Extracorporeal Membrane Oxygenation to Support COVID-19 Patients: A Propensity-Matched Cohort Study

Background In patients with severe respiratory failure from COVID-19, extracorporeal membrane oxygenation (ECMO) treatment can facilitate lung-protective ventilation and may improve outcome and survival if conventional therapy fails to assure adequate oxygenation and ventilation. We aimed to perform a confirmatory propensity-matched cohort study comparing the impact of ECMO and maximum invasive mechanical ventilation alone (MVA) on mortality and complications in severe COVID-19 pneumonia. Materials and Methods All 295 consecutive adult patients with confirmed COVID-19 pneumonia admitted to the intensive care unit (ICU) from March 13th, 2020, to July 31st, 2021 were included. At admission, all patients were classified into 3 categories: (1) full code including the initiation of ECMO therapy (AAA code), (2) full code excluding ECMO (AA code), and (3) do-not-intubate (A code). For the 271 non-ECMO patients, match eligibility was determined for all patients with the AAA code treated with MVA. Propensity score matching was performed using a logistic regression model including the following variables: gender, P/F ratio, SOFA score at admission, and date of ICU admission. The primary endpoint was ICU mortality. Results A total of 24 ECMO patients were propensity matched to an equal number of MVA patients. ICU mortality was significantly higher in the ECMO arm (45.8%) compared with the MVA cohort (16.67%) (OR 4.23 (1.11, 16.17); p=0.02). Three-month mortality was 50% with ECMO compared to 16.67% after MVA (OR 5.91 (1.55, 22.58); p < 0.01). Applied peak inspiratory pressures (33.42 ± 8.52 vs. 24.74 ± 4.86 mmHg; p < 0.01) and maximal PEEP levels (14.47 ± 3.22 vs. 13.52 ± 3.86 mmHg; p=0.01) were higher with MVA. ICU length of stay (LOS) and hospital LOS were comparable in both groups. Conclusion ECMO therapy may be associated with an up to a three-fold increase in ICU mortality and 3-month mortality compared to MVA despite the facilitation of lung-protective ventilation settings in mechanically ventilated COVID-19 patients. We cannot confirm the positive results of the first propensity-matched cohort study on this topic. This trial is registered with NCT05158816.


Introduction
Coronavirus disease 2019 (COVID- 19) is a viral infection caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). A high viral load will cause both a direct viral cytopathic efect as well as an immune response with a cytokine storm, potentially resulting in severe pneumonia and acute respiratory distress syndrome (ARDS) [1]. Treatment of these patients includes intensive care unit (ICU) admission and therapy with conventional methods established for ARDS, including lung-protective mechanical ventilation, neuromuscular blockade, and prone positioning [2].
In the selected patients with severe respiratory failure from COVID-19, treatment with extracorporeal membrane oxygenation (ECMO) can facilitate lung-protective ventilation which may improve outcome and survival if conventional therapy fails to assure adequate oxygenation and ventilation [3][4][5]. Terefore, International guidelines recommend, depending on the availability of resources, considering venovenous (VV) ECMO in the selected patients with COVID-19 who develop severe ARDS and hypoxemia refractory to prone positioning and optimal ventilator management [2,6,7]. More specifcally, VV ECMO should be considered in the selected patients with the sustained PaO2/Fi02 ratio (P/F Ratio) < 60 mmHg or pH < 7.20 + PaCO 2 > 80 mmHg despite maximizing conservative therapies [7]. VA ECMO should be timely considered before the development of multiple organ failure in the selected patients with the coexistence of refractory cardiogenic shock [7]. Besides its known benefts, ECMO support also carries an increased risk of bleeding and thromboembolic events [8] and, therefore, may have a negative impact on the survival rate.
