Left ventricular unloading during VA-ECMO for refractory cardiogenic shock: a target trial emulation multicenter analysis Authors: Angela Dettling, Ouriel Saura, Lisa Dilange, David Levy, Manuela Lucenteforte, Benedikt N Beer, Charles Juvin, Jonas Sundermeyer, Guillaume Hekimian, Marvin Kriz, Alexander Bernhardt, Stefan Kluge, Matthieu Schmidt, Pascal Leprince, Guillaume Lebreton, David Hajage, Benedikt Schrage, Alain Combes Published: 2026 ABSTRACT AbstractBackgroundConflicting data exist regarding the survival benefit of mechanical left ventricular (LV) unloading in patients receiving peripheral venoarterial extracorporeal membrane oxygenation (VA-ECMO) for refractory cardiogenic shock (CS). In this study, we aimed to assess the effect of LV unloading (Impella or intra-aortic balloon pump [IABP]) on 60-day mortality in patients with refractory CS treated with VA-ECMO.MethodsWe conducted a retrospective bi-center study including consecutive patients requiring VA-ECMO for ≥ 48 h in two high-volume ECMO centers between 2019 and 2023, using the target trial emulation framework. Patients were classified as receiving VA-ECMO alone or VA-ECMO with LV unloading if the unloading strategy was applied within 24 h after ECMO implantation. Propensity-score overlap weighting was applied to adjust for baseline differences. The primary endpoint was 60-day mortality. Secondary outcomes included ECMO weaning rates and device-related complications.ResultsAmong 264 patients (76% male, mean age 53 ± 14 years), 138 (52.2%) received VA-ECMO alone and 126 (47.8%) underwent LV unloading (78 IABP, 48 Impella), 2 were lost to follow-up. Unloading was not associated with improved 60-day mortality (weighted risk difference: 1.4% [95% CI: -11.8%, 14.6%], p = 0.84). No significant differences were observed in ECMO weaning rates, but device-related complications were more frequent in the unloading group.ConclusionIn this propensity-score weighted analysis, LV unloading during VA-ECMO was not associated with improved survival at 60 days. These hypothesis-generating findings highlight the need for a refined, phenotype-driven approach to patient selection for LV unloading in ECMO-supported cardiogenic shock.Graphical abstract Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s13054-026-06213-4.Keywords: Venoarterial extracorporeal membrane oxygenation (VA-ECMO) , Mechanical left ventricular (LV) unloading, Intra-aortic balloon pump [IABP], Microaxial flow pump, Target trial emulation framework Introduction Cardiogenic shock (CS), the most severe form of heart failure, is marked by profound cardiac output reduction leading to poor perfusion, organ dysfunction, and often multi-organ failure [1]. Despite advances in diagnosis and management, short-term mortality remains high at around 50% over the past two decades [2–5]. Temporary mechanical circulatory support (MCS) devices are increasingly used to address circulatory failure as a bridge to recovery, transplantation, or durable ventricular assist devices [6]. Among these, peripheral veno-arterial extracorporeal membrane oxygenation (VA-ECMO) is frequently employed due to rapid deployment, minimally invasive cannulation, and high-flow support capabilities [7]. However, recent randomized trials have not shown a mortality benefit in unselected refractory cardiogenic shock populations, emphasizing the need for better patient selection and ECMO timing [8–10]. Beyond selection, optimizing management during ECMO support is crucial. VA-ECMO’s retrograde arterial flow can increase left ventricular afterload, exacerbating myocardial strain, hindering ventricular recovery, and causing complications such as pulmonary congestion [11–13]. These adverse effects highlight the need for strategies that enhance myocardial protection and unloading. Recently, several retrospective studies have evaluated combined VA-ECMO and left ventricular unloading, but their findings remain conflicting, warranting further investigation [14–18]. Therefore, the aim of this study was to assess the impact of combining VA-ECMO with left ventricular unloading, using either an intra-aortic balloon pump (IABP) or a microaxial flow pump (Impella®, Johnson & Johnson, Norderstedt, Germany), on 60-day mortality in patients with refractory cardiogenic shock treated at two high-volume ECMO centers in Europe, utilizing a propensity score–weighted analysis emulating a target trial. Methods Patients and design This bi-center, retrospective cohort study was conducted at two European ECMO centers: University Heart and Vascular Center in Hamburg, Germany and the Pitié-Salpêtrière Hospital in Paris, France. The study included all consecutive adult patients (≥ 18 years) hospitalized for cardiogenic shock between January 2019 and January 2023. Clinical and demographic data were collected at baseline, during VA-ECMO support, and at follow-up. Key data points included : baseline characteristics of patients at cannulation (age, sex, body mass index, comorbidities and etiology of cardiogenic shock); severity scores of cardiogenic shock before VA-ECMO (SCAI, SAVE and SAPS II scores) along with hemodynamic and biological parameters ; mechanical circulatory support (MCS) details and unloading strategy (timing and duration of ECMO and unloading device) ; in ICU complications (devices related