Evolving outcomes of extracorporeal membrane oxygenation support for severe COVID-19 ARDS in Sorbonne hospitals, Paris

All consecutive adults, with laboratory-confirmed SARS-Cov-2 infection admitted to Paris–Sorbonne University Hospital Network ICUs, who received venoarterial (VA)- or venovenous (VV)-ECMO for severe ARDS from March 8, 2020, to January 28, 2021, were included, and provided 90-day survival status. The characteristics and outcomes of the first 83 cohort patients were reported previously [2]. We chose to split that cohort into two periods, before and after July 1, 2020, respectively. That date corresponds to the publication of the RECOVERY trial [8], and the transition period between the first and second COVID-19 waves in France. ECMO support was provided in four ICUs at Pitié–Salpêtrière and Tenon Hospitals before July 1, and in one Pitié–Salpetrière Hospital ICU thereafter. The Sorbonne University Ethics Committee approved the study protocol (CER-SU-2020–46).

ECMO indications for COVID-19 patients are detailed elsewhere [2, 4] and did not change throughout the study period. Briefly, indication(s) for ECMO implantation were centralized and evaluated in staff meetings, including at least two intensivists. Severe ARDS patients eligible for ECMO had to fulfill EOLIA trial respiratory severity criteria [5], with pre-ECMO use of neuromuscular-blocking agents and prone-positioning strongly recommended. Similarly, ECMO contraindications were still age > 70 years, severe comorbidities, cardiac arrest, refractory multiorgan failure or Simplified Acute Physiology Score (SAPS) II [9] > 90, irreversible neurological injury and/or invasive mechanical ventilation for > 10 days. Once the indication was retained, the Pitié–Salpêtrière Mobile ECMO Retrieval Team, comprising a cardiovascular surgeon and a perfusionist, was sent at the patient’s bedside for ECMO cannulation and retrieval to our department, as described previously [10, 11].

Throughout the study period, femoral–jugular percutaneous cannulation under ultrasonography guidance with a large drainage femoral cannula (25–29Fr) was strongly recommended for VV-ECMO. Pump speed was adjusted to obtain > 90% arterial oxygen saturation. Optimal cannula positioning was verified by ultrasonography and chest X-ray. Because of frequent thromboembolic events on-ECMO, including massive pulmonary embolism [12], and based on our early experience [2], the targeted activated partial thromboplastin time (aPTT) for VV-ECMO anticoagulation with unfractionated heparin was 60–75 s or anti-Xa activity 0.3–0.5 IU/mL. The hemoglobin threshold for red-cell transfusion was 7–8 g/dL and platelet transfusions were discouraged except for severe thrombocytopenia (< 50 G/L) or thrombocytopenia < 100 G/L with bleeding. To enhance protection against ventilator-induced lung injury, ultra-protective lung ventilation on-ECMO was recommended [6], and early prone-positioning on-ECMO was encouraged in the absence of hemodynamic instability and contraindications [13]. Based on the first descriptions [1, 2, 14] of very long ECMO runs and mechanical ventilation of COVID-19 patients, our team decided to decrease on-ECMO neuromuscular blocking-agent use for these patients. Patients were assessed daily for possible ECMO-weaning using the EOLIA clinical and physiological criteria [5, 15]. On-ECMO tracheostomy was considered after having identified, before the procedure, that the patient would likely tolerate decreased sedation. All tracheostomy decisions were made after discussion within the medical team of the balance between risks and benefits of that procedure on-ECMO [16]. When VA-ECMO was initiated, the ipsilateral limb received percutaneous cannulation via an anterograde perfusion catheter.

Our first-wave, ECMO-treated COVID-19 patients’ data were reported previously [2]. Briefly, they included pre-ECMO demographic information; severity scores (i.e., SAPS II [9]; Sequential Organ-Failure Assessment (SOFA) score [17]; Respiratory Extracorporeal Membrane Oxygenation Survival Prediction (RESP) score [18]); rescue therapies; pre-ECMO ventilatory mechanics, including driving pressure (ΔP), mechanical power [19] and ventilatory ratio [20]; arterial blood-gas parameters, and routine laboratory values. Because we expected that pre-intubation management had changed over the study period, HFNO and NIV use and their durations were also noted.

Mechanical ventilation settings, arterial blood gases, adjuvant therapies on-ECMO and ECMO-related complications were recorded daily from day 1 to 7, then every 7 days until ECMO-day 60, ECMO-weaning or death, whichever occurred first. Major bleeding was defined as the need for ≥ 2 units of packed red blood cells for an obvious hemorrhagic event, necessitating a surgical or interventional procedure, an intracerebral hemorrhage or being fatal, while massive hemolysis was defined as plasma-free hemoglobin > 500 mg/L associated with clinical signs of hemolysis.

Lastly, specific COVID-19 treatments were recorded. We specified whether dexamethasone (6 mg/day) was initiated before or during ECMO, and if the patient received high-dose corticosteroids, defined as > 1 mg/kg/day of prednisone or equivalent since ARDS onset, during the ECMO run [21].