Despite the growing body of literature on ECMO therapy in COVID-19 patients, randomized controlled trials comparing outcomes and adverse events of ECMO therapy versus conventional respiratory support are lacking due to ethical concerns. In the absence of randomized controlled trials (RCTs), propensity-matched cohort studies comparing the outcomes of ECMO therapy versus maximum ventilation alone (MVA) for ECMO-eligible COVID-19 patients in homogeneous cohorts are the best available study design. Diferent propensitymatched cohort studies has been published with a demonstrated 3-fold improvement in survival with ECMO (75%) compared to MVA (26.2%) [9]. Te high survival rate in the ECMO group of this study is not reported in other publications, and as result, confrmatory studies are necessary. Others show an absolute mortality reduction of 18.2% (44% vs. 25.8%) for treatment with ECMO compared to MVA [5].
JESSA hospital, Hasselt, was situated at the epicenter of the Belgian outbreak during the frst COVID wave with the highest incidence across the country [10]. From the frst admission to ICU on March 13 th , admissions of critically ill COVID-19 patients to ICU grew exponentially [11] which resulted in very high thresholds for initiating ECMO during the frst wave. Indeed, surge conditions result in decreased utilization of EMCO, as constrained resources must be utilized efciently to ensure an acceptable level of care in all patients [12]. Tis high threshold for ECMO during the frst wave, however, may enhance the selection of a well-matched cohort of COVID-19 patients treated with MVA.
Hence, this study aimed to perform a confrmatory propensity-matched cohort study comparing the impact of ECMO and MVA on mortality and complications in severe COVID-19 pneumonia. Te main hypothesis was that the initiation of ECMO therapy in selected patients would reduce mortality and improve clinical outcomes in critically ill COVID-19 patients admitted to ICU.

Materials and Methods
Tis single-center, longitudinal, retrospective, investigatorinitiated, propensity-matched cohort study was performed at Jessa Hospital, Hasselt, Belgium. Tis study is approved by the Ethical Committee of Jessa Hospital, Hasselt, Belgium, on 8 th September, 2021, and registered on clinicaltrials.gov. (NCT05158816). Te requirement for informed consent from the study subjects was waived by the Ethical Committee of Jessa Hospital due to the urgent need to collect data on the ongoing pandemic and the retrospective nature of this study. Tis study was performed in accordance with all relevant guidelines and regulations and in accordance with the Declaration of Helsinki. Te study is reported according to the STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) guidelines [13].

Study Population.
All adults (>18 years) with acute hypoxemic respiratory failure due to diagnosed COVID-19 pneumonia and admitted to ICU from 13th March 2020 until 30th June 2021 were included in this analysis. Following the World Health Organisation (WHO) protocol [14], laboratory confrmation of COVID-19 infection was defned as a positive result on polymerase chain reaction (PCR) assays of nasopharyngeal swab samples or bronchoalveolar lavage. Only laboratory-confrmed patients were included in the analysis. Data from 295 consecutive patients admitted to the ICU from March 13 th , 2020, until July 31 st , 2021, were prospectively entered into a customized database that included medical history, demographic data, clinical symptoms and signs, laboratory results, ventilator settings, ventilator-derived parameters, and clinical outcomes [15]. Tis database was retrospectively reviewed [15]. APACHE II and APACHE IV scores were calculated on ICU admission [16,17]. Te sequential organ failure assessment (SOFA) score [18] was evaluated on a daily basis.
