. Moreover, given the substantial differences between VA- and VV-ECMO in indications, interventions, and underlying pathophysiology, key risk factors may be diluted or obscured. High-quality, population-specific evidence in emergency ECMO is scarce, and rigorous epidemiologic data remain limited. Accordingly, large, multicenter studies are needed to evaluate in-hospital mortality among ECMO recipients, identify independent risk factors, and compare outcomes across ECMO modes. Using multicenter retrospective data from emergency ECMO cases in China, we assessed in-hospital mortality and its determinants to support precise risk stratification and individualized management.

(1) Age ≥ 18 years; (2) Underwent ECMO at one of the three participating centers during the study period; (3) In-hospital outcome data available.

(1) Documented brain death before ECMO initiation (e.g., organ donors);

(2) ECMO initiated for planned short-term postoperative support, such as after lung transplantation;

(3) Underlying conditions likely to confound outcome assessment, including end-stage malignancy or severe trauma;

(4) Incomplete key clinical or laboratory data that preclude statistical analysis.

All data were de-identified prior to extraction and were stored and managed in encrypted form to ensure security, confidentiality, and regulatory compliance.

Participants were categorized by in-hospital outcome (survivors vs non-survivors). Subgroup analyses were prespecified by ECMO mode (VV or VA). The primary endpoint was in-hospital all-cause mortality. The secondary endpoints included early mortality (≤ 5 days after ECMO initiation) and the difference in in-hospital mortality between ECMO modes.

This multicenter study included 304 ECMO recipients from three emergency ECMO centers, constituting one of the largest cohorts reported in China. The overall in-hospital mortality was 36.2%, and more than half of the deaths occurred within the first 5 days after ECMO initiation. Multivariable analyses identified higher APACHE II score, lactate, and systolic blood pressure as independent predictors of in-hospital mortality, whereas higher diastolic blood pressure and albumin were protective. Subgroup analyses demonstrated significantly poorer survival among patients receiving VA-ECMO than those treated with VV-ECMO. These findings underscore the importance of meticulous patient selection and early-phase management in emergency ECMO. Although outcomes in this population have improved compared with earlier domestic reports, mortality remains substantial, highlighting the need for further optimization of treatment strategies and providing new evidence on mode-specific prognostic differences.

The in-hospital mortality in our cohort (36.2%) was higher than that reported in a multicenter Chinese study by Cheng et al. (29.6%) [11] but lower than in several international cohorts [4, 12, 13]. Several factors may account for these differences. First, the case mix differed: VA-ECMO accounted for 65% of our cohort—higher than in some prior Chinese series—and, given its indication for circulatory failure and higher mortality than VV-ECMO, this likely increased the overall mortality rate [11, 13]. Second, baseline illness severity was high among emergency ECMO candidates. APACHE II is an independent predictor of mortality in ECMO recipients [4, 14]. In our cohort, the median APACHE II exceeded 20; among non-survivors, the median was 28 and 25% had scores > 33—consistent with greater baseline severity and plausibly contributing to higher mortality. Third, CRRT exposure was substantial. Prior studies report ~ 3.7-fold higher in-hospital mortality among ECMO patients receiving CRRT, with mortality exceeding 60% [15]. Mechanistically, concomitant CRRT may worsen hemocompatibility-related cellular injury and amplify systemic inflammation [16, 17]. In our cohort, ~ 50% received CRRT, which was associated with a significantly higher risk of death. Finally, ECMO center practices have become increasingly standardized. About 15 years ago, a Taiwanese cohort reported overall mortality approaching 60% [18]. More recently, the establishment of specialized ECMO teams and progressively standardized protocols for patient selection, anticoagulation, and complication management across centers—domestically and internationally—has markedly improved care quality, which has been closely linked to better outcomes [5, 11–13].

In both the overall cohort and the VA-ECMO subgroup, APACHE II remained an independent predictor of in-hospital mortality, consistent with Cui et al [4], supporting its continued prognostic utility. By contrast, APACHE II showed limited discrimination in the VV-ECMO subgroup. plausibly because VV-ECMO candidates typically have isolated respiratory failure and relatively preserved circulation, whereas APACHE II weights hemodynamic and neurologic components. Accordingly, ELSO guidance recommends tools tailored to respiratory ECMO (e.g., RESP) to improve survival estimation in VV-ECMO [19]. Regarding lactate, prior studies highlight its early-warning value [20, 21]. In our data, lactate predicted adverse outcomes in the overall cohort and in VV-ECMO, but not in VA-ECMO, likely because hyperlactatemia in VA-ECMO reflects not only hypoperfusion but also vasoactive support and impaired clearance, reducing specificity. Consistent with prior reports, lactate clearance or trajectories appear more informative than single time points [20, 22]. Blood-pressure indices also diverged: higher systolic pressure correlated with worse outcomes–suggesting “pseudo-stability” under vasopressors amid low cardiac output and microcirculatory dysfunction–whereas higher diastolic pressure was protective, in line with its role in coronary/renal perfusion and with reports of diastolic pressure–guided resuscitation benefiting VA-ECMO patients [23, 24].