We retrospectively analyzed data from all consecutive patients supported with Impella 2.5/CP or vaECMO for CS in our institution, a European tertiary University Hospital, from September 2014 to September 2019. Patients with refractory cardiac arrest in whom insertion of the device took place under ongoing cardiopulmonary resuscitation were excluded from the analysis. CS was defined as the need for continuous infusion of catecholamines to maintain systolic blood pressure > 90 mmHg with clinical signs of pulmonary congestion and signs of end-organ hypoperfusion indicated by an elevated lactate level > 2  mmol/L. Operators used MCS in patients with CS, according to our institutional practices. However, as there is insufficient evidence for the choice of the circulatory support device type in CS and neither the superiority of one device over the other was proven in the setting of CS, the decision to implant an Impella or vaECMO was based on operator’s clinical evaluation (which was based on individual patients’ parameters, including severity of shock, prior resuscitation, bleeding risk, isolated left ventricular failure, biventricular failure, combined cardiorespiratory failure). In brief, the Impella device is our first choice of MCS in patients with CS due to isolated left ventricular (LV) failure. In cases with biventricular failure or cases with combined LV and pulmonary failure we implant a vaECMO first. If vaECMO leads to LV distension and worsening pulmonary edema, an Impella device for venting is additionally inserted. The development of respiratory failure, right heart failure, hemodynamic deterioration, progressive multiorgan failure while on Impella support is, according to our institutional algorithm, an indication for additional implantation of vaECMO. Besides mechanical circulatory support, all patients received standardized medical care and management in accordance with guidelines. If patients received ultimately both devices, they were analyzed according to the device first implanted.

The investigation conforms with the principles outlined in the Declaration of Helsinki. The study was approved by the local ethics committee of the Philipps University of Marburg. The need for informed consent was waived due to the retrospective nature of the study.

All MCS devices were implanted percutaneously in the catheterization laboratory by experienced operators. The Impella pump (Abiomed) was implanted through the femoral artery and placed via the retrograde approach through the aortic valve into the left ventricle under fluoroscopic control. The vaECMO (Maquet Getinge Group) was inserted percutaneously using arterial (17F) and venous (21F for female and 23F for male) femoral cannulas with an additional antegrade femoral limb perfusion cannula. All cardiac arrest patients were treated with targeted temperature management (mild hypothermia of 34 °C) for 24 h with an endovascular cooling device (Thermogard Temperature Management System, Zoll). Inotropes and vasopressors were used to obtain a mean arterial pressure ≥ 65 mmHg. Circulatory support flow was adjusted to maintain mean arterial pressure ≥ 65 mmHg with the lowest possible dose of catecholamines and to cover metabolic needs as assessed by central venous oxygen saturation (≥ 70%) and serum lactate levels (< 2.0 mmol/L). The decision to wean the circulatory support device was based on resolution of shock and clinical assessment. Weaning process was performed by gradually decreasing support. Once the support of the device was reduced to low levels (for Impella performance level 1 and for vaECMO < 1.5 L/min) with stable mean arterial pressure ≥65 mmHg, no or low doses of catecholamines, central venous oxygen saturation ≥7 0% and serum lactate levels < 2.0 mmol/L the device was removed in ICU and hemostasis was achieved with mechanical compression (St. Jude Medical FemoStop).

Intrahospital data, outcomes and follow-up data were collected from the medical charts. Our primary outcomes were survival to hospital discharge and at 6-month. Secondary endpoints were complications. Complications included device-related vascular complications (bleeding at access-site requiring transfusion, limb ischemia requiring removal of the device, surgery or interventional repair), myocardial reinfarction, stroke and other non-device-related bleeding. Access-site bleeding requiring transfusion was defined as bleeding at the cannulation site with need of transfusion of at least two red blood cell (RBC) units. Cerebral functional status in cardiac arrest patients was determined according to the Pittsburgh cerebral performance category (CPC) based on medical records or discharge summary abstracts.