All patients were classifed into 3 categories on admission based on their medical history, age, and clinical frailty index: (1) full code including potential initiation of ECMO therapy (AAA code), (2) full code excluding ECMO (AA code), and (3) do-not-intubate (A code). Inclusion criteria for ECMO candidacy during the frst wave were as follows: prone ventilation, neuromuscular blockade, age <60 years, sustained severe hypoxemia (P/F-ratio <60 mmHg) or hypercapnia (pH < 7.20 + PaCO2 > 80 mmHg) despite maximum ventilator support, and clinical frailty scale of 1 or 2 [15]. After the frst wave in May 2020, inclusion criteria were extended to clinical frailty scale <5 and age <70 years (with age between 70 and 80 years only a relative contraindication) [15]. Exclusion criteria for ECMO were known active malignancy, severe chronic organ failure (i.e. hepatic cirrhosis Child-Pugh B or C or COPD GOLD IV), signs of acute, cardiac arrest, severe bleeding, and known severe neurological injury or cognitive impairment (including stroke or dementia) or multiple organ failure involving three or more organ systems [15]. [15]. All COVID-19 patients were treated according to the COVID protocol of the JESSA hospital based on the latest insights on COVID-19 at that timepoint [2,15,19]. According to this protocol, all patients admitted to our ICU received an intravenous (IV) infusion with glucose 5% at 60 ml/h as maintenance fuid and stress ulcer prophylaxis with pantoprazole 40 mg intravenously daily. Prophylactic antibiotic therapy was initiated for 5 days, using amoxicillin-clavulanic acid 1 g IV 4 q.i.d. or moxifoxacin 400 mg IV QD in case of known allergy to penicillin. Prophylactic administration of antibiotics was abandoned on 08 th April 2020. Initially, corticosteroids were administered with caution and minimally after 1 week of ICU admission based on the clinical judgment of the attending intensivist. After the publication of the frst results of the RECOVERY trial in July 2020, all patients received intravenous dexamethasone at a dose of 6 mg once daily for ten days after admission. Ventilatory support was initiated with a high-fow nasal cannula or noninvasive mechanical ventilation as long as the patient was cooperative. Awake-prone positioning was also applied in cooperative patients who required support with a high-fow nasal cannula. In case of respiratory fatigue, patients were sedated and intubated and invasive mechanical ventilation (IMV) was started according to the ARDS network guidelines. Tis was based on the frst reports that viral pneumonia caused by SARS-CoV-2 mimicked an ARDS-like pattern [2]. Sedation was performed by a combination of propofol, midazolam, and piritramide in selected cases in association with ketamine, clonidine, or dexmedetomidine, always aiming for the lowest level of sedation required to tolerate IMV. Te intermittent use of neuromuscular blocking agents was applied when required. Adjustments were made guided by pulse oximetry levels, which were continuously monitored, and arterial blood gasses took every 4 hours. In case of hypotension due to vasoplegia, norepinephrine was used as the frst choice vasopressor.

Anticoagulation.
Between March 13th, 2020, and March 30th, 2020, all patients received routine low dose pharmacological VTE prophylaxis, i.e., QD subcutaneous injection of nadroparin calcium 2850 IU. On March 30 th , 2020, a high incidence of deep venous thrombosis was discovered [20], for which we changed our prophylactic anticoagulation protocol from prophylactic to intermediate dosages of low molecular weight heparin (LMWH) with plasma anti-Xa activity monitoring [11]. Te anti-Xa activity was measured daily and targeted at 0.3 to 0.5 IU/ml in patients without echographic fndings of deep venous thrombosis (DVT) and 0.4 to 1 IU/ml in patients with screening duplex positive for DVT. Patients were routinely screened for DVT, using ultrasonography twice per week.
Other haemostasis parameters were also measured daily and included the activated partial thromboplastin time (aPTT), international normalized ratio (INR), platelet count, and fbrinogen.
