Predictors of in-hospital mortality after successful weaning of venoarterial extracorporeal membrane oxygenation in cardiogenic shock

Limited knowledge exists regarding the predictors of mortality after successful weaning of venoarterial extracorporeal membrane oxygenation (ECMO). We aimed to identify predictors of in-hospital mortality in patients with cardiogenic shock (CS) after successful weaning from ECMO. Data were obtained from a multicenter registry of CS. Successful ECMO weaning was defined as survival with minimal mean arterial pressure (> 65 mmHg) for > 24 h after ECMO removal. The primary outcome was in-hospital mortality after successful ECMO weaning. Among 1247 patients with CS, 485 received ECMO, and 262 were successfully weaned from ECMO. In-hospital mortality occurred in 48 patients (18.3%). Survivors at discharge differed significantly from non-survivors in age, cardiovascular comorbidities, cause of CS, left ventricular ejection fraction, and use of adjunctive therapy. Five independent predictors for in-hospital mortality were identified: use of continuous renal replacement therapy (odds ratio 5.429, 95% confidence interval [CI] 2.468–11.940; p < 0.001), use of intra-aortic balloon pump (3.204, 1.105–9.287; p = 0.032), diabetes mellitus (3.152, 1.414–7.023; p = 0.005), age (1.050, 1.016–1.084; p = 0.003), and left ventricular ejection fraction after ECMO insertion (0.957, 0.927–0.987; p = 0.006). Even after successful weaning of ECMO, patients with irreversible risk factors should be recognized, and careful monitoring should be done for sign of deconditioning.


Patient management, outcome measurement and definition of variables
Venoarterial ECMO was performed at the physician's discretion for short-term mechanical circulatory support of refractory CS.ECMO was removed (i) when the patient had successfully recovered from the critical phase of CS, or (ii) when it was deemed by a physician and legal representative that the patient was unlikely to recover from shock (hopeless removal).ECMO was weaned and removed with comprehensive consideration of clinical criteria, and successful ECMO weaning was defined as survival with minimal mean arterial pressure (> 65 mmHg) for more than 24 h after ECMO removal.The primary outcome was in-hospital mortality after successful weaning from ECMO, while the secondary outcome was all-cause mortality at the 1-year follow-up.The maximal usage of vasoactive agents was quantified with the inotropic and vasoactive-inotropic scores using the formula suggested by Gaies et al. 9 .Left ventricular ejection fraction (LV EF) was measured at the presentation of shock and was followed up upon clinical need.LV EF was specified by lowest LV EF before and after ECMO insertion.Further clinical definitions for each variable are provided in the supplemental material (Supplementary Table S1).

Statistical analysis
Categorical variables were described as numbers and percentages, and continuous variables were described as means and standard deviations.To compare variables, the Student's t-test, Mann-Whitney U test, chi-square test, or Fisher's exact test was used as indicated.Logistic regression was used to identify the predictors of inhospital mortality, and Kaplan-Meier analysis and the Cox proportional hazards model were used to assess time-dependent variables.Potential outliers and missing patterns of variables were examined, and variables with < 5% missing values were considered for the multivariable analysis (Supplementary Fig. S1).One variable (LV EF after ECMO insertion) revealed higher missing value, but was included for analysis regarding its clinical significance.Continuous variables with missing values were imputed using their mean value.Clinically intercorrelated variables were excluded from the multivariable regression analysis, and backward selection was used to identify statistically significant and 0.1 level for inclusion.Least Absolute Shrinkage and Selection Operator

Ethics approval and consent to participate
The current study was approved by the Institutional Review Board of all participating hospitals (Korea University Anam Hospital, Samsung Medical Center, Severance Cardiovascular Hospital, Ewha Woman's University Seoul Hospital, Konkuk University Medical Center, Chungbuk National University College of Medicine, Ilsan Paik Hospital, Chung-Ang University Hospital, Chungnam National University Hospital, Inha University Hospital, Dankook University Hospital).The ethical guidelines of the 2013 Declaration of Helsinki and legal medical regulations of Republic of Korea were strictly undertaken throughout the study.Written informed consent was obtained from all participants or their legal representatives in patients that were prospectively enrolled, and written infomed consent was waived in the retrospectively enrolled patients.

Study population
A total of 1247 patients with CS were included in the RESCUE registry from 2014 to 2018.Among them, 485 underwent venoarterial ECMO for refractory CS (Fig. 1).Of these, 173 patients did not survive or were transferred to another hospital before ECMO removal.Consequently, 312 patients underwent ECMO removal, either after recovery from shock or after hopeless removal.ECMO was successfully weaned in 262 patients, whereas 50 failed to wean.Among 262 patients with successful ECMO weaning, 48 (18.3%) did not survive to discharge.

