Can levosimendan reduce ECMO weaning failure in cardiogenic shock?: a cohort study with propensity score analysis

We conducted a retrospective observational study between January 2012 and December 2018 at Louis Pradel University Hospital (Hospices Civils de Lyon, France). The study protocol was approved by our Institutional Review Board (N°20-54; Chair: Prof. JF Guerin) and registered with ClinicalTrial.​gov (NCT04323709). Given the retrospective and non-interventional design of the study, the need for written informed consent was waived. All consecutive adult patients admitted to the cardiothoracic intensive care unit (ICU) who underwent VA-ECMO for refractory cardiogenic shock were eligible for the study. Exclusion criteria were age < 18 years, VA-ECMO duration < 48 h, VA-ECMO for refractory cardiac arrest, right heart or veno-venous ECMO, and VA-ECMO for circulatory failure following lung transplantation.

The following data were collected at the admission: age, gender, body mass index, Simplified Acute Physiology Score (SAPS-II), Sequential Organ Failure Assessment (SOFA), hypertension, diabetes, hypercholesterolemia, smoking status, history of stroke or congestive heart failure, coronary or peripheral artery disease, renal failure with dialysis, left ventricular ejection fraction (LVEF), tricuspid annular plane systolic excursion (TAPSE), mean arterial pressure, heart rate, central venous pressure, ScvO₂, presence of an intra-aortic balloon pump, and biochemical parameters. During the hospitalization, the following variables were collected: the reason for initiation of VA-ECMO and VA-ECMO characteristics (duration, type, flow (L/min), RPM, FiO₂), length of stay in ICU, catecholamines and inotrope maximal doses and durations of administration, and patients receiving heart transplantation or left ventricular assist device (LVAD). In patients receiving levosimendan, the timing of administration regarding VA-ECMO initiation was also collected.

During the study period, all patients were managed according to international guidelines for cardiogenic shock [14]. Timings of administration of levosimendan (Zimino®, Orion Pharma, Issy-les-Moulineaux, France) and catecholamines were at the entire discretion of the physicians. The administration of levosimendan was started at a dose of 0.1 μg/kg/min for 1 h, followed by a continuous infusion of 0.1 to 0.2 μg/kg/min for 24 h. VA-ECMO flow rate was initially set at the theoretical cardiac output owing to the body surface area of the patient (2.2 L/min/m²). Inotropic support was usually provided in order to maintain both a left ventricular ejection and an aortic valve opening. Anticoagulation with unfractionated heparin was used to maintain anti-Xa factor activity between 0.30 and 0.35 IU/ml during mechanical support. Serial transesophageal echocardiography was performed after a progressive reduction of VA-ECMO flow to a minimum of 1.0–1.5 L/min to assess myocardial recovery. When the weaning trial was hemodynamically well tolerated without the need for increasing inotropic or vasoactive support and echocardiographic criteria were fulfilled (LVEF > 20–25%, time-velocity integral > 10 cm, lateral mitral annulus peak systolic velocity > 6 cm/s, satisfactory right ventricular systolic function without dilatation [15]), the weaning procedure was performed.

The primary endpoint was VA-ECMO weaning failure defined as death occurring during VA-ECMO support or within 24 h after VA-ECMO removal [12, 16]. Secondary endpoints were mortality at day 28 and at 6 months after VA-ECMO implantation. Based on the available literature [17‐24] and on the analysis of patient outcomes in our institutional database [25, 26], we classified indications for VA-ECMO into three categories (high, intermediate, or low) according to the potential for myocardial recovery.

Continuous variables were summarized as mean ± standard deviation and compared using Student’s t test or Mann–Whitney U test depending on their normality. Categorical variables were summarized as counts and percentages and compared using Pearson’s chi-squared test or Fisher’s exact test, as appropriate. Survival at 28 days was reported with Kaplan–Meier curves and compared between the two groups with the log-rank test. We conducted a multivariable logistic regression with propensity score matching [27, 28], which was defined as the probability of exposure to levosimendan. We selected only the covariates most likely to introduce a confounding bias based on clinical expertise and inputs from the literature [20, 29‐31]: potential for myocardial recovery, age, gender, SAPS-II, SOFA, LVEF, duration of VA-ECMO, and lactate level. Next, we performed matching with replacement between patients from the levosimendan group and those from the control group in a 1:10 ratio. Finally, we undertook multivariate weighted logistic regression with weaning failure as an outcome variable and the treatment group and the matched variables as explanatory variables. Results were reported as odds ratios (ORs) together with 95% confidence intervals (CIs) assuming a 5% level of statistical significance. All analyses were conducted in STATA 16.0 (Stata Corp, College Station, Texas 77845 USA).