or not) and outcomes (60-day survival, weaning attempts and bridge strategies). All clinical events and endpoints were recorded from the time of VA-ECMO initiation until hospital discharge or death. MCS management The decision to initiate VA-ECMO or to implement a LV unloading strategy was not standardized or protocolized across centers. Instead, mechanical circulatory support was employed based on conventional clinical indications [3]. VA-ECMO was indicated for patients with refractory cardiogenic shock, defined as persistent hypotension (systolic blood pressure < 90 mmHg or mean arterial pressure < 65 mmHg) despite high-dose vasopressor and inotropic therapy, signs of end-organ hypoperfusion (e.g., oliguria, elevated lactate >2mmol/L), and/or severe left ventricular dysfunction (ejection fraction < 20%) unresponsive to medical management. Left ventricular unloading devices were implemented at the discretion of the treating physician but were not used routinely. Instead, unloading was reserved for patients with objective evidence of inadequate left ventricular decompression during VA-ECMO support, including increased pulmonary artery occlusion pressure, pulmonary edema, left ventricular dilation, markedly reduced residual cardiac output (LVOT-VTI < 6–7 cm), or persistent absence of aortic valve opening despite optimization of inotropic therapy [11, 19]. The choice of the unloading device was left to the discretion of the treating physician, in accordance with locally established practices. Transitioning from one device to another was similarly determined at the physician’s discretion. ECMO weaning or bridge to a long term assist device or heart transplantation was not standardized across centers but followed local protocols and general recommendations [20]. Patients were categorized into two groups based on the MCS strategies implemented the day of VA-ECMO initiation: VA-ECMO alone or VA-ECMO plus unloading device (VA-ECMO with IABP), or VA-ECMO with Impella) and were compared from baseline to 60 days after VA-ECMO implantation or death accordingly. Patients who received an unloading device more than 24 h after ECMO initiation were classified in the ECMO-alone group. The 24-hour threshold was selected for two reasons: (1) to avoid immortal time bias associated with defining exposure as unloading at any time during ECMO, consistent with the intention-to-treat framework of target trial emulation [21] and (2) because planned unloading was typically performed shortly before, simultaneously with, or shortly after ECMO initiation, whereas unloading after 24 h usually represented rescue treatment for late LV distension and pulmonary edema. This approach is consistent with Schrage et al. who also distinguished early from delayed unloading and observed the greatest survival benefit with early unloading [22]. This retrospective study was conducted in accordance with the ethical principles of the Declaration of Helsinki and all applicable national and institutional regulations governing retrospective research involving human participants, including data protection and confidentiality requirements. The study was approved by local ethics and institutional review committees (Société de Réanimation de Langue Française ethics committee CE-SRLF-23-047; Ethikkommission der Ärztekammer Hamburg, PV5607). The need for informed consent was waived by the main ethics committee because this was a retrospective analysis and only completely anonymized data were collected and analyzed. Endpoints The primary endpoint of this study was 60-day mortality. Other endpoints included VA-ECMO weaning success rate, ICU length of stay, durations of mechanical ventilation (MV) and VA-ECMO support, as well as device-related complications. These complications, which served as safety endpoints, encompassed MCS site infections, hemolysis, leg ischemia, stroke, and bleeding events. VA-ECMO weaning success was defined as the successful removal of ECMO, without subsequent death, requirement for additional circulatory support, or heart transplantation within 30 days of weaning. Device-related sepsis was defined according to international sepsis criteria, requiring local signs of infection (e.g., abscess, exudate at the cannulation site) and microbiological confirmation through positive local or blood cultures for a pathogenic microorganism. Hemolysis was identified as lactate dehydrogenase (LDH) levels ≥ 1,000 U/L and plasma-free hemoglobin levels > 100 mg/dL in two consecutive samples within 24 h. Leg ischemia was clinically suspected based on the presence of pain, pallor, pulselessness, or paresthesia in the affected limb, with radiological confirmation via computed tomography demonstrating arterial occlusion or significant vascular compromise. Stroke required CT-scan confirmation to be included in the analysis. Bleeding events were classified as moderate or severe based on the GUSTO (Global Utilization of Streptokinase and Tissue Plasminogen Activator for Occluded Arteries) classification [23]. Statistics To mitigate biases inherent to the observational study design, we emulated a target trial [24]. First, we defined a hypothetical randomized trial protocol that would answer our research question. Next, we emulated this target trial step-by-step using retrospective observational