ECMO Therapy for Critically Ill Coronavirus Disease 2019 Patients in Wuhan, China: A Retrospective Multicenter Cohort Study

The coronavirus disease 2019 (COVID-19) pandemic has led to surges in the demand for extracorporeal membrane oxygenation (ECMO) therapy. However, little in-depth evidence is known about the application of ECMO therapy in COVID-19 patients. This retrospective multicenter cohort study included 88 patients who had been diagnosed with COVID-19 and received ECMO therapy at seven designated hospitals in Wuhan, China. The clinical characteristics, laboratory examinations, treatments, and outcomes were extracted from electronic medical records and compared between weaned and non-weaned ECMO patients. The patients were followed until June 30, 2020. Logistic regression analyses were performed to identify the risk factors associated with unsuccessful ECMO weaning. Propensity score matching was used to match patients who received veno-venous ECMO with those who received invasive mechanical ventilation (IMV)-only therapy. The primary endpoint, 120-day all-cause mortality after intensive care unit (ICU) admission during hospitalization, was compared using a mixed-effect Cox model.

Background On April 26, 2020, the number of hospitalized COVID-19 cases decreased to zero in Wuhan, the Chinese city most severely affected by the coronavirus disease 2019 (COVID- 19) pandemic. However, the global pandemic situation remains critical, as the number of new severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection cases continues to increase without signs of alleviation. Recent studies have revealed that up to 12.2% of hospitalized COVID-19 patients in the USA were refractory to oxygen therapy and required advanced forms of respiratory support, such as invasive mechanical ventilation (IMV) [1]. COVID-19-induced acute respiratory distress syndrome (ARDS) is characterized by a decrease in lung compliance, which necessitates a high IMV driving pressure to maintain oxygenation and might consequently causes ventilator-induced lung injury. This led to disappointing outcomes of IMV in critically ill COVID-19 patients and mortality rates as high as 88.1% [1]. Extracorporeal membrane oxygenation (ECMO) therapy can support not only ventilation but also gas exchange. Such a complete lung replacement therapy would guarantee both oxygen intake and CO 2 removal. This aspect is particularly important because a de ciency in gas exchange has been identi ed as the dominant cause of hypoxemia in COVID-19 patients who develop ARDS. Moreover, the use of ECMO may protect the lungs, in contrast to IMV, and this could better enable viral depletion and lung recovery.
ECMO has been used widely as a life-saving option in critically ill patients with viral pneumonia since the demonstration of its superiority over IMV during the 2009 H1N1 in uenza pandemic [2,3]. Although sporadic studies have reported the use of ECMO in COVID-19 patients, the results were generally disappointing [4,5]. This study aimed to systematically and comprehensively elaborate the application, e cacy, therapeutic considerations, and outcomes of ECMO application in COVID-19 patients in Wuhan, the largest cohort in China. We sought to identify the risk factors that accounted for unsuccessful ECMO weaning, so as to facilitate decision making regarding eligibility for ECMO implementation. Using propensity score matching, we matched ECMO and IMV patients to compare the effectiveness and outcomes of the two approaches, using 120-day in-hospital mortality after ICU admission as the primary end point. We hypothesized that in critically ill COVID-19 patients, ECMO use would be associated with an improved outcome when compared to IMV-only therapy. Propensity score-matched analysis Propensity score-matched cohorts of patients who received ECMO therapy or IMV-only therapy were created based on variables expected to be confounders associated with the outcomes of ECMO or IMVonly treatment. A detailed description about the propensity score matching is available in the online supplement. Only patients who received veno-venous (V-V) ECMO treatment were included in the ECMO group for this cohort analysis (Fig. 1). Moribund patients who were extremely ill and died within 24 hours after ECMO or IMV initiation were excluded. ECMO and IMV-only patients were paired at a 1:1 ratio according to the propensity scores. A mixed-effect Cox model was used to compare the two treatment groups in terms of the primary endpoint, which was 120-day all-cause mortality after ICU admission during hospitalization.

