Background
Severe coronavirus disease 2019 (COVID-19) predominantly presents with the clinical picture of acute respiratory distress syndrome (ARDS) and a high likelihood of multiple organ failure and death [
1]. Venovenous (VV) extracorporeal membrane oxygenation (ECMO) is a known therapeutic option in life-threatening respiratory failure and may improve outcome in ARDS [
2]. Although there has been some doubt about the adequacy of ECMO in the context of limited ICU resources, the scientific community has advocated for the use of ECMO in patients with severe COVID-19 ARDS, and notes the crucial role of specialised high-volume centres [
3].
During the COVID-19 pandemic, intensive care capacities have repeatedly been exceeded in several regions around the world. From March until May 2020, restrictions placed on the Austrian population prevented the local healthcare system from reaching its capacity limits. However, the resurgence of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infections starting in September 2020 led to the widespread conversion of ordinary ICUs into COVID-19-ICUs to cope with the high influx of COVID-19 patients with respiratory failure. In times of high infection rates, accurate decision-making is vital when it comes to ideal patient allocation. This includes crucial considerations with respect to contraindications and limiting factors, including the duration of invasive mechanical ventilation (IMV) prior to ECMO start. ECMO initiation within 7 days following intubation is considered optimal [
4‐
6] as longer pre-ECMO IMV durations increase mortality in general ARDS populations [
7‐
10]. However, there is no clinically useful cut-off for the maximum antecedent period on IMV.
There is an urgent need for greater understanding of the risk factors influencing mortality in COVID-19 patients receiving ECMO in order to accurately allocate limited ICU and ECMO capacities and avoid triage situations. This retrospective study investigated the effect of pre-ECMO IMV duration on survival and risk factors for dismal outcome.
Methods
Study design and setting
Investigator-initiated, retrospective, observational cohort study. This investigation was carried out at the Medical University of Vienna, Austria. In order to meet demand, the Medical University of Vienna converted up to six ICUs into COVID-19 wards designed primarily to provide ECMO support. We included all adult patients treated with ECMO for confirmed COVID-19 in these six ICUs from January 2020 until May 2021. Most of our patients were transferred to our centre from hospitals with no ECMO capability. The observational period ran from ECMO start to ICU discharge at the Medical University of Vienna. This study was approved by the local Ethics Committee of the Medical University of Vienna (EK 2024/2020) and performed in accordance with the Declaration of Helsinki as well as the applicable laws and regulations currently in force. Study design as well as data handling and reporting followed the STROBE guidelines to ensure a maximum level of research quality.
ECMO management
With respect to the clinical consideration of ECMO, the consultants in charge followed the official Medical University of Vienna consensus recommendations [
11]. Details on ECMO evaluation, eligibility assessment, decision-making, implantation technique and management are described ibidem. In accordance with the opinions of international experts [
12,
13] and following conventional selection criteria, the use of ECMO in COVID-19-related ARDS was advocated as a last resort option. Thus ECMO was initiated when other strategies including lung protective ventilation, prone positioning, high positive end-expiratory pressure (PEEP), or neuromuscular blocking agents had failed, or in life-threatening hypoxia to avoid cardiopulmonary resuscitation. Our centre adopts a protective ventilation strategy in ARDS patients on ECMO, using a volume-limited controlled ventilation mode pursuing tidal volumes of ≤ 6 ml/kg ideal body weight (IBW), a driving pressure limited to 15 cm H
2O, and a target peak pressure of ≤ 30 cm H
2O. We titrate ECMO blood flow to at least 60% of the patient’s cardiac output to maintain peripheral saturation at between 88 and 92%.
Data sources
Patient identification and data collection were conducted using the patient data management system’s routine documentation (ICCA©, Philips, Amsterdam, Netherlands).
The documentation of clinical routine included patient demographic data, underlying disease, reason for hospital/ICU admission, severity of illness on admission expressed by APACHE II score, extent of organ dysfunction expressed by sequential organ failure (SOFA) score, severity of ARDS expressed by respiratory ECMO survival prediction (RESP) score prior to ECMO start, ICU length of stay (LOS), ICU survival, hospital survival, IMV duration prior to ECMO start, and ventilator settings during the course of admission.
Details of ECMO therapy, including duration, and reason for ECMO cessation (e.g., successful weaning, therapy withdrawal, lung transplantation [LTX], death) were extracted. Standard laboratory parameters were routinely documented on a daily basis. Baseline values were collected at the closest timepoint prior to ECMO start, except for one patient whose data were only available from day three onwards.