data. Details of the target trial design and its emulation are provided in the Supplemental Material (Table S2-S4, Fig. S1). In the target trial, randomization, unloading device (for the combined strategy group) and follow-up would approximately coincide with the time ‘zero’ of VA-ECMO initiation. In the retrospective cohort, left ventricular unloading was initiated simultaneously with VA-ECMO for the majority of patients. To avoid immortal time bias, patients initiating left ventricular unloading > 24 h after VA-ECMO initiation were considered unexposed to the ECMO + unloading strategy (thus only receiving VA-ECMO). Conversely, patients for whom left ventricular unloading was initiated before VA-ECMO were considered exposed to the ECMO + unloading strategy. Our target estimand was an observational “analogue” of an “intention-to-treat” effect. Descriptive analysis of the baseline characteristics of the patients We conducted a descriptive analysis of the initial characteristics of patients according to their baseline exposure i.e. VA-ECMO alone or in combination with an unloading device. A descriptive analysis of post-exposure outcome was also performed. Categorical variables are presented as number with percentage, while continuous variables are presented as median with interquartile range or as mean with standard deviation. Propensity score analysis Our objective was to estimate the effect on all-cause mortality of the strategy combining VA-ECMO and left ventricular unloading compared with VA-ECMO alone. First, we estimated a propensity score for each individual in the study using a logistic regression model. The variables included in this model were pre-specified based on clinical expertise: age, sex, BMI, history of atrial fibrillation, diabetes, hypertension, peripheral arterial disease, ischemic heart disease, valvular heart disease, other chronic diseases, cause of cardiogenic shock, SCAI score, SAVE score, SAPS II score, mechanical ventilation prior to ECMO, Glasgow Coma Scale score prior to ECMO, cardiac arrest before ECMO, myocardial infarction before ECMO, serum lactate, use of vasopressors, and measurement of left ventricular ejection fraction. Then, each participant was weighted using the estimated propensity score. Covariate balance between the two groups was assessed after weighting, and we considered an absolute standardized difference of less than 0.1 as evidence of balance. We first calculated ATE (Average Treatment Effect) weights [25]. This weighting aims to estimate the effect of the ECMO + unloading VA-ECMO and left ventricular unloading strategy if the entire population were to receive it, compared to if the entire population were to receive VA-ECMO only. Because the ATE weighting did not result in a satisfactory balance, we also calculated overlap weights [26]. This weighting estimates the treatment effect of the ECMO + unloading VA-ECMO and left ventricular unloading strategy compared to VA-ECMO only, if applied to the overlap population—the subgroup with the most balanced characteristics between the two treatment groups—. This approach is particularly useful in the presence of extreme propensity scores values. The effect of the ECMO + unloading VA-ECMO and left ventricular unloading strategy versus VA-ECMO alone on all-cause mortality was estimated within the weighted pseudopopulation using risk difference at Day 60. Weighted Kaplan-Meier estimator was used. Estimates were obtained first from complete cases and then by imputing missing data [27, 28]. Five copies of the dataset were created with the missing values replaced by imputed values, on the basis of observed data, including outcomes and baseline characteristics of participants. Each dataset was then analyzed using previously described statistical methods, and the results from each dataset were pooled into a final result using Rubin’s rule. Finally, subgroup analyses were performed, according to the center and the etiology of cardiogenic shock. Results Patients’ characteristics at baseline During the study period, 264 patients (76% male, mean age 53 ± 14 years) received VA-ECMO. Patients’ baseline characteristics according to mechanical circulatory support (MCS) strategies are presented in Table 1. VA-ECMO alone was applied to 138 (52.2%) patients, while an IABP or an Impella was associated with ECMO in 78 (29.5%) and 48 (18.2%) patients, respectively (see Flowchart Fig. 1). Acute myocardial infarction (AMI) was the most frequent etiology of cardiogenic shock, occurring in 118 (44.6%) patients (53 (38.4%) and 65 (51%) patients, in the ECMO alone and ECMO + Unload groups, respectively). Patients in the ECMO + Unload group more frequently presented with AMI (51% vs. 38%) or idiopathic cardiomyopathy (21% vs. 9%), had lower LVEF (10% vs. 15%) and were more often mechanically ventilated before ECMO initiation (58% vs. 43%), whereas cardiac arrest was less common (43% vs. 57%). Severity scores were well-balanced between the groups, with most patients (67%) classified as SCAI Stage D. Cardiac arrest had occurred pre-ECMO in 132 (50%) patients (65 were cannulated while undergoing cardiopulmonary resuscitation CPR). Table 1Baseline characteristics of patients according to their MCS strategyAllN = 264ECMO alone*N = 138ECMO and UnloadN = 126ECMO + IABPN = 78ECMO + ImpellaN = 48 