Study design
We emulated a target trial assessing the impact of TMCS on mortality in CS using prospective observational data.24 This emulated trial was elaborated to mirror each protocol component from the randomized controlled trial as closely as possible.25 The specifications of target-emulated trials are summarized (see Additional file 1).
This study was conducted at 3 TMCS centers organized into a regional cardiac assistance network (see Additional file 2). The study was designed and led by the heart team of the expert center with input from the hub center teams. The research protocol (ClinicalTrials.gov ID: NCT03528291) was approved by a national ethical committee (Ouest II-Angers IRB).
Study population
Patients or relatives were informed of their inclusion in the study during the intensive care unit (ICU) stay. Given the non-interventional design of the study, a written informed consent was not needed.
The eligibility criteria included adult patients admitted to an ICU for CS in the Languedoc-Roussillon region of France. CS was defined according to the criteria established by the European Society of Cardiology16 which includes systolic blood pressure <90 mmHg despite adequate filling pressure or need of catecholamine to maintain systolic blood pressure >90 mmHg, signs of right sided and/or left sided congestion, signs of tissue hypo-perfusion (oliguria, cold sweats in the extremities, mental confusion, dizziness or blood lactate levels of >2mmol/L), which corresponded to stages C and above (D, and E) according to the clinical expert consensus statement on CS severity (SCAI statement).26 Included patients had no formal contraindication to TMCS (severe aortic regurgitation, severe peripheral artery disease, high comorbidity with a life expectancy lower than one-year, moribund patient with irreversible end organ failure or pregnancy).
This study did not include patients with CS after cardiac surgery or after cardiac arrest if it was refractory or with a no flow >3min and/or out-of-hospital cardiac arrest with non-shockable rhythm. Moreover, CS in the context of myocardial infarction complications (rupture of papillary muscle, left ventricle rupture or acquired ventricular septal defect) and/or massive pulmonary embolism were also exclusionary factors.
Treatment strategy
The decision of TMCS implantation for a patient was made by the heart team after ICU admission. It was mainly based on the initial severity of CS or if CS was refractory to medical treatment (persistent signs of CS despite increasing doses of inotrope and vasopressor). TMCS consisted of VA-ECMO and/or Impella CP (Abiomed Europe GmbH, Aachen, Germany) according to the severity of the CS and/or the need for a left ventricle discharge (see Additional file 2).27
Treatment allocation in the emulated trial
Participants were allocated to the TMCS group if they received TMCS within the first 24 hours after admission. Participants who received TMCS after 24 hours (rescue TMCS), or those who never received TMCS were allocated to the control group. A grace period (24 hours in our emulated trial) corresponded to the maximum time allowed between enrollment and treatment initiation in a pragmatic trial.25 This allocation scheme and the very short grace period restricted the possibility of an immortality time bias.25
Demographic characteristics, medical history and etiology of the CS patients were recorded. At ICU admission, we collected the patients’ arterial blood pressure, left ventricular ejection fraction (LVEF), aortic Doppler velocity time integral (VTI), inotrope and/or vasopressor treatments, bio-chemical markers, TMCS indication (type, and time delay from onset of CS to TMCS request), Simplified Acute Physiology Score II (SAPS II), and we used the Sepsis-related Organ Failure Assessment (SOFA) score to assess illness severity.
Outcomes
The primary outcome was in-hospital mortality. Secondary outcome included mechanical ventilation duration, incidence of renal replacement therapy, ICU length of stay, hospital length of stay, mortality at Day-28 and at Day-180.
Safety end points included thromboembolic events (stroke, peripheral or mesenteric ischemia), transfusion (number of packed red blood cells transfused) and TMCS-associated complications, namely leg ischemia, cannula thrombosis, cannula insertion site infection, and circuit-related hemolysis (plasma lactate dehydrogenase concentration> 500 UI/ml).
Statistical Analysis
Mortality rate without TMCS when CS is extremely severe is about 60%, and around 40% with TMCS.9,28 With an alpha risk of 5%, the recruitment of 206 participants provided 80% power to detect a 20% lower mortality rate in the assisted group (assuming a 1:1 ratio allocation). The number of participants was increased to 240 to account for adjustments to the propensity score and any potential deviation from a 1:1 ratio.
The analysis of all outcomes followed the intention-to-treat principle whereby all participants (no expected missing values for the primary outcome) were analyzed in the initial allocated group. The baseline was the date of ICU hospitalization in both groups. In order to mimic randomization, the treatment effect was estimated using Inverse Probability of Treatment Weighting (IPTW) to account for relevant baseline covariates. This IPTW method included all participants in the analysis. IPTW were calculated as the inverse of propensity score among participants of the intervention group, and the inverse of (1-propensity score) among participants of the control group. The propensity scores consisting of conditional probabilities of receiving TMCS were calculated by multivariable logistic regression. TMCS was the predicted variable, and the predictive variables were selected variables of interest such as age, LEVF, VTI, serum lactate level, tracheal intubation, serum bicarbonate level, SOFA score and cardiac arrest.
A per group baseline descriptive analysis was executed. Raw data and weighted data were expressed as numbers and percentage (n, %) or median and Interquartile range (m, [IQR]) when appropriate. Baseline raw data were compared between the two groups using Chi-square tests (qualitative variables) and Wilcoxon-Mann-Whitney tests (quantitative variables). Weighted data were compared between the two groups using standardized mean differences (SMD). A SMD < 0.25 was considered as evidence of balance.29,30 If the difference was outside the range for one or more variables, the latter were included in the Cox proportional hazard model to assess whether they modified the association between TMCS and mortality.
The emulated trial mortality outcomes were compared between arms using hazard ratio (HR) estimated from a IPTW Cox proportional hazard model. Furthermore, the primary outcome (in-hospital survival) and survival at Day 28 and Day 180 were estimated within the weighted population using adjusted Kaplan-Meyer estimators and the two groups were compared using log-rank tests.
IPTW secondary outcomes were compared between the two groups using absolute difference for proportions, and Hodges-Lehmann median differences for quantitative variables. IPTW safety outcomes were compared using Chi-square tests (qualitative variables) and Wilcoxon-Mann-Whitney tests (quantitative variables). Crude and multiplicity-adjusted p-values were calculated using the false discovery rate correction for secondary outcomes.
A sensitivity analysis was carried out by allocating in the control group all patients who had rescue TMCS during the grace period.
Finally, sub-group analyses were planned on the main CS etiologies and for TMCS with VA-ECMO only. For each of these sub-group analyses, the same approach using IPTW was used, and the propensity scores were recalculated.
Statistical analyses were programmed using SAS V7.2 (SAS Institute Inc., Cary, North Carolina, USA).