Patient outcomes on days 28, 40, 50, 60, 70, 80 or 90 post-ECMO implantation included the following endpoints: on-ECMO, in-ICU and weaned-off ECMO, alive and out of ICU or died. The time spent in each of those four states until the specified days were also calculated. Other outcomes included ICU- and ECMO-related complications.

To describe patients’ in-ICU trajectories over time, a multi-state model was used as in our previous study [2, 22]. Briefly, this framework considers that a patient can go through different states during follow-up. Herein, the starting time was the ECMO-initiation day, making on-ECMO the initial state for all patients, potentially followed by two intermediate states: in-ICU & weaned-off ECMO or alive & out of the ICU. Because patients could die at any time during follow-up, either in-ICU or after discharge, the death is the only final absorbing state (the final state that a patient can enter but once entered cannot be left). In this four-state model (Additional file 1), each box represents a state and each arrow represents possible transitions from one state to another.

After assessing patient status, participants who did not reach the final absorbing state were right-censored. A Cox model stratified on each possible transition state was fitted to estimate transition (from one state to another) and state-occupation (for each of the four states) probabilities over time; the percentages of patients occupying each possible state were represented simultaneously over time with a stacked probability plot and reported with their 95% confidence interval (CI) on days 28, 40, 50, 60, 70, 80 and 90 post-ECMO initiation. Another figure (Additional file 2) individually displays all possible transition probabilities from one state to another over time. Mean state-occupation times (i.e., the expected length of stay in each possible state of the multi-state model) were also reported at the same times. Finally, median on-ECMO duration and length of ICU stay were established. These analyses were computed separately for the first and second waves.

Pre-ECMO risk factors for 90-day mortality were assessed for the entire cohort using univariate and multivariate Cox regression models. The variables included in the multivariate model were defined a priori, without any variable selection. Multiple imputations were used to replace missing values when appropriate. Briefly, 10 copies of the dataset were created with the missing values replaced by imputed values, based on observed data including participants’ outcomes and pre-ECMO characteristics. Each dataset was then analyzed and the results from each dataset were pooled into a final result applying Rubin’s rule [23]. Hazard ratios and their 95% CIs were estimated.

Finally, adjusted Kaplan–Meier probabilities of survival were estimated from the multivariate Cox regression model. Each subject’s survival probability over time was estimated from the model, first considering all subjects included during the first wave, and second considering all subjects included during the second wave. Then, the survival probabilities were averaged across all individuals. Finally, unadjusted and adjusted Kaplan–Meier curves were plotted on the same figure.

All the analyses were computed at a two-sided α level of 5% with R software, version 4.0.3.

Eighty-eight patients were admitted before July 1, 2020, and 71 thereafter (Table 1). Briefly, patients admitted after July 1 were significantly older, while the numbers and types of failing organs at ECMO cannulation were comparable. The first COVID-19-symptoms-to-intubation interval was longer during the second period. Respective before and after HFNO (19% vs. 82%) and/or NIV (7% vs. 37%) use and their durations before intubation were significantly higher after July 1, while the intubation-to-ECMO interval was comparable for the two periods. At ECMO cannulation, ventilation parameters, respiratory mechanics and blood gases did not differ between groups. More than 90% of the patients had received neuromuscular blocking agents and undergone prone-positioning before ECMO during both periods, while inhaled nitric oxide or prostacyclin (35% vs. 52%, p = 0.032) and dexamethasone (18% vs. 82%, p < 0.001) were used significantly more frequently after July 1. Notably, d-dimer concentrations were significantly lower in patients admitted during the second period.

ECMO management during the two periods is described in Table 2 and Additional file 3. During the second period, more patients received airway pressure-release ventilation/bilevel-mode ventilation and two remained non-intubated, awake and on HFNO. Ventilation parameters were comparable during the two periods, with major decreases of driving pressure, static compliance, mechanical power and ventilatory ratio on ECMO-day 1. More than 80% of the patients were prone-positioned during ECMO, with comparable numbers of sessions during both periods, while a significantly lower percentage of patients received continuous neuromuscular blockade during the second period (94% vs. 27%, p < 0.001). After July 1, 60 (85%) patients received dexamethasone during the first 15 days on-ECMO and a significantly higher percentage of patients (15% vs. 37%, p = 0.001) received high-dose corticosteroids for non-resolving ARDS [21].

On April 28, 2021, complete 90-day follow-up was obtained for all patients. The estimated state-occupation probabilities (95% CI) of being on-ECMO, in-ICU & weaned-off ECMO, alive and out of ICU or dead 90 days post-ECMO initiation, respectively, were: 1% (0.2–8%), 3% (1–10%), 59% (49–69%) and 36% (27–47%) during the first period, and 3% (0.7–11%), 1% (0.2–10%), 48% (37–60%) and 48% (37–60%) for patients admitted after July 1 (Fig. 1 and Additional file 4). Kaplan–Meier estimates of 90-day survival were 64% and 52%, respectively, for the first and the second periods (log-rank test p = 0.108) (Fig. 2). After adjusting for confounders, patients treated after July 1 had a significantly higher probability of death by day 90 (HR 2.27, 95% CI 1.02–5.07).

Median (IQR) ECMO durations during the first and second