Statistical methods
Metric variables were reported using mean and standard deviation (SD) or median and interquartile range (IQR), and ICU survivors compared to non-survivors using t-tests or Mann–Whitney] U tests, according to their distribution, to identify potential risk factors for ICU death. Categorical variables are reported by absolute and relative frequencies, and ICU survivors compared to non-survivors using Chi-squared tests or Fisher’s exact tests, according to their distribution. The primary objective was to determine whether duration of IMV prior to ECMO start influenced ICU mortality. In order to assess the primary objective, a logistic regression model was fitted using ICU mortality as dependent variable, IMV duration prior to ECMO insertion as an independent variable, and age, SOFA score, and RESP score as confounders as these variables showed the greatest differences between survivors and non-survivors in univariate analyses. We used a Chi-squared test to compare the survival of the respective subgroups to address the commonly utilised cut-offs of 7 and 10 days of antecedent IMV duration. In addition, survival analysis was performed to investigate the effects of IMV duration on the hazard. To identify trends in pre-ECMO IMV duration and ICU mortality over time, we analysed a logistic regression model including age, a modified SOFA score (excluding PaO2/FiO2 ratio), a modified RESP score (excluding age and pre-ECMO IMV duration), and all comorbidities, performing stepwise selection while forcing pre-ECMO IMV to remain in the model. We considered p values < 0.05 statistically significant. P values from secondary and exploratory analyses serve only descriptive purposes, hence no multiplicity corrections were applied. Calculations were performed using R statistics software (version 4.0.5, The R Foundation for Statistical Computing, Vienna, Austria).
Discussion
In our patient population, the median duration of pre-ECMO IMV was 7.7 days in survivors and 6.8 days in non-survivors. Similar to other observations [
14‐
17], we found no correlation between pre-ECMO IMV duration and survival.
Although some data show the lowest mortality of COVID-19 patients when ECMO initiation takes place within the first three to four days following intubation [
18,
19], the current literature provides no clear cut-off for the maximum antecedent time on IMV. In non-COVID-19-associated ARDS, a duration > 7 days has been associated with increased mortality, which is why ECMO initiation, once indicated, should not be delayed [
7‐
9,
20,
21]. Exceeding 7 days from intubation to ECMO is also an integral component in well-established risk prediction scores such as RESP [
7] and PRESERVE [
8], and has therefore often been considered a relative contraindication for ECMO therapy and thus centre admission [
4‐
6,
22]. During the COVID-19 pandemic, some institutions have considered a limit of 10 days to be more appropriate [
23]. Against this background, we compared patients mechanically ventilated for < 7 (and < 10) days with those ventilated for ≥ 7 (and ≥ 10) days in sub-analyses and found no significant difference in ICU survival. It remains uncertain whether ECMO timing in COVID-19 patients should follow commonly utilised entry criteria as stated in the EOLIA trial and our own COVID-19 ECMO guidelines, including long IMV duration as a (relative) contraindication [
11,
24]. According to the current norms of practice, prolonged duration of IMV may lead to denial of ECMO therapy, especially in a pandemic context with significant resource constraints [
4,
22]. Our findings challenge the applicability of general ECMO entry criteria for COVID-19 patients and the role of pre-ECMO IMV duration of > 7 days as a relative contraindication in commonly utilised recommendations.
In our cohort, ICU survival was 59%. This is similar to previous outcomes for COVID-19 ARDS [
5,
14,
15] as well as for severe ARDS resulting from other causes and treated with ECMO [
2,
25]. Median duration of IMV before ECMO was 7.7 days, notably exceeding that reported by Barbaro et al. (2.7 to 4 days), Schmidt et al. (4 days), Lebreton et al. (5 days), and Diaz et al. (4 days) [
5,
6,
15,
26]. Time of IMV prior to ECMO start was ≥ 7 days in as many as 53 patients and ≥ 10 days in 35 patients, with a maximum of 42 days.
One explanation for the prolonged duration of IMV reported here may be the protracted and often complicated course of severe COVID-19 pneumonia itself [
27,
28], although other studies have reported shorter periods on IMV until ECMO was initiated, as mentioned above. However, conservative management had commonly reached its limits in the referring hospitals, potentially leading to late presentation at our institution. Furthermore, longer periods of IMV were regarded as a relative contraindication and therefore tended to be accepted by our consultants where the patient was otherwise eligible for ECMO. This approach could have led to a selection bias, by accepting less sick patients for ECMO treatment. However, pre-ECMO severity of illness expressed by APACHE II showed no differences between survivors and non-survivors. Also, median PaO
2/FiO
2 ratio of 74.2 at cannulation expressed profound ARDS severity, similar to other COVID-19 cohorts [
2,
14,
25], and did not differ significantly between survivors and non-survivors.