At the initiation of ECMO therapy, LMWH therapy was stopped and unfractionated heparin (UFH) was started. In these patients, aPTT was measured six times per day and targeted at 60-80 seconds in patients without clot formation in the ECMO circuit or echographic fndings of DVT and 80-100 seconds in patients with documented thrombus formation. [15]. A standard ECMO/ECLS circuit was used for all patients, including a Hico Variotherm 550 heater/cooler, Sechrist gas blender, a LivaNova Stöckert console with a Revolution centrifugal pump system, and a Medtronic Biotrend SvO2 meter. Te disposables consisted of a LivaNova Revolution centrifugal pump head with line reassure control in 3 places: P1 negative drainage pressure, P2 preoxygenator pressure, and P3 postoxygenator pressure and a coated VA-tubing set with a PMP fber ECMO oxygenator (LivaNova EOS ECMO or Eurosets A.L. ONE ECMO). 5000 IU of UFH were administered IV before cannulation according to our protocol. A cardiac surgeon performed the venous drainage cannulation. After disinfection and preparing the groin, a 21Fr. or 25Fr. Medtronic multistage venous cannula was inserted percutaneously with Seldinger technique into the RFV (right femoral vein) under ultrasound guidance. Te tip of the cannula was placed into the VCI to avoid recirculation. Simultaneously, the venous return cannula (Edwards Optisite 20Fr. or 22Fr.) was inserted by a cardiac anaesthesiologist into the RIJV (right internal jugular vein) with the tip positioned towards the tricuspid valve. After ultrasound control of the position of both cannulas and ACT check, ECMO was initiated. Blood fow was increased with a target of 2,4 LPM CI (cardiac index), taking the limitations of negative venous drainage pressures into account. Fine tuning of ECMO ventilation/oxygenation settings was performed led by arterial blood gas sampling. A rather high level of PEEP (>10 cm H 2 O) was maintained during ECMO. Te pressurecontrolled mode of ventilation was preferred with an RR 10-12/min, FiO 2 tapered to 0.4, a tidal volume target of 4-6 ml/ kg, PIP <30 cm H 2 O, and plateau pressures <25 cm H 2 O.

Outcome Parameter.
Te primary endpoint is ICU mortality indicating the study population was divided into patients who died at the ICU and patients who were discharged from the ICU. Te key secondary outcome in this study is 3-month mortality. Other secondary outcomes include the incidence of acute kidney injury and continuous renal replacement therapy (CRRT), other complications during ECMO, length of stay (LOS) in the ICU, and hospital LOS. All patients were followed for at least 3 months after submission to ICU. Te data set was closed on October 31st, 2021, ensuring that all patients reached the primary and key secondary outcomes.
2.6. ICU Scoring Systems. APACHE II, APACHE IV, and SOFA scores were calculated https://www.via.mdcalc.com within the frst 24 hours after admission to our ICU. Te data with the highest severity were used to calculate these scores [19].
Critical Care Research and Practice 2.7. Defnitions. Acute kidney failure was diagnosed according to the KDIGO clinical practice guidelines [21]. ARDS was diagnosed according to the Berlin defnition [22]. Sepsis and septic shock were defned according to the 2016 Tird International Consensus Defnition for Sepsis and Septic Shock.

Statistical
Analysis. For descriptive purposes, continuous data are shown as mean ± standard deviation (SD) and categorical data are presented as frequencies (%). For the 271 non-ECMO patients, match eligibility was determined based on the following applied criteria: patients with AAAcode (ECMO candidacy) on admission and treated with IMV. Subsequently, propensity score matching was performed using a logistic regression model including the following variables: gender, P/F ratio, SOFA score at admission, and date of ICU admission. More specifcally, the worst P/F-ratio during IMV therapy in the non-ECMO group was compared with the P/F-ratio before starting ECMO in the ECMO group. To minimize the risk of selecting a falsely reduced P/F ratio due to sputum plugs or other mechanical problems, the worst P/F ratio was only selected taking into account the global evolution of P/F ratios over time. "Date of ICU admission" or "wave" was included in the propensity score model to prevent an asymmetrical distribution of patients across groups over time in an attempt to match groups for evolving treatment strategies and diferent virus variants. Te date of ICU admission was categorized according to the COVID-19 wave (supplementary table 1). Waves 1 and 2 in Belgium were caused by the D614G variant, wave 3 by the alpha variant, and wave 4 by the delta variant. Tese virus variants difer in disease severity and consequently the mortality rate. Furthermore, in the course of the pandemic, we also adapted therapy strategy to several domains. After nearest neighbour calliper matching with a calliper of 0.2 [23], 24 patients without ECMO treatment acted as the matched control group. Comparisons between the groups were performed with the Student's t-tests for normally distributed data and with Mann-Whitney U-test for not normally distributed data. Categorical variables were analyzed with a Chi-Square test or, if appropriate, with Fisher's exact test. A p value <0.05 was considered statistically signifcant. All analyses were performed with SPPS version 27. Figure 1 Figure 2. Figure 3 presents a timeline with the mean duration of treatment phases in the COVID-19 ECMO and matched cohort, and no signifcant diferences were demonstrated in the timeline of onset of symptoms to hospitalization (14.29 ± 8.59 days vs. 10.79 ± 6.99 days, p � 0.21) and in the timeline of hospitalization to intubation (7.12 ± 6.90 days vs. 5.17 ± 4.59 days, p � 0.23) between COVID-19 ECMO patients and the matched cohort group. Te duration of mechanical ventilation was shorter in the matched cohort compared to the COVID-19 ECMO group (15.74 ± 11.71 vs. 24.9 ± 16.16, p � 0.01). All other treatment phases were equal in both groups.