The flow chart of the study is depicted in Fig. 1. Of 399 patients admitted to the ICU who received VA-ECMO, 199 patients were excluded, leaving a total of 200 patients who met eligibility criteria: 53 in the group levosimendan and 147 in the control group. The use of levosimendan in that specific indication started in 2013 and climbed up over time, reaching 40% of VA-ECMO patients in 2018 (Fig. 2).

No significant difference in baseline characteristics was found between groups except for VA-ECMO duration which was longer in the levosimendan group (10.6 ± 4.8 vs. 6.5 ± 4.7 days, p < 0.001) (Table 1). Indications for VA-ECMO were mainly represented by post-cardiotomy low cardiac output syndrome (29.5%), acute myocardial infarction (22.5%), and graft dysfunction (16.5%). Mean LVEF at admission was 19.6 ± 11.3%, and 30.8% of patients had a significant right ventricular failure. Peripheral VA-ECMO cannulation was performed in 87.5% of cases and 27% of patients had IABP associated with VA-ECMO. Fifty-three (26.5%) patients received levosimendan, the administration starting at 6.6 ± 5.4 days after implantation. Rates of weaning failure were 28.3% and 29.9% in the levosimendan and control groups, respectively (OR 0.92; 95% CI 0.46–1.85). The mortality rate at 28 days was 44.2% in the levosimendan group and 37.5% in the control group (OR 0.69; 95% CI 0.39–2.51) (Fig. 3). Heart transplantation was more frequent in the levosimendan group (13.7% vs. 4.0% respectively, p = 0.017), as was LVAD implantation (11.3% vs. 2.7% respectively, p = 0.014). SOFA score and LVEF at admission were the only covariates associated with weaning failure after multivariate analysis (see Appendix).

After matching of 48 patients in the levosimendan group to 78 in the control group (Fig. 1), the balance of covariates was improved with no statistical difference on any of the covariates, including VA-ECMO duration (Table 2). Rates of weaning failure were 29.1% and 35.4% in the levosimendan and control groups, respectively (OR 0.69; 95% CI 0.25–1.88). No significant difference was found between both groups for all secondary outcomes. Rates of death at 28 days were 41.0% and 41.6% in the levosimendan and the control groups, respectively (OR 1.08; 95% CI 0.42–2.81). Rates of death at 6 months were 50.0% and 54.3% in the levosimendan and control groups, respectively (OR 0.79; 95% CI 0.30–2.07).

Our study suffers several obvious limitations such as its observational nature and a possible lack of power due to the low number of patients included in the matched analysis. Although we have used propensity score matching to reduce selection bias, there is still a risk that our two groups may not be comparable due to the presence of confounding variables not accounted for in our model (unknown or unmeasured confounders) [34]. Thus, the results observed here may not be reproducible within the scope of a randomized controlled trial (RCT). More broadly, observational studies and RCT can generate heterogenous or even conflicting results [35]. Given this limitation, our findings should be viewed cautiously and call for future clinical research using a more robust design. Moreover, in the levosimendan group, 34.8% of patients presented a TAPSE < 12 mm at ICU admission compared to 29.1% in the control group. Although not significant, this result may suggest a higher proportion of patients with right ventricular dysfunction in the levosimendan group and may have contributed to limit the effect of levosimendan on VA-ECMO weaning success in patients presenting bi-ventricular failure [36]. Also, we observed an increasing use of levosimendan over the most recent period with the risk that the variation in team performance over time may have led to minimize the drug effect on VA-ECMO weaning. Another main limitation is that administration of levosimendan occurred late after VA-ECMO implantation. This timing could be too late when compared with other studies [12, 33]. We postulate that physicians started levosimendan in a second step, only in patients demonstrating a weak probability of ECMO weaning success. Many arguments support that hypothesis. First, the duration of VA-ECMO was significantly longer in the levosimendan group in the unmatched analysis. Second, even if not statistically significant, a greater proportion of patients with a high potential for myocardial recovery did not receive levosimendan (control group). Third, more patients in the levosimendan group had heart transplantation or LVAD. By contrast, levosimendan was administered only 3 days after VA-ECMO start in the study reported by Vally et al. [33], a shorter delay that may have contributed to their positive results. Finally, as a tertiary care university hospital with high volumes for transplantation and LVAD, we are more exposed to treat patients with severe refractory cardiogenic shock and supported with VA-ECMO for many days prior to admission in our institution.