All data were analyzed retrospectively. Data are presented as absolute variables and percentages (%) for categorical variables and either median with interquartile range [IQR 25th–75th percentile] or mean with standard deviation according to the distribution of the variables. We assessed normality using Shapiro–Wilk test as well as Pearson Tests. After testing for normal distribution, Student’s t test or Mann–Whitney test was implemented to test for differences between the various characteristics. For categorical variables Fisher’s exact test or Chi-square test were used, as appropriate. Interactions between nominal variables were measured with lambda coefficient. Patients at risk were assessed with the log-rank test of the survival analysis. The variables were dichotomized according to median in overall population, when they were not linearly distributed. An initial analysis was performed in order to identify the variables associated with outcome mortality in the overall population. A separate analysis was performed in order to identify the variables with a different distribution among the groups of devices. All these variables are presented in the Table 1. Age, Charlson Comorbidity Index (CCI), vasoactive score, creatinine, GFR, pH, PaO₂/FiO_2, etiology of cardiogenic shock and prior CPR were included in the model as significantly associated with outcome in univariate analysis or as clinically meaningful. Propensity score matching was used to balance observed covariates in treatment groups [4, 5]. In this study, the propensity score was the conditional probability for getting vaECMO for CS, as a binary dependent variable, under a set of measurements. Age, CCI, vasoactive score, creatinine, pH, etiology of shock, PaO₂/FiO₂ and prior CPR were added into a multivariable logistic regression model. The predicted probability derived from the logistic equation was used as the propensity score for each individual. The model revealed a c-statistic of 0.88 (C.I. 0.81–0.95). Then we performed a conditional logistic regression after matching on the propensity score in a 1:1 in order to identify the matched pairs. Given the frequency of the outcome in total population, our power calculation showed that we would need a minimum of 75 pairs in order to establish our superiority or non-inferiority hypothesis with a non-inferiority or superiority margin of − 0.1 with a power of 0.8 and type α error of 0.05. All analyses were considered statistically significant for p < 0.05. All analyses were two-sided. Statistical analysis was performed using SPSS 24 and Graphpad Prim 6.0.

From September 2014 to September 2019, a total of 423 patients with CS were treated with MCS. Among them, 300 patients (71%) underwent Impella implantation and 123 patients (29%) were supported with vaECMO. Within the Impella group, 227 patients (75.7%) were supported with an Impella 2.5 and 73 (24.3%) patients with Impella CP. Forty-two (9.9%) patients were ultimately supported by both devices simultaneously (20 patients Impella first and 22 patients vaECMO first).

Demographic and baseline characteristics of patients are presented in Table 1. On admission, the groups were similar regarding gender distribution, cardiovascular risk factors, etiology of CS, mean blood pressure, LVEF, inotropic/vasopressor therapy and creatinine values. Compared to patients in the Impella group, patients in the vaECMO group were younger (68.96 ± 11.56 versus 61.25 ± 10.4, p < 0.001) and presented more often with cardiac arrest (58.9% versus 39%, p = 0.0004) (Table 1). Also, vaECMO supported patients had significantly worse pH (7.27 ± 0.17 versus 7.21 ± 0.19, p = 0.001) and significantly higher lactate levels (6.5 ± 4.7 versus 9.18 ± 5.63 mmol/l, p < 0.001). Using propensity score, 83 pairs of patients were matched. The characteristics of the propensity-matched cohort were well balanced and evenly distributed regarding covariates (Table 1). The procedural characteristics of the vaECMO and Impella supported patients (overall and matched cohort) are displayed in Table 2.

Information on hospital and 6-month mortality were available for all patients. Hospital survival rates were 47.7% in the Impella group and 37.3% in the vaECMO group (p = 0.07) (Table 3). Six-month survival rates were 45.7% and 35.8% (p = 0.08) for the Impella and vaECMO group, respectively. In the matched cohort, the hospital and 6-month survival rates were also comparable for the Impella and vaECMO group (hospital survival: 50.6% versus 38.6%, p = 0.16 and 6-month survival: 45.8% versus 38.6%, p = 0.43) (Fig. 1 and Table 3). Furthermore, survival rates were comparable across all tested subgroups except for patients with prior cardiac arrest where Impella support improved the primary endpoint of hospital survival (Fig. 2). Patients with pre-percutaneous coronary intervention (PCI) Impella (n = 94) had better survival rates than patients with post-PCI Impella (n = 151) implantation (hospital survival and 6-month survival: 63.8% versus 51%, p = 0.03).

Regarding complications, similar rates of myocardial reinfarction (Impella: 1.3% versus vaECMO: 1.6%, p = 0.1) and stroke (Impella: 1.7% versus vaECMO: 2.4%, p = 0.7) were observed (Table 3). Device-related access-site vascular complications occurred more frequently in the vaECMO group than in the Impella group. Access-site bleedings rates requiring transfusion were significantly higher within the vaECMO group (17% versus 7.3%, p = 0.004). Vascular access-site ischemic complications requiring removal of the device, surgery or percutaneous intervention occurred in 17% and 7.7% (p = 0.008) of patients in the vaECMO and Impella group, respectively (Table 3).