Predictors for in-hospital mortality
Predictors of in-hospital mortality were identified using multivariable logistic regression analysis.Statistically significant variables in the univariable logistic regression analysis were as follows: age, hypertension, diabetes mellitus, dyslipidemia, chronic kidney disease, inotropic score, ischemic origin, peak CK-MB level, LV EF after ECMO insertion, use of IABP, use of CRRT, use of a mechanical ventilator, and extracorporeal cardiopulmonary resuscitation (Supplementary Table S3).In multivariable regression analysis, use of CRRT was the strongest predictor of in-hospital mortality (odds ratio  S4).

Discussion
This study investigated the clinical characteristics of patients who underwent successful ECMO weaning and identified the predictors of in-hospital mortality in this population.Although in-hospital mortality of patients with successful ECMO weaning was relatively low, it was not negligible (18.3%).Moreover, patients with successful ECMO weaning demonstrated distinct clinical characteristics compared with those who died in the early phase of CS.Namely, clinical parameters reflecting the severity of the early phase of CS were not identified as predictors of in-hospital mortality after successful weaning from ECMO.Predictors in this population underscored the significance of (i) risk factors that were not modifiable by ECMO support (age, pre-existing diabetes mellitus, prolonged pump failure measured as LV EF after ECMO insertion) and (ii) the need for adjunctive end-organ support even after ECMO insertion (IABP, CRRT).
Venoarterial ECMO has emerged as a salvage strategy for refractory CS that provides short-term cardiopulmonary support for patients with CS who do not respond to conventional medical therapies.Although the use of ECMO for refractory CS has increased over the past two decades, the outcome of refractory CS remains suboptimal 10 .Successful weaning is conventionally defined as having no requirement for further mechanical circulatory support 30 days after ECMO removal 11,12 .However, successful weaning is not always equivalent to successful hospital discharge and survival 11 .Therefore, the procedure does not guarantee promising outcomes following ECMO device removal.As a result, constant efforts are needed to extend successful weaning to survival after discharge and long-term survival after discharge.The aforementioned clinical differences in this population underpin the different approaches to risk assessment in patients who have undergone successful ECMO weaning.
Several risk prediction scores have been developed to assess short-term mortality in patients with CS requiring ECMO 5,[13][14][15] .However, previous studies focused on the outcomes of ECMO-treated patients with CS as a whole, which comprised patients with two distinct clinical courses: those who did not survive until ECMO weaning and those who survived after ECMO weaning.While it is not fully established, distinct differences exist between patients who died in the early phase of CS and those who survived until ECMO weaning.Thus, previous risk scores, while useful for providing a generalized risk assessment for ECMO-treated patients with CS, may not be suitable for predicting outcomes in patients who survive until ECMO weaning.

Comparison with previous studies of predictors of in-hospital mortality
Several studies on CS have identified the risk factors for in-hospital mortality.Schmidt et al. identified pre-ECMO factors that predict in-hospital survival and developed a risk calculation score called the survival after venoarterial ECMO score 5 .The independent predictors of in-hospital survival included demographic factors (age and weight), cause of CS, presence of organ failure, and several parameters assessed during the early phase of CS (pre-ECMO cardiac arrest, diastolic blood pressure before ECMO, peak inspiratory pressure, serum bicarbonate level, and duration of intubation before ECMO).Previous studies from the RESCUE registry reported similar findings.Yang et al. reported that age, body mass index, cardiac arrest at presentation, vasoactive-inotropic score, use of CRRT, use of a mechanical ventilator, use of IABP, and use of ECMO were independent predictors of in-hospital mortality for CS 8 .Seong et al. focused on ECMO-treated populations from the same registry and suggested the following predictors of in-hospital mortality: body mass index, lactic acid level, shock-to-ECMO time, cardiopulmonary resuscitation, use of a mechanical ventilator, use of CRRT, LV EF, and distal limb perfusion 6 .Namely, previous studies have highlighted (i) parameters that indicate the severity of the initial phase of CS, (ii) presence of other organ failures, and (iii) patients' underlying demographic risk factors (age, weight, and comorbidities) as risk factors for mortality from CS.
In this study, in-hospital mortality for patients who underwent successful ECMO weaning was improved significantly (18.3%) compared with that in previous studies in the same cohort (33.6% in all CS cohort 8 , and 52.2% in ECMO-treated patients with CS 6 ) and other studies that reported in-hospital mortality after ECMO weaning as high as 30% 16,17 .However, in-hospital mortality was not negligible even after successful ECMO weaning.Interestingly, parameters indicative of the early phase of CS that were significant in previous studies (lactic acid level, shock-to-ECMO time, initial cardiac arrest, blood pressure, and vasopressor requirement) did not predict in-hospital mortality in this specific group.This finding implies that the mortality of patients with more severe CS (such as lower blood pressure with higher vasopressor requirement, higher lactic acid level, prolonged shock time, and a history of cardiac arrest) is determined during the early phase of CS.Accordingly, the mortality of those who survive the early phase of shock and proceed to successful ECMO weaning is not determined by early parameters.Rather, patients who die even after successful weaning of ECMO have significant risk factors, such as age, diabetes mellitus, or prolonged pump failure, that are not modifiable with ECMO support.In addition, these patients require adjunctive therapies in addition to ECMO, such as IABP and CRRT.A comparison of non-survivors who underwent successful ECMO weaning and those who did not (Supplementary Table S3) supports the robustness of the aforementioned predictors.