Interestingly, for both cut-offs (7 days and 10 days), the survival probability curve for patients with shorter pre-ECMO IMV duration is below the curve for patients with longer ventilation time, even though this difference is not found to be significant (Figs.
1 and
2). However, this finding is supported by the negative effect of longer IMV on the hazard in survival analysis. In combination with the non-significant effect of IMV duration on ICU mortality, this indicates that pre-ECMO IMV duration does not predict the risk of death in the ICU, but that non-survivors would die sooner if they had shorter IMV durations. In our experience, some patients deteriorate quickly, presenting ECMO indication within a few hours following intubation, while others tolerate IMV for prolonged periods. One explanation for our finding might be that some patients experienced less swift but still aggressive courses of ARDS, leading to slower deterioration and therefore delayed ECMO indication (and thus longer pre-ECMO IMV duration), but still with protracted death. However, this could also happen in the context of variable responses to other treatment cornerstones such as steroid use or prone positioning.
In our patient population, older age was associated with ICU mortality. Age is known to be one of the most important risk factors in COVID-19 patients, with [
5,
14] or without ECMO [
29]. Usually, patients receiving extracorporeal gas exchange represent a younger population. In our study, however, mean age was 56 years and thus higher than in previously described cohorts [
30], reflecting a more liberal attitude to accepting older patients for ECMO implantation. Indeed, our individual decision-making naturally includes an intuitive rating of the biological age, rather than strictly following the chronological age [
11]. However, recent observations underline the strong correlation between chronological age and risk of in-hospital mortality [
18]. Under pandemic pressure, patient-centred approaches should cautiously consider all aspects known to influence outcome, including both age and health condition, as patients with a high mortality risk commonly require extended ICU resources.
At baseline, RESP score was significantly lower and SOFA score significantly higher in non-survivors. SOFA score is designed to assist prediction of outcome in critically ill patients with organ failure. RESP score has been developed to predict hospital survival at the time of ECMO initiation. Our findings indicate that, even after eliminating components of the RESP score which represent the effects of age and pre-ECMO IMV duration, RESP score is informative for the prediction of ICU mortality in a model including age and pre-ECMO IMV duration separately.
LTX is not a recommended treatment option for COVID-19 ARDS, but may serve as an ultima ratio alternative in highly selected patients with irreversible lung damage [
31]. Remarkably, 15 patients in our population received LTX during ECMO therapy, of which 12 patients survived ICU. Indeed, the complexity and risks of post-LTX management may be equal to ongoing ECMO management. The therapy principle of transplantation, however, differs considerably from conventional management, which is why overall comparability may be limited.
For the purposes of generalisation, an additional regression model was fitted using a combined endpoint of either ICU death or lung transplantation (see Additional file
1: Tables S8 and S9) which corresponds to a worst case scenario in which all transplanted patients would have died without LTX. We could not see any effect of IMV on this composed event. Effects of age and SOFA score on prediction of the composed event were comparable to prediction of the event of ICU death alone.
Moreover, when patients receiving LTX were excluded from our statistical analysis, with respect to group differences the results were qualitatively the same throughout, indicating at least no major confounding effect within our population.
One major strength of our study is the high number of patients with longer pre-ECMO IMV durations at our centre compared to previous studies. This accounts for the novel data. Furthermore, our institution does not strictly adhere to predetermined time limits for ECMO support as the decision for therapy cessation is usually based upon individual factors and interdisciplinary discourse.
We acknowledge the following limitations to our study: firstly, the majority of our patients were transferred to our tertiary care centre from a variety of referring hospitals in which IMV had often already been initiated. Detailed information about respiratory management and quality of lung protection is therefore fragmentary. It should be stressed that the strategy and duration of non-invasive ventilation prior to intubation may affect the duration of subsequent IMV, possibly confounding our findings. Secondly, due to our retrospective study design, outcome evaluation allowed for a complete ICU but an incomplete in-hospital survival analysis, the result of missing post-ICU values for 14 patients. It should be emphasised that patients surviving ARDS and ECMO often suffer from relevant post-intensive care sequelae which impair health-related quality of life. In a (post)COVID-19 condition, an even broader range of symptoms may persist which impair daily life and possibly require prolonged hospital stay or rehabilitation [
32,
33]. Prospective evaluations with quality-adjusted life years as a patient-centred outcome measure are warranted to depict this highly relevant interval from discharge to recovery. Thirdly, the retrospective nature of our study also accounts for missing data noted in the respective tables. And fourthly, all patients were treated in a high-volume tertiary centre within different departments which may limit the ability to generalise our findings.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.