Levosimendan in patients undergoing extracorporeal membrane oxygenation after cardiac surgery: an emulated target trial using observational data

Abstract

Background: Retrospective cohorts have suggested that levosimendan may facilitate the weaning of veno-arterial extracorporeal membrane oxygenation (VA-ECMO). We therefore studied this clinical question by emulating a randomized trial with observational data.

Methods: All patients with refractory postcardiotomy cardiogenic shock and assisted with VA-ECMO, admitted to a surgical intensive care unit at La Pitié-Salpêtrière Hospital between 2016 and 2019, were eligible. To avoid immortal-time bias, we emulated a target trial sequentially comparing levosimendan administration versus no levosimendan administration in patients treated with VA-ECMO. The primary outcome was time to successful ECMO weaning. The secondary outcomes were 30-day and 1-year mortality. We performed a multivariable analysis to adjust for confounding at baseline.

Results: Two hundred and thirty-nine patients were included in the study allowing building a nested trials cohort of 1434 copies of patients. No association of levosimendan treatment and VA-ECMO weaning was found (HR = 0.91, [0.57; 1.45], p = 0.659 in multivariable analysis), or 30-day mortality (OR = 1.03, [0.52; 2.03], p = 0.940) and 1-year mortality (OR = 1.00, [0.53; 1.89], p = 0.999).

Conclusions: Using the emulated target trial framework, this study did not find any association of levosimendan treatment and ECMO weaning success after postcardiotomy cardiogenic shock. However, the population of interest remains heterogeneous and subgroups might benefit from levosimendan.

Fig. 1

Sequential emulation of the target trial. The target trial (as defined in Additional file 1: Table S2) was emulated several times in a sequence of nested trials, to make coincide treatment assignment and start of follow-up, thus avoiding immortal-time bias. This figure describes the construction of the observations in each nested trial with their respective treatment assignment, measurement of confounders, and follow-up. A hypothetical cohort is presented for simplicity. Panel A: The figure depicts the construction of the first emulated trial. At day 0 (first day of ECMO support), no patient received levosimendan; thus, the target trial was not emulated as the comparison was not feasible between patient receiving and not receiving levosimendan. At day 1 (D1), five patients (P5 to P9) started levosimendan, and so could be compared to the four patients (P1 to P4) that did not receive levosimendan. Their treatment assignment is in line with the received strategy at the day of the target trial emulation (here day 1). In addition, baseline confounders for the first nested trial are measured on day 1 (red rectangle), that is, day 1 since beginning of ECMO. Follow-up is restarted at day 1 for the first emulated trial to make coincide eligibility, treatment assignment and time-zero. Panel B: The figure depicts the construction of the second emulated trial. At day 2 after beginning of ECMO support, three patients did not receive levosimendan (P1, P2, and P4) and one patient received levosimendan (P3). Patients that received levosimendan in the previous nested trial at day 1 are not eligible anymore, as defined in the target trial protocol (new-user design). As for day 1 nested trial, treatment assignment of cloned copies of eligible patients is based on the treatment received the day of the emulated trial (blue rectangle) and baseline confounders for the second nested trial are measured on day 2. Follow-up is again restarted on the day 2 of the nested trial. Each of these nested emulated trials data is then stacked in a unique dataset, built with cloned copies of participants with assigned treatment, reset follow-up and baseline confounders in each nested trial. A multivariable Cox model is then fitted on the stacked cohort of observations and adjusted for confounders measured at ICU admission and at baseline of each nested trial (lactatemia, ECMO output, VIS, and SOFA score)

Fig. 2

Flow diagram of the population included at the beginning of ECMO support

Fig. 3

Flow diagram of the nested trials cohort. A nested trial was emulated each day (D) a patient received levosimendan since beginning of ECMO support

Fig. 4

Cumulative incidences of ECMO weaning success in the presence of competing risks of death and ECMO weaning failure after ECMO removal. The cumulative incidence curves of ECMO weaning success were plotted with the Gray’s estimator that accounts for the competing risk of death and the competing risk of ECMO failure after ECMO removal [32]. The x-axis represents the days since the beginning of ECMO support