Demographic Age, years53 (14)53 (14)55 (13)49 (12)61 (12)Men199 (75.5%)86 (67.7%)105 (83.3%)71 (84.5%)44 (80%)BMI, kg/m226.7 (5.3)27.0 (5.3)26.5 (5.3)26.3 (5.7)26.6 (4.4) Comorbidities Atrial Fibrillation46 (17.5%)25 (18.1%)21 (16.7%)10 (12.9%)11 (22.9%)Diabetes mellitus55 (20.8%)33 (23.9%)22 (17.5%)13 (16.7%)9 (18.8%)Hypertension76 (28.9%)46 (33.6%)30 (23.8%)9 (11.5%)21 (43.8%)Periph arterial disease10 (3.8%)6 (4.3%)4 (3.2%)1 (1.3%)3 (6.3%)Chronic cardiomyopathy131 (49.6%)77 (55.7%)44 (34.9%)42 (53.3%)12 (25.0%) Ischemic cardiomyopathy44 (16.6%)26 (18.8%)18 (14.3%)11 (14.1%)7 (14.6%) Valvular cardiomyopathy17 (6.4%)10 (7.2%)7 (5.6%)5 (6.4%)2 (4.2%) Other cardiomyopathy70 (26.5%)41 (29.7%)19 (15.2%)26 (33.3%)3 (6.3%) Cause of Cardiogenic shock Acute myocardial infarction112 (42.4%)44 (31.9%)65 (51.0%)34 (43.6%)34 (70.9%) LAD occlusion74 (62.7%)28 (52.8%)46 (70.8%)23 (74.2%)23 (67.6%) Attempted PCI108 (91.5%)46 (86.8%)62 (95.4%)29 (93.5%)33 (97.1%) Successful PCI91 (84.2%)39 (78.2%)52 (83.8%)20 (68.9%)32 (96.9%)Idiopathic dilated cardiomyopathy40 (15.2%)13 (9.4%)27 (21.4%)25 (32.1%)2 (4.2%)Post-cardiotomy19 (7.2%)11 (8.0%)8 (6.3%)6 (7.7%)2 (4.2%)Myocarditis12 (4.5%)8 (5.8%)4 (3.2%)4 (5.1%)0 (0.0%)Other81 (30.7%)62 (44.9%)19 (15.1%)9 (11.5%)10 (20.8%)State of cardiogenic shock Low-output Cardiogenic shock255 (96.6%)129 (93.5%)126 (100%)78 (100%)48 (100%) High-output Cardiogenic shock9 (3.4%)9 (6.5%)0 (0%)0 (0%)0 (0%)SCAI classification C14 (5.3%)7 (5.1%)7 (5.6%)0 (0%)7 (14.6%) D177 (67.0%)89 (64.5%)88 (69.8%)72 (92.3%)16 (33.3%) E73 (27.7%)42 (30.4%)31 (24.6%)6 (7.7%)25 (52.1%)SOFA11 [8 ; 14]12 [8 ; 13]11 [8 ; 14]11 [8 ; 14]10 [10 ; 10]SAVE Score-9 [-11 ; -5]-9 [-12 ; -6]-9 [-11 ; -5]-9 [-11 ; -4]-9 [-11 ; -6]SAPS II62 [41.25 ; 77]68 [45 ; 80.5]56 [38 ; 72]51 [35.75 ; 65]65 [53 ; 80.5]Pre ECMO MV134 (50.8%)80 (58.0%)54 (42.9%)16 (20.5%)38 (79.2%)ECMO Retrieval62 (23.5%)34 (24.6%)28 (22.2%)22 (28.2%)6 (12.5%)Pre ECMO cardiac arrest132 (50.0%)78 (56.5%)54 (42.9%)22 (28.2%)32 (66.7%) No-flow0 [0 ; 0]0 [0 ; 0]0 [0 ; 0]0 [0 ; 0]0 [0 ; 0] Low-flow29 [15 ; 50]30 [17 ; 51]25 [13.5 ; 50]15 [10 ; 25]47 [20 ; 60] Shockable rythm61 (52.1%)30 (44.1%)31 (57.4%)10 (50.0%)21 (72.4%) ECMO under CPR65 (49.2%)38 (48.7%)27 (50.0%)5 (22.7%)22 (68.8%) Pre ECMO biology pH7.25 [7.12 ; 7.3]7.23 [7.08 ; 7.31]7.27 [7.16 ; 7.38]7.30 [7.21 ; 7.43]7.20 [7.08 ; 7.28]Lactate, mmol/L6.7 [3.6 ; 11.1]7.7 [4.18 ; 12.12]5.7 [3.3 ; 10]4.7 [3.05 ; 8.05]8.95 [5 ; 13.4]Troponin (ng/L)1656 [226 ; 21666]2134 [297 ; 21776]1239 [181 ; 21338]479 [82 ; 4465]35,595 [3438; 163932]Creatinine, µmol/L126 [94 ; 174]126 [96 ; 200]211 [82.5 ; 595.5]124 [93 ; 160]130 [101 ; 167]ASAT (UI)201 [60 ; 630]167 [51 ; 637]127 [94 ; 162]211 [72 ; 577]203 [84 ; 812]Prothrombin activity (%)62 [49 ; 72]62.5 [50 ; 72]62 [48.75 ; 75.25]62 [48 ; 75]62 [62 ; 62]aPTT (sec)52 [40 ; 107]51 [37 ; 117]54 [42 ; 92]48 [41 ; 61.5]103 [46.75 ; 160]PaO2/FiO2342 [162 ; 496]303 [152 ; 434]398 [202.5 ; 536.5]428 [232 ; 555]338 [161 ; 457] Pre ECMO hemodynamics SAP, mmHg94 [84 ; 107]95 [83 ; 111]94 [84 ; 104.25]95 [87 ; 104]86 [70 ; 102]DAP, mmHg63 [53 ; 73]66 [54 ; 74]61 [53 ; 71.25]63 [56 ; 72]54 [44 ; 60]MAP, mmHg73 [65 ; 82]75 [66 ; 83]73 [65 ; 81]73 [67 ; 81]63 [56 ; 74]HR, bpm111 [90 ; 124]112.5 [90 ; 127]110 [89 ; 122]111 [94 ; 124]89 [75 ; 120]Vasopressor use254 (96.2%)124 (98.4%)118 (94.2%)80 (95.4%)50 (92.9%)Vasoactive inotropic score88 [32 ; 146]93 [55 ; 169]72 [24 ; 134]70 [24 ; 130]390 [390 ; 390]Epinephrine use51 (19.3%)37 (26.8%)14 (11.1%)3 (3.8%)11 (22.9%)Norepinephrine use228 (86.3%)114 (82.6%)114 (90.4%)65 (83.3%)48 (100.0%)Dobutamine use137 (51.9%)57 (41.3%)80 (31.5%)71 (91.0%)9 (18.7%) Pre ECMO Echo Data LVEF, %10 [5 ; 20]15 [10 ; 25]10 [5 ; 15]10 [10 ; 20]5 [5 ; 10]Aortic VTI, cm7 [6 ; 10]8 [7 ; 10]7 [6 ; 8]7 [5.75 ; 8]6 [6 ; 6]Grade > II mitral regurgitation44 (17.0%)18 (13.2%)29 (23.0%)18 (23.1%)8 (17.8%)Values are expressed as n (%), median [IQR] or mean (± sd)* Patients who received Unloading device > 24 h (N = 13) after ECMO initiation were considered in the ECMO alone group (6 patients received IABP > 24 h after ECMO insertion and 7 patients received Impella > 24 h after ECMO insertion)Inotrope score (µg/kg per min) = dobutamine dose (µg/kg per min) + 100x(epinephrine dose [µg/kg per min] + norepinephrine dose [µg/kg per min])MCS mechanical circulatory support, ECMO Extracorporeal membrane oxygenation, BMI body mass index, LAD left anterior descending artery, PCI percutaneous coronary intervention, MV mechanical ventilation, CPR cardiopulmonary resuscitation, SAP systolic arterial pressure, DAP diastolic arterial pressure, MAP mean arterial pressure, HR heart rate, LVEF left ventricular ejection fraction, VTI velocity-time integral, IABP intra-aortic balloon pump Baseline characteristics of patients according to their MCS strategy Values are expressed as n (%), median [IQR] or mean (± sd) * Patients who received Unloading device > 24 h (N = 13) after ECMO initiation were considered in the ECMO alone group (6 patients received IABP > 24 h after ECMO insertion and 7 patients received Impella > 24 h after ECMO insertion) Inotrope score (µg/kg per min) = dobutamine dose (µg/kg per min) + 100x(epinephrine dose [µg/kg per min] + norepinephrine dose [µg/kg per min]) MCS mechanical circulatory support, ECMO