STROBE fowchart depicting inclusion and exclusion is presented in
In total, 3 COVID-19 ECMO patients (12.5%) sufered from a CVA or stroke. Major bleeding occurred in 17 patients undergoing ECMO treatment (70.83%), and heparininduced thrombocytopenia was diagnosed in 2 patients after ECMO (8.33%). At last, 5 patients (20.83%) required a second ECMO run of which 3 patients were deceased.

Discussion
In this propensity-matched cohort study comparing the impact of ECMO and MVA on mortality and complications in severe COVID-19 pneumonia, 24 MVA patients were identifed as suitable matches for the 24 ECMO patients in terms of similar sex, P/F-ratio, and date of ICU admission. Te analysis also showed that the two groups were not statistically signifcantly diferent in terms of age, medical antecedents, clinical frailty score, BMI, and SOFA score on admission. Tis study demonstrated an almost three-fold risk of ICU mortality (p < 0.01) in patients supported with ECMO (45.8%) compared to MVA (16.67%). Te diference in three-month mortality between groups was even higher (50% vs. 16.67%; p � 0.02). Nonetheless, MVA patients were as expected exposed to less lung-protective ventilation settings, including higher peak inspiratory pressures (33.42 ± 8.52 mm; Hg vs. 24.74 ± 4.86 mmHg; p < 0.01). Also, maximal PEEP levels (13.52 ± 3.86 mmHg vs. 11.05 ± 2.20 mmHg; p � 0.01) were higher in the MVA group. ICU-LOS and hospital-LOS were comparable in the ECMO group and the matched cohort group.
Tese results are not consistent with the fndings of the frst recently published retrospective propensity-matched cohort study on this topic. Mustafa et al. reported a 3fold improvement in survival with ECMO, with a mortality rate of 25% with EMCO (n � 80) compared to 74% in the MVA cohort (n � 80) [9]. In latter study, data were collected from patients treated between March 1st, 2020, and June 9th, 2021 [9].
Te frst results of ECMO therapy in COVID-19 patients from small Chinese cohorts were discouraging, reporting a very high mortality [24,25]. A more recently published systematic review including 1896 COVID-19 patients supported with ECMO reported a pooled in-hospital mortality of 37.1% [26]. Te variation in study outcomes, however, was rather high with a reported heterogeneity I 2 of 52.8% [26]. Tis heterogeneity may be explained by diferences in the study population, sample size, and/or publication bias. A retrospective analysis of data of the Extracorporeal Life Support Organization Registry and COVID-19 Addendum including 4812 COVID-19 patients receiving ECMO in 2020 across 349 centers within 41 countries showed that inhospital mortality 90 days after ECMO initiation was between 36.9% and 58.9% [27]. More specifcally, earlyadopting centers that used ECMO therapy throughout 2020, reported a mortality rate of 36.9% in 1182 patients receiving ECMO on or before May 1st and 51.9% in 2824 patients after May 1 st [27]. Late-adopting centers that provided ECMO for COVID-19 only after May 1st, 2020, reported a mortality rate of 58.9% in 806 patients [27]. A large epidemiologic study reporting on hospital mortality in severe COVID-19 patients requiring admission into Belgian ICUs concluded that ECMO therapy is an independent predictor of in-hospital mortality (OR 8.83 (4.50-17.34)) [28]. Tis reported odds ratio is in line with the odds ratio reported in the present study (5.91 (1.55, 22.58)).