Statistical analysis
Continuous variables are presented as medians and interquartile ranges (IQRs), and categorical variables are expressed as numbers (N) and corresponding percentages. Continuous variables and ordered categorical variables were analyzed using the Mann-Whitney U test, whereas proportions of unordered categorical variables were compared using the χ 2 test or Fisher's exact test when the sample sizes were small. Univariate logistic regression models were initially applied to analyze the association between each single variable and the nal weaning of ECMO, and indices with p values < 0.1 were included in a further multivariate logistic regression. The odds ratios (ORs) and 95% con dence intervals (CIs) were reported.
The risk of reaching the primary endpoint and the corresponding hazard ratio (HR) were calculated using the mixed-effect Cox model and compared between the ECMO and IMV-only groups, as detailed in the online supplement. The cumulative death rates were compared using the Kaplan-Meier method. A two-sided p value of < 0.05 was considered statistically signi cant. The data were analyzed using the R platform version 3.6.3 with the "tableone" and "stats" packages (R Foundation for Statistical Computing,

Results
Clinical characteristics of COVID-19 patients who received ECMO therapy in Wuhan Ninety-four critically ill COVID-19 patients received ECMO therapy at seven designated hospitals in Wuhan before June 30, 2020. Six patients without nal results due to incomplete medical records were excluded.
Of the remaining 88 patients, 27 were successfully weaned (weaned group) and 61 were unsuccessfully weaned (non-weaned group) from ECMO ( Fig. 1). Figure 2 presents the number of patients concurrently treated with ECMO in Wuhan between January 1 and June 30. ECMO was rst implemented on January 2, and concurrent ECMO cases peaked on March 7 (n = 36 patients; Fig. 2). The rst patient in the weaned group received ECMO support on January 28 (Fig. 2). Patients in the non-weaned group had a median (IQR) age of 62.00 (52.00-68.00) years and were signi cantly older than those in the weaned group (median: 50.00 [42.00-64.00] years; Table 1). The male sex was predominant in both groups ( Table 1). The median (IQR) duration from symptom onset to rst admission was 12.00 (7.00-20.25) days (Table 1). Approximately half of the patients had comorbidities, of which hypertension and diabetes mellitus were the most frequent (Table 1). NT-proBNP=N-terminal pro-brain natriuretic peptide. cTnI=cardiac troponin I. IL-6=interleukin-6. SOFA=sequential organ failure assessment. PEEP=positive end expiratory pressure. PaO 2 /FiO 2 =partial pressure of arterial oxygen/fraction of inspiration oxygen. PaCO 2 =partial pressure of arterial carbon dioxide. IVIG=intravenous immunoglobulin.
The majority of patients presented with lymphocytopenia at the time of ICU admission, and was more severe in the non-weaned group (Table 1). Similarly, small proportions of patients in both groups demonstrated thrombocytopenia and elevated total bilirubin, ALT, and creatinine concentrations (Table 1). Similarly, large proportions of patients in both groups exhibited substantially elevated serum C-reactive protein, LDH, D-dimer, and IL-6 concentrations ( Table 1). Nearly half of the patients exhibited elevated NT-proBNP and cTnI concentrations indicative of an abnormal cardiac functional status ( Table 1). The median Murray score, which is used to assess the severity of ARDS, was 3.0 in both groups ( Table 1). The median sequential organ failure assessment (SOFA) score, which is used to evaluate the functions of extra-pulmonary organs, was 8.00 (IQR: 6.00-10.00).
Compared with the weaned group, the non-weaned group had a signi cantly higher peak respiratory rate and peak PaCO 2 at 24 hours before ECMO initiation and a signi cantly lower minimum PaO 2 /FiO 2 and arterial pH ( Table 1). The peak PEEP and arterial lactic acid values were similar between the two groups ( Table 1). Approximately 30% of patients in both groups had an unstable hemodynamic status necessitating inotropes or vasopressors before intubation (Table 1). Anti-viral, anti-microbial, and antifungal agents; corticosteroids; and intravenous immunoglobulin were administered at similar frequencies in both groups (Table 1). Notably, many more patients in the weaned group (44.4%) than in the nonweaned group (11.5%) received convalescent plasma (Table 1).
Descriptive data related to the use of ECMO Of the 88 enrolled patients, 79 received V-V ECMO therapy (Table 2). Of the seven patients who received veno-arterial (V-A) ECMO, four were weaned and three failed to wean ( Table 2). The two patients who received veno-arterio-venous (V-A-V) ECMO failed to wean from ECMO (Table 2). One patient in each group underwent a switch of the ECMO mode (Table 2). One patient in the weaned group received ECMO therapy while awake and without mechanical ventilation (Table 2). Three patients in the weaned group and two in the non-weaned group had undergone repeated ECMO (Table 2). Five patients in the weaned group and seven patients in the non-weaned group experienced interhospital transportation with ECMO run ( Table 2). The blood ow velocities of ECMO were equivalent in the two groups ( Table 2). Outcomes and complications associated with the use of ECMO Of the 27 patients weaned from ECMO, 15 were further weaned from IMV, 15 were discharged from hospital, 4 died during hospitalization, and 8 remained in hospital (Table 2). Of the 61 patients in the nonweaned group, all patients died during hospitalization ( Table 2). The median times to intubation and ECMO initiation after the onset of symptoms were 20.00 and 23.00 days, respectively, and these were similar between the two groups ( Table 2). The median (IQR) duration of IMV support before ECMO initiation was 2.00 (1.00-6.00) days in the weaned group and 4.00 (2.00-8.00) days in the non-weaned group ( Table 2). The median (IQR) time from intubation to tracheotomy was 25.00 (15.50-30.50) days ( Table 2). The median (IQR) time from ECMO initiation to death was 67.50 (53.50-80.75) days in the weaned group, and was 12.00 (3.00-21.00) days in the non-weaned group ( Table 3). The cumulative ECMO support, IMV support, ICU stay, and hospital stay durations were signi cantly shorter in the nonweaned group than in the weaned group (Table 2), which was largely ascribed to the higher mortality rate in the former group. Microbial culture-con rmed nosocomial pulmonary infection was evident in 73.9% of patients, and this rate was similar between the two groups ( Table 2). Cardiac injury was the most common type of acute organ injury, followed by renal and liver injury ( Table 2). Disseminated intravascular coagulation was observed in 6.8% of all patients (Table 2). Gastrointestinal bleeding was the major hemorrhagic complication ( Table 2). Pneumothorax was observed exclusively in the non-weaned group ( Table 2). The major causes of death were multiple organ dysfunction syndrome and heart failure ( Table 2).
Factors associated with the odds of successful weaning from V-V ECMO After excluding nine patients who received V-A ECMO or V-A-V ECMO support (Fig. 1), a logistic regression analysis was performed to identify factors associated with the likelihood of successful weaning from V-V ECMO. Factors identi ed as statistically signi cant (p < 0.1) in a univariate analysis were entered into the multivariate logistic regression analysis. Notably, a lymphocyte count ≤ 0.5 × 10 9 /L and D-dimer concentration > 4 × upper limit of normal (ULN) at ICU admission and a peak PaCO 2 > 60 mmHg at 24 hours before ECMO initiation were independently correlated with lower odds of successful weaning (Table 3). A tracheotomy during the ICU stay was independently correlated with higher odds of ECMO weaning (Table 3). In contrast, convalescent plasma infusion was signi cant in the univariate analysis but was not independently associated with the odds of weaning ( Table 3).