Extracorporeal membrane oxygenation, BMI body mass index, LAD left anterior descending artery, PCI percutaneous coronary intervention, MV mechanical ventilation, CPR cardiopulmonary resuscitation, SAP systolic arterial pressure, DAP diastolic arterial pressure, MAP mean arterial pressure, HR heart rate, LVEF left ventricular ejection fraction, VTI velocity-time integral, IABP intra-aortic balloon pump Flow chart of the study population. * 11 patients in the ECMO and mechanical unloading received the unloading device before ECMO (7 Impella and 4 IABP). IABP intra-aortic balloon pump The unloading strategy was an IABP in Paris and an Impella in Hamburg in 99% and in 100% of the patients, respectively. Patients included in Hambourg were older, more frequently in SCAI Stage E class, had more frequent AMI as the initial cause of CS and had more cardiac arrest pre-ECMO (Table S1). Association between LV unloading and 60-day mortality Thirteen patients received their unloading device more than 24 h after ECMO initiation and were reclassified into the ECMO alone group per the pre-specified ITT classification (7 Impella, 6 IABP), while 11 had received their device prior to ECMO initiation (7 Impella, 4 IABP) and were retained in the ECMO + Unloading group. Two patients in the ECMO alone group were lost to follow-up and were excluded from the survival analysis. Before weighting and imputation of missing data, analysis showed no difference in Day 60 mortality between the VA-ECMO alone group and the ECMO + Unloading group (53,6% vs 53,2%, Risk Difference [RD]: -0.4%, 95%CI [-25.0%, 17.8%]) (Tables 2 and 3; Fig. 2). Table 2Patient outcomes during ICU stay according to their MCS strategyAll PatientsN = 264ECMO Alone*N = 138ECMO and UnloadN = 12643 (16.3%)15 (10.9%)28 (22.2%)Device related sepsis ECMO site infection42 (15.9%)15 (10.9%)27 (21.4%) IABP site infection0 (0%)0 (0%)0 (0%) Impella site infection1 (1.4%)0 (0%)1 (2.1%)Hemolysis79 (30.0%)39 (28.5%)40 (31.7%)Leg ischemia42 (15.9%)25 (18.1%)17 (13.5%)Ischemic stroke25 (9.5%)11 (8.0%)14 (11.1%)Hemorrhagic stroke11 (4.2%)4 (2.9%)7 (5.6%)Thrombocytopenia102 (38.6%)49 (35.5%)53 (42.1%)Moderate bleeding38 (14.4%)18 (13.0%)20 (15.9%)Severe bleeding81 (30.7%)35 (25.4%)46 (36.5%)Renal replacement therapy127 (48.1%)62 (44.9%)65 (51.6%)Pulmonary edema on ECMO115 (43.7%)49 (35.8%)66 (52.4%)Ventilator associated pneumonia126 (47.7%)67 (48.6%)59 (46.8%)Atrial septostomy0 (0%)0 (0%)0 (0%)IABP to impella5 (1.9%)0 (0.0%)5 (3.9%)LVAD/BiVAD13 (4.9%)5 (3.6%)8 (6.3%)Heart transplant33 (12.5%)13 (9.4%)20 (15.9%)ECMO successfully weaned109 (41.3%)56 (40.6%)53 (42.1%)Mechanical ventilation duration, days6 [2 ; 15]6 [2 ; 17]8 [3 ; 17]Time on ECMO, days6 [3 ; 10]5 [3 ; 9]6 [4 ; 12]Time on IABP, days4 [0 ; 8]0 [0 ; 0]7 [4 ; 12]Time on impella, days5 [3.5 ; 11.5]0 [0 ; 0]5 [3.25 ; 11]ICU stay, days13 [6 ; 28]12 [5 ; 25]15 [7 ; 35]Hospitalization stay, days19 [9 ; 41]18 [7 ; 35]22 [9 ; 48]60 days survival status1,2 Alive123 (46.6%)54 (45.6%)66 (47.4%) Alive with MCS11 (4.2%)6 (4.3%)5 (4.0%) Alive with transplant27 (10.2%)11 (8%)16 (12.7%) Alive without MCS or transplant85 (32.2%)47 (34.1%)38 (30.2%) Dead141 (53.4%)74 (53.6%)67 (53.2%)Values are expressed as n (%) or median [IQR]* Patients who received Unloading device > 24 h (N = 13) after ECMO initiation were considered in the ECMO alone group (6 patients received IABP > 24 h after ECMO insertion and 7 patients received Impella > 24 h after ECMO insertion)1 After ECMO initiation.2 Two patients in the ECMO alone group were lost to follow upMCS mechanical circulatory support, ECMO Extracorporeal membrane oxygenation, IABP intra-aortic balloon pump, LVAD left ventricular assist device, BiVAD biventricular assist device, ICU intensive care unit Patient outcomes during ICU stay according to their MCS strategy Values are expressed as n (%) or median [IQR] * Patients who received Unloading device > 24 h (N = 13) after ECMO initiation were considered in the ECMO alone group (6 patients received IABP > 24 h after ECMO insertion and 7 patients received Impella > 24 h after ECMO insertion) 1 After ECMO initiation.2 Two patients in the ECMO alone group were lost to follow up MCS mechanical circulatory support, ECMO Extracorporeal membrane oxygenation, IABP intra-aortic balloon pump, LVAD left ventricular assist device, BiVAD biventricular assist device, ICU intensive care unit Table 3Survival status 60 days after ECMO initiation before and after weighting: ATO weightsVariableUnweighted sampleWeighted sampleECMO,N = 1381ECMO and unload,N = 126ECMO,N = 43ECMO and unload,N = 43Alive64 (46.4%)59 (46.8%)22 (50.1%)21 (48.7%)Dead74 (53.6%)67 (53.2%)21 (49.9%)22 (51.3%)Values are expressed as n (%)1 Includes 13 patients with Unloading support insertion > 24 h after ECMO initiation (7 Impella and 6 IABP), 2 patients lost to follow-up Survival status 60 days after ECMO initiation before and after weighting: ATO weights Values are expressed as n (%) 1 Includes 13 patients with Unloading support insertion > 24 h after ECMO initiation (7 Impella and 6 IABP), 2 patients lost to follow-up Kaplan-Meier 60 days survival curves before (A) and after (B) weighting on propensity score - ATO weigths Propensity score weighting with ATE weights improved overall balance of baseline covariates, although some covariates