Two large systematic reviews assessing outcomes in COVID-19 patients supported with IMV alone reported overall mortality rates of 43% and 45% and a mortality rate of 36% in the Europe cohort [29,30]. A large international cohort study comparing the outcome of patients admitted to seven large ICUs in the Euregio Meuse-Rhine, one region across Belgium, Te Netherlands, and Germany found the lowest mortality rate in the Belgian subgroup (i.e., 22%, 42%, and 44%, respectively) [3]. Te mortality rate in the mechanically ventilated subcohort was 29%, 45%, and 44%, respectively [3].  Tus, it has to be emphasized that compared to the literature, Mustafa et al. reported a very low mortality rate with ECMO and a very high mortality rate with MVA [9]. In contrast, the present study found a mortality rate with ECMO well in the range of values found in the literature but a mortality rate in the matched MVA cohort at the very low end of values reported in the literature. Tese conficting results may be explained by a combination of selection bias, intervention bias, and the use of diferent concomitant treatment strategies. First, despite the application of propensity score matching in both studies, the selection of controls and cases is performed in diferent ways. In this study, patients were already selected for ECMO candidacy at ICU admission resulting in rather homogeneous control and case groups in terms of age, medical history, and clinical frailty. Tis selection procedure may also partially explain the low observed mortality rate in the matched cohort group. Te mortality rate of the patients not selected for ECMO candidacy in our study was indeed much higher, up to 41.4% in the mechanically ventilated subcohort. In contrast, in the study of Mustafa et al., the selection of controls was performed in a post hoc manner, based on age, ventilatory settings, arterial blood gas results, and presence of severe chronic organ dysfunction or acute multiorgan failure [9].
Tis selection method may have been associated with a high probability of a selection bias, favoring the use of ECMO in younger patients or those with fewer comorbidities. Second, applied ECMO support strategies also difer signifcantly between studies. In this study, 23 patients were treated with VV ECMO and only one patient with proven right ventricle (RV) failure with VA ECMO. In contrast, Mustafa et al. utilized single access, dual-stage right atrium to the pulmonary artery cannula resulting in additive right heart support in all patients [9]. However, relatively new and infrequently used, the single access dual-stage right atrium to the pulmonary artery cannula indeed classifes as a right ventricular assist device and has proven beneft for unloading of the failing right ventricle in various clinical settings [31,32]. Tis may partially explain the favorable outcomes in the ECMO group of the latter study since numerous studies have demonstrated that COVID-19 infection and ARDS are independent promotors of RV failure [33,34]. Tis hypothesis is supported by the fndings of another retrospective study that RVAD support at the time of ECMO initiation results in higher in-hospital and 30 days survival versus IMV in specially selected patients with severe COVID-19 ARDS [35]. Another potentially benefcial difference in ECMO strategy in the latter study is the extubation and mobilization of patients while on ECMO [9]. Tird, several treatment strategies applied at the JESSA hospital may also explain the low observed mortality rate in the non-ECMO cohort of this study. Early during the frst wave, we observed a prevalence of DVT in more than 65% of the intubated and mechanically ventilated COVID-19 patients [20]. Terefore, we implemented a more aggressive thromboprophylaxis protocol including close to therapeutic LMWH dosing, individually tailored with routine anti-Xa measurements and systematically ultrasonography screening for DVT. A before-after study suggested a signifcant decrease in one-month mortality after the implementation of this more aggressive thromboprophylaxis protocol [11]. It might even be hypothesized that the low mortality rate in the ECMO cohort of Mustafa et al. is partially the result of a more aggressive anticoagulation strategy in this cohort. Conversely, the high mortality rate in the MVA cohort of the latter study might have been partially due to the utilization of a more conservative thromboprophylaxis strategy in this cohort. Tese hypotheses are supported by the results of a large RCT evaluating the efects of therapeutic LMWH versus standard