Comparison of combined ECMO therapy and IMV-only therapy using propensity score-matched analysis
Recent EOLIA study questioned the necessity for ECMO to rescue ARDS [6]. We then evaluated the effectiveness of ECMO support (ECMO group) and IMV-only support (IMV-only group) for the treatment of COVID-19-related ARDS. To avoid confounding variables that might affect the outcomes of critically ill patients, we performed a propensity score-matched analysis (Fig. 1). Moribund patients who died within 24 hours after ECMO or IMV initiation were excluded (Fig. 1). We matched 70 patients from the IMV-only group with 70 patients from the ECMO group at a ratio of 1:1 in the propensity score-matched analysis ( Table 4). The crude 120-day in-hospital mortality rates after ICU admission were 74.3% in the ECMO group and 80.0% in the IMV-only group (Table 4) Table 4. Data are median (IQR) or n/N (%). Upper limit of normal (ULN) and lower limit of normal (LLN) were de ned according to the normal ranges of tests in each hospital. ALT=alanine transaminase. LDH=lactate dehydrogenase. NT-proBNP=N-terminal pro-brain natriuretic peptide. BNP=brain natriuretic peptide. cTnI=cardiac troponin I. IL-6=interleukin-6. PaO 2 =partial pressure of arterial oxygen.
*Standardized difference (SD) values were calculated to compare the mean of baseline covariate between ECMO and IMV-only groups.