remained suboptimally balanced between groups (Table S2 and S3, Fig. S1). In this weighted sample, 60-day mortality did not significantly differ between the ECMO alone group (51.4%) and ECMO + Unload (47.8%) groups (RD -3.6%, 95% CI [-25%, 17.7%, p = 0.74) (Fig. S2, Table S3). After multiple imputation of missing data, RD was − 6.4% (CI95%:[-29.2%, 16.3%], p = 0.58). Due to the non-optimal balance between groups, weighting using ATO weights was also performed. Balance was optimal in the resulting weighted sample for all characteristics included in the analysis (Fig. S3). In this weighted sample, 60-day mortality did not significantly differ between the ECMO alone group (49.9%) and ECMO + Unload (51.3%) groups (RD 1.4%, 95% CI [-11.8%, 14.6%], p = 0.84) (Fig. 2, Table S3). After multiple imputation of missing data, RD was 2.2% (CI95%: [-10%, 14.5%], p = 0.72). In a sensitivity analysis excluding the 11 patients who received unloading prior to ECMO initiation, the 60-day risk difference remained non-significant, with both ATE weighting (RD = -11.1%, 95% CI [-25.0%, 17.7%], p = 0.53) and ATO weighting (RD = 2.5%, 95% CI [-11.1%, 16.0%], p = 0.72). Day 60 mortality did not differ in sensitivity analyses restricted to Hamburg both ATE weighting (RD = 15.0%, 95% CI [-25.1%, 55.5%], p = 0.46) and ATO weighting (RD = 6.1%, 95% CI [-24.0%, 11.2%], p = 0.50) and Paris both ATE weighting (RD = 10.1%, 95% CI [-13.4%, 34.2%], p = 0.41) and ATO weighting (RD = -6.8%, 95% CI [-11.7%, 25.3%], p = 0.47) patients and to myocardial infarction and idiopathic dilated cardiomyopathy patients. Safety endpoints and ICU outcomes ICU outcomes according to the MCS strategy are summarized in Table 2 (device-stratified outcomes in Table S4). Overall, device-related sepsis occurred in 43 patients (16%), more frequently in the ECMO + Unload group than in the VA-ECMO alone group (28 [22.2%] vs. 15 [10.9%]), and was almost exclusively related to the ECMO cannulation site. The incidence of hemolysis, leg ischemia, stroke, bleeding, and ventilator-associated pneumonia was otherwise comparable between groups. Pulmonary edema while on ECMO was observed in 115 patients (43.7%) overall, with a higher occurrence in the ECMO + Unload group (54% vs. 33%). When stratified by device type, hemolysis, leg ischemia, stroke, and bleeding were predominantly observed in the Impella subgroup, while device-related sepsis was more frequent in the IABP subgroup, likely reflecting the longer duration of VA-ECMO support in patients managed with this strategy. In the overall cohort, 13 (5%) patients were transitioned from short-term to long-term MCS, including 5 (3.6%) patients in the VA-ECMO alone group and 8 (6.3%) patients in the ECMO + Unload group and 33 (12.5%) patients were bridged to heart transplantation. The median duration of VA-ECMO support was 6 [3–10] days. Median ICU length of stay was 13 [6–28] days, shorter in the VA-ECMO alone group (12 [5–25] vs. 15 [7–35] days, respectively). Discussion In this multicenter, propensity-score weighted analysis emulating a target trial, LV unloading in patients supported by VA-ECMO was not associated with a significantly lower mortality at day 60. The rationale for left ventricular (LV) unloading during peripheral VA-ECMO is based on its potential to mitigate ECMO-induced increases in LV afterload, thereby reducing LV distension, elevated pulmonary capillary wedge pressure, and pulmonary edema [12, 29]. In the largest retrospective multicenter cohort to date, based on the ELSO registry, Grandin et al. reported improved survival associated with LV mechanical unloading in over 12,000 patients [15], despite a higher incidence of complications such as hemolysis and cannulation site bleeding. Compared to microaxial pumps, unloading with an IABP was associated with similar mortality but lower complication rates. Similarly, Schrage et al. observed a survival benefit with the addition of Impella to VA-ECMO compared to VA-ECMO alone, with sensitivity analyses suggesting that this effect was largely driven by patients who received Impella before or shortly after ECMO initiation [22]. Similarly, in the study by Grandin et al. [15], the unloading device was implanted prior to ECMO initiation in 86% of patients. Although these and other observational studies have suggested a potential survival advantage with LV unloading, whether through Impella or IABP, these findings are inherently limited by confounding [14, 15, 30, 31]. Significant selection bias related to indication, as well as heterogeneity in shock etiology and variability in device timing and choice, complicates the interpretation of these results. In our cohort, the adjusted 60-day mortality was approximately 50% in both groups, consistent with prior reports [32]. However, our neutral findings stand in contrast to the pathophysiological rationale advocating for left ventricular unloading to mitigate ECMO-associated increases in afterload, as well as to recent data from large registries of VA-ECMO patients, where unloading was applied without standardized hemodynamic criteria, clearly defined patient selection, or consistent device strategies across heterogeneous populations. Emerging data further suggest that unloading may not be universally beneficial, particularly when applied without standardized protocols