prophylactic or intermediate-dose heparins for thromboprophylaxis [36]. Tis RCT concluded that therapeutic LMWH reduces major thromboembolism and death in high-risk hospitalized COVID-19 patients. In contrast, the observation that all patients who died in the ECMO group of the present study experienced major bleeding during their ICU stay may suggest that the applied anticoagulation strategy in this cohort was too aggressive. Another treatment strategy applied at the JESSA hospital to explain both the low observed mortality rate in the MVA group and the high observed mortality rate in the ECMO group of the present study might be the early adoption of awake-prone positioning. A recently published large multicenter RCT evaluating the efcacy of awake prone positioning concluded that this treatment strategy is safe and has a favorable efect on the primary composite outcome of  Critical Care Research and Practice intubation or death within 28 days of enrolment [37]. However, this randomized controlled trial was not able to detect a statistically signifcant diference in death between the aforementioned groups 37. It might be hypothesized that awake-prone positioning may have caused exhaustion of the most afected patients in the present study, eventually resulting in a worse outcome for ECMO patients. Finally, in contrast to the present study, Mustafa et al. did not include "date of ICU admission" or "wave" into the propensity score model increasing the likelihood of an asymmetrical distribution of patients across groups over time. An asymmetrical distribution results in an increased risk of both selection bias and treatment bias because patients infected with diferent virus variants and disease severity are compared to each other and because of changing therapy strategies over time. Subanalysis showed no association between BMI categories and ICU mortality in both ECMO and MVA groups. In contrast, ICU mortality seems to increase with increasing age (>60 years) with ECMO. Tese results echo those of the previous studies [38,39].
Tis study has several limitations. First, the retrospective single-center design with relatively low numbers of patients negatively impacts the generalizability of our fndings. Second, there is a potential impact of the increasing knowledge of pathophysiology and treatment options in COVID-19 over time, which leads to frequent changes in therapeutic strategies, creating a heterogeneous patient population. Tird, during the frst wave, national Belgian guidelines for admission to ICU and ECMO initiation became more stringent to secure sufcient ICU capacity, also causing heterogeneity in the patient population. Nonetheless, this high threshold for ECMO during the frst wave may have enhanced the selection of a well-matched cohort of COVID-19 patients treated with MVA.
In conclusion, we were not able to reproduce the positive results of ECMO therapy in the frst propensity-matched cohort study comparing ECMO and MVA in critically ill patients with COVID-19 pneumonia. In contrast, the results of the present propensity-matched cohort study suggest that ECMO therapy may be associated with an up to a three-fold increase in ICU mortality and three-month mortality compared to MVA despite the facilitation of lung-protective ventilation settings. Te results of our analysis and the conficting data in the literature demonstrate the need for randomized controlled multicenter clinical trials on the clinical impact of ECMO therapy in refractory COVID-19induced ARDS.

Data Availability
Due to the applicable privacy regulation (GDPR) and Good Clinical Practices (GCPs) legislation, the full underlying dataset supporting the study cannot be provided. Tis dataset contains potentially identifying information, for example, age, BMI, and comorbidities such as diabetes mellitus leading to a unique subject in the dataset. Terefore, descriptive statistics have been used for a general overview of our study population, and all other relevant information is provided in Table 1. Anonymized data are available on motivated request and can be sent to Prof. Dr. Björn Stessel; Stadsomvaart 11; 3500 Hasselt, Belgium; bjorn.stessel@jessazh.be; AND Jessa Ziekenhuis, Data Protection Ofcer (DPO); Stadsomvaart 11; 3500 Hasselt, Belgium; DPO@ jessazh.be.

Ethical Approval
Tis study was approved by the Ethical Committee of JESSA Hospital, Hasselt, Belgium, on 8 th September 2021 and registered on clinicaltrials.gov. (NCT05158816).

Consent
Written informed consent was waived in light of the urgent need to collect data in the ongoing pandemic.