Discussion
Herein, we report the descriptive data collected from seven ECMO centers in Wuhan during the COVID-19 pandemic, which accounted for almost all of the ECMO cases in Wuhan. To the best of our knowledge, this is the largest cohort of ECMO-treated patients that has been reported since the COVID-19 outbreak.  [2] and another study that reported a mortality rate of 65% during the Middle East respiratory syndrome pandemic in 2012 [7]. The relatively high mortality rate in this study might be attributable to the following factors. 1) Expert ECMO centers and experienced ECMO specialists were lacking during the COVID-19 outbreak, especially during the early phases when medical systems were overwhelmed. From January 24 to March 8, multiple medical teams and additional ECMO resources were dispatched from all over the nation to Wuhan to combat COVID-19. On February 2, a stringent quarantine policy was implemented, resulting in a decreasing trend in the daily numbers of con rmed cases thereafter [8]. At that time, the shortages of medical resources began to be alleviated, especially in the ICUs. However, ECMO as a salvage therapy might not have been effective in critically ill patients for whom timely treatment had been delayed, leading to substantial irreversible hypoxemic injuries to various organs. 2) Knowledge regarding the etiology, progression dynamics, and effective therapies was scarce during the early phases of the COVID-19 outbreak. Deaths related to this limited experience were re ected by an ECMO-related mortality rate that approached 92% in January 2020. 3) Expertise in ECMO varied among hospitals, which led to signi cant 3-to 4-fold differences in successful weaning between hospitals (data not shown). Hence, centralized management of multidisciplinary care at high-volume ECMO therapy centers would be ideal. During the later phases of the COVID-19 pandemic in Wuhan, ECMO patients were transferred between hospitals, leading to improvements in survival.
Our study revealed that a pre-ECMO PaCO 2 > 60 mmHg was an independent risk factor for a failure to wean from ECMO. This result was consistent with previous studies in which hypercapnia was identi ed as a marker of poor prognosis related to ECMO therapy [9,10]. Moreover, the pre-ECMO PaCO 2 was included in the Respiratory ECMO Survival Prediction-Score model used to predict the outcome of ECMO therapy [11]. As the COVID-19-induced ARDS progresses, disrupted vasoregulation and massive alveolar microthrombi increase the alveolar dead space and cause a high ventilation-perfusion ratio [12].
Additionally, sputum arising from proteinaceous and brinous exudates, pulmonary hemorrhage, cell debris, and nosocomial infection all contribute to airway obstruction and lobular consolidation [13].
Therefore, hypercapnia is an indicator of progression from type 1 to type 2 respiratory failure and a potential signal of ventilation disorders and disease severity. These phenomena suggest that the reversibility of a COVID-19 patients with severe hypercapnia should be evaluated cautiously before ECMO implementation. In contrast, the implementation of ECMO at an early stage before hypercapnia may lead to a better prognosis.
Zhang et al. reported that a D-dimer concentration 4 × ULN was the cutoff value for predicting the inhospital death of a COVID-19 patient [14]. Our results showed that this D-dimer cutoff was independently correlated with decreased odds of weaning from ECMO, consistent with recent ndings suggesting that an elevated D-dimer level predicts a poor outcome in COVID-19 patients [14][15][16]. Moreover, Tang et al.
demonstrated a correlation of prophylactic heparin use with reduced mortality in patients with D-dimer levels exceeding 3.0 mg/L [16]. D-dimer is a rational marker for disease severity and prognosis because i) an elevated D-dimer level is caused mainly by in ammation associated with COVID-19 and subsequent coagulation activity [14], and ii) massive microvascular thrombosis formation in the alveolar capillaries secondary to vasculitis can trigger ARDS during COVID-19 [13,14].
We and other researchers consistently noted that lymphopenia was an indicator of disease severity and a substantial predictor of an unfavorable outcome of COVID-19 [15,17]. Although the underlying mechanisms remain to be elucidated, the main reason likely involves the ability of COVID-19 to induce the exhaustion of lymphocytes, particularly T lymphocytes [18]. As lymphocyte repletion is pivotal in the recovery from COVID-19, ECMO-induced lymphopenia and impaired immunity should be carefully evaluated during treatment [19].
Supportive care remains the mainstay of COVID-19 treatment in the absence of proven effective drugs.
Most patients on ECMO also received prolonged IMV support, which requires meticulous airway management. In these patients, tracheotomy would reduce the need for sedation, enable oral intake, facilitate the e cient suction of secretions, allow for early rehabilitation, and facilitate weaning from IMV.
We demonstrated that patients who received a tracheotomy had independently increased odds of ECMO weaning, even in the presence of greatly delayed therapy (median duration: 25 days post-intubation).
Prolonged intubation and associated sedation would inevitably render the patient vulnerable to ventilatorassociated pneumonia, which would further exacerbate the disease. The observed high mortality rate in the present study may be partially ascribed to the delay in tracheotomy. This delay may be attributable to i) concerns about tracheotomy-provoked aerosol-mediated transmission due to a lack of knowledge about the transmission dynamics of SARS-CoV-2 during the early phases of the pandemic and ii) concerns about uncontrollable bleeding due to heparization. In fact, tracheotomy might be considered before ECMO initiation. Otherwise, the anticoagulation intensity should be reduced, and deep anesthesia would be required at the time of tracheotomy during an ECMO run [20].
Our propensity score-matched analysis revealed that critically ill COVID-19 patients who received ECMO had a better survival outcome than those who received only IMV. Here, ECMO was frequently used as a last resort when IMV therapy failed to provide oxygenation support, and not as a prophylactic therapy.
Our nding was consistent with those of the CESAR study [21] and an analysis of the application of ECMO for H1N1-related ARDS [3]. IMV strategies such as a higher PEEP, lung recruitment maneuvers, and a prone position might help during the early phase of refractory hypoxemia. However, most patients with COVID-19-related ARDS require a high driving pressure to maintain satisfactory oxygenation and overcome a low lung compliance, which increases the risk of ventilator-induced lung injury. Although ECMO is expertise-and cost-intensive and carries a higher likelihood of complications, this option provides a full support for oxygenation and CO 2 removal, enables a "lung rest" ventilatory strategy, and provides time for virus clearance and lung recovery.
Our study has some limitations. First, the retrospective and uncontrolled nature increases the risk of bias in the study. Patients from seven hospitals were enrolled, and each patient was treated at the discretion of the treating hospital or even the attending physicians, without a consensus or a standard clinical pathway. Second, the respirator parameters (e.g., plateau pressures, tidal volumes, or PEEPs) were not routinely recorded. This limitation prohibited us from evaluating the role of IMV in the outcome of ECMO use. Third, although we included nearly all ECMO patients in Wuhan, the sample size remained relatively small. Therefore, the strength of our ndings might be limited.

Conclusion
Our study provides evidence that lymphocytopenia, a higher D-dimer concentration, and pre-ECMO hypercapnia predict a poor outcome of ECMO use in COVID-19 patients. Tracheotomy, even when delayed, could facilitate weaning from ECMO. Compared with IMV-only support, ECMO therapy was associated with a better survival outcome in critically ill patients with COVID-19-induced ARDS. Flow diagram of patient enrollment and propensity score matching in this study.

Figure 2
Histogram of the number of patients concurrently treated with ECMO in Wuhan, China, from January 1 to June 30, 2020. Green and red histograms indicate the patients nally successfully weaned (weaned group) and failed to wean (non-weaned) from ECMO, respectively.

Figure 3
Kaplan-Meier curves of the cumulative probability of in-hospital mortality during the 120-day follow-up period after ICU admission (the primary endpoint) in the ECMO cohort and propensity-score matched IMVonly cohort. The blips on the curve indicate the censoring of cases during the 120-day follow-up period.
An adjusted mixed-effect Cox model with the hospital site as a random effect and adjusting the imbalanced parameters were used to detect the hazard ratio (HR) regarding the primary endpoint. p value was calculated based on mixed-effect Cox model. HR, hazard ratio. CI, con dence interval.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. SupplementalData.docx