Clinical implications
Our findings have several implications in clinical practice.In patients with CS who survive until successful weaning from venoarterial ECMO, a non-negligible degree of in-hospital mortality should be perceived.That is, even after successful weaning of ECMO, approximately one out of five patients do not survive to discharge.Patients with higher risk of in-hospital mortality after successful ECMO weaning-including those with older age, presence of diabetes mellitus, prolonged pump failure, history of using IABP or CRRT-should be recognized, and be informed of their risk.Consequently, more thorough monitoring should be done in this population to detect sign of deterioration, and prompt clinical intervention should be pursued.Careful monitoring should be maintained until successful discharge and follow-up after discharge.Although we have identified extensive organ failure as a predictor for in-hospital mortality, specific risk factors that contribute to extensive organ failure need to be clarified.Ischemic CS is the most common cause of CS, which is known to have worse prognosis 18 .In our cohort, 85.4% of non-survivors had CS of ischemic origin.Coronary hypoperfusion with myocardial ischemia could contribute to multi-organ failure, and degree of coronary hypoperfusion is highly influenced by factors such as severity of coronary artery disease, time to revascularization, and revascularization strategy (culprit-only revascularization or multi-vessel revascularization) 19,20 .Therefore, focusing on ischemic CS and identifying contributors to the myocardial ischemia may verify novel, modifiable risk factors that contributes to in-hospital mortality after successful weaning of ECMO.

Limitations
This study had several limitations.First, although we specified successful weaning of ECMO as maintaining blood pressure after ECMO removal, the detailed reasons for weaning were not obtained.Thus, patients who underwent ECMO weaning after recovery from the shock phase were not distinguished from those who underwent unsuccessful ECMO removal.Consequently, patients with weaning failure (n = 50) may have included those who failed ECMO weaning after recovery from shock and those who underwent hopeless removal of ECMO.Nonetheless, only nine patients from the weaning failure group (n = 50) survived until discharge, implying that the survival rate was much lower in the weaning failure group than in the successful weaning group.Second, some clinical parameters may vary greatly according to the phase of shock and the use of supportive therapy.For instance, the assessment of LV EF is highly influenced by inotropic agents as well as the use of ECMO and its www.nature.com/scientificreports/setting.Laboratory marker levels also change dynamically during shock; therefore, several variables were specified according to the temporal period and severity (i.e., highest lactic acid level before and after ECMO insertion, and lowest LV EF before and after ECMO insertion).We also attempted to specify clinical variables that reflect the most severe phase of shock, such as the lowest blood pressure measured at the date of shock, highest dose of vasoactive agents at the initial phase of shock (inotropic score, vasoactive-inotropic score), and highest level of cardiac markers (peak troponin-I and peak CK-MB levels).Finally, although diabetes mellitus was identified as an independent risk factor for mortality, further clinical information about diabetes mellitus, such as duration of diabetes or severity (glycosylated hemoglobin type A1c level, diabetic complications, and need for insulin), was not provided in this study.A similar shortcoming was seen in other parameters, such as the use of CRRTspecific clinical data, indications for CRRT, and serial changes in creatinine levels.Further studies with more specific clinical data may provide additional supporting evidence for our findings.

Conclusion
Distinctive clinical characteristics exist between patients who survived ECMO weaning and those who did not, thus requiring a different approach to risk assessment.Independent predictors of in-hospital mortality identified after successful ECMO weaning include underlying irreversible risk factors and the need for adjunctive therapy for organ failure.Even after successful weaning of ECMO, patients with higher risk of mortality-that have irreversible risk factors or history of adjunctive therapy for organ failure-should be recognized.In this population, signs of deconditioning should be promptly discerned, and further thorough assessment should be done till successful discharge.

Figure 3 .
Figure 3. Kaplan-Meier curves for all-cause mortality at 1-year follow up (Binary variables).Kaplan-Meier curves of all-cause mortality according to presence of DM (a), use of IABP (b), and use of CRRT (c).DM diabetes mellitus, IABP intra-aortic balloon pump, CRRT continuous renal replacement therapy.