in diverse ECMO populations [7]. To date, two randomized controlled trials have compared VA-ECMO alone to a combined strategy involving direct left atrial venting via an additional cannula [33, 34]. Consistent with our findings, neither trial demonstrated a survival benefit or improvement in ECMO weaning. However, both studies included a limited number of patients, applied systematic unloading across heterogeneous populations, and experienced a high cross-over rate to LV unloading in the control group (approximately 50%), thereby limiting the ability to detect a mortality benefit. In a meta-analysis of over 4,000 patients, Vallabhajosyula et al. found no overall survival benefit of the IABP when added to ECMO [16], although the combined strategy was associated with improved outcomes in the subgroup of patients with post-myocardial infarction (MI) cardiogenic shock. Indeed, the systematic increase in LV end-diastolic pressure and volume under VA-ECMO has recently been challenged. In a prospective physiological study of 80 patients with refractory cardiogenic shock supported by peripheral VA-ECMO, Saura et al. showed that increasing ECMO flow from 2 to 4 L/min led to a ≥ 20% reduction in pulmonary capillary wedge pressure (PCWP) in 36% of patients, with no change in 58%, and an increase in only 6%. These findings confirm prior observations by Kalra et al. [35, 36]., suggesting a paradoxical unloading effect of ECMO, likely mediated by right-sided preload reduction and subsequent lowering of LV filling pressures. Collectively, these data challenge traditional assumptions regarding cardiac physiology under VA-ECMO and underscore the need to re-evaluate current approaches to LV unloading. Indiscriminate use of unloading devices may expose patients to unnecessary device-related complications, potentially offsetting the intended benefits of the ECMO + unloading strategy. Consistent with prior reports, we observed higher rates of hemolysis, limb ischemia, ischemic stroke, and bleeding in the ECMO + unloading group, predominantly driven by the Impella subgroup [14, 15]. In contrast, the IABP subgroup experienced more device-related sepsis, largely attributable to prolonged ECMO duration. While grouping IABP and Impella under a single “unloading” category may have diluted some of these differences, subgroup analysis suggests distinct risk profiles associated with each device. Besides, this additional risk burden, unevenly distributed between unloading modalities, should be carefully considered when selecting a venting approach. While Impella provides more robust LV unloading and may allow for earlier ECMO weaning in some patients, it is consistently associated with a higher incidence of device-related adverse events, particularly hemolysis and acute kidney injury [15, 37]. In contrast, the IABP provides a more modest reduction in left ventricular pressures and volumes but appears to offer a more favorable safety profile during prolonged use [19]. However, these findings remain observational and are subject to residual confounding. To address this limitation, several randomized controlled trials are currently underway to assess the impact of left ventricular unloading during VA-ECMO. These include trials evaluating Impella-assisted unloading (UNLOAD ECMO, NCT05577195) and IABP-supported strategies (EULODIA, NCT07027202; EASE-ECMO, NCT06872021; REMAP-ECMO, NCT05913622), each comparing combined support to conventional VA-ECMO alone. The results of these studies are anticipated to provide high-quality evidence to inform clinical decision-making. Our study has several limitations. First, despite rigorous propensity score-based weighting methods, residual unmeasured confounding is likely given the observational design. Second, patients were classified as “ECMO + Unloading” if they received an unloading device within 24 h of ECMO initiation (including before ECMO), and as “ECMO alone” otherwise, consistent with an intention-to-treat approach within the target trial emulation framework. Reclassifying patients who underwent delayed unloading (> 24 h) according to the therapy they ultimately received would have introduced immortal time bias. Nonetheless, this inevitably created heterogeneity within the " ECMO + unloading strategy” group, encompassing patients unloaded before, simultaneously with, or shortly after ECMO initiation. A sensitivity analysis excluding the 11 patients unloaded prior to ECMO confirmed robustness of the primary neutral result, but the observational design precludes reliable identification of the patients most likely to benefit from unloading or determination of the optimal timing of support, questions that will require prospective randomized trials with pre-specified timing strata. Third, grouping IABP and Impella under a unified “LV unloading” strategy assumes a class effect despite major mechanistic differences between devices. Because device selection was almost entirely center-dependent, center and device effects are inseparable in this dataset, and the pooled analysis may mask divergent signals between modalities. Patients treated with ECMO plus Impella had greater illness severity than those treated with ECMO plus IABP, and bridging strategies were unevenly distributed across device groups, likely reflecting center-specific expertise and differential patient selection. The neutral finding was nonetheless consistent across both centers. Importantly, the wide confidence intervals around the primary estimate (risk difference approximately − 12% to + 15%) do not exclude clinically meaningful benefit in selected patient phenotypes, such as those with severe LV distension, absent aortic valve opening, or very low aortic VTI. Accordingly, our findings should be interpreted as an absence of evidence supporting routine, non-selective unloading rather than evidence that unloading is ineffective in all patients. Fourth, both centers’ substantial experience with LV unloading may have introduced confounding by indication. Consistent with the pre-specified clinical criteria used to guide unloading (LV dilation, aortic VTI < 6 cm, pulmonary edema, or absent aortic valve opening despite inotropic therapy), patients receiving ECMO + Unloading generally presented with more severe LV dysfunction, including lower baseline LVEF and aortic VTI. The absence of standardized documentation of each unloading trigger further precludes exploration of whether specific hemodynamic phenotypes preferentially benefit, leaving uncertain whether the neutral result reflects a true absence of benefit or dilution by confounding by indication. Fifth, overlap (ATO) weighting targets patients with intermediate probabilities of receiving unloading, which may not correspond to a clearly identifiable clinical population, and the resulting effective sample size is substantially smaller than the original cohort, limiting the generalizability of these findings. Accordingly, the present study estimates the effectiveness of an early unloading strategy rather than the efficacy of a specific unloading device or of delayed rescue unloading. Conclusion In this propensity score–weighted analysis of patients with refractory cardiogenic shock supported with VA-ECMO, the addition of left ventricular unloading was not associated with a survival benefit at 60 days, despite an increased incidence of device-related complications. These hypothesis-generating findings suggest that a non-selective approach to LV unloading may expose patients to unnecessary device-related risks without necessarily improving outcomes. However, given the wide confidence intervals, the observational design, the inseparability of device and center effects, and the significant heterogeneity within the exposure definition, clinically meaningful benefits in carefully selected patient phenotypes cannot be excluded. The results of ongoing randomized controlled trials (UNLOAD-ECMO, EULODIA, EASE-ECMO, REMAP-ECMO) are eagerly awaited to provide high-quality evidence on the optimal indications, timing, and modality of LV unloading in VA-ECMO-supported cardiogenic shock. Take home message In this propensity score–weighted, bi-center analysis of 266 patients with refractory cardiogenic shock supported with VA-ECMO, early left ventricular unloading (Impella or IABP) was not associated with improved 60-day survival, while device-related complications were more frequent. These results emphasize the need for refined patient selection and randomized controlled trials to define optimal unloading strategies. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1. Acknowledgements None. Abbreviations Acute myocardial infarction Average treatment effect Average treatment effect in the overlap Cardiogenic shock Intra-aortic balloon pump Intensive care unit Left ventricle Left ventricular assist device Mechanical circulatory support Veno-arterial extracorporeal membrane oxygenation Author contributions Author’s contributions: AD, OS, BS, AC wrote the manuscript. All authors contributed to the revision, read and approved the final version of the manuscript. AD, OS, BS, AC take responsibility for the integrity of the work, from inception to published article. Funding None. Data availability No datasets were generated or analysed during the current study. Declarations Ethics approval and consent to participate This retrospective study was conducted in accordance with the ethical principles of the Declaration of Helsinki and all applicable national and institutional regulations governing retrospective research involving human participants, including data protection and confidentiality requirements. It was approved by local ethics and institutional review committees (Société de Réanimation de Langue Française ethics committee CE-SRLF-23-047; Ethikkommission der Ärztekammer Hamburg, PV5607). Consent for publication Not applicable. Competing interests AC reports grants from Getinge and personal fees from Getinge, and Vantive outside the submitted work. MS reports lecture fees from Getinge, Dräger, Baxter, and Fresenius Medical Care outside the submitted work. MS received grants from the French Ministry of Health, outside the submitted work. ACM reports Lecture and consulting fees from Abbott, Alliance BMS-Pfizer, Bayer, Boehringer Ingelheim, Carmat, Novartis, Sanofi. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Angela Dettling, Ouriel Saura, Benedikt Schrage and Alain Combes contributed equally to this work. References Associated Data Supplementary Materials Supplementary Material 1. Data Availability Statement No datasets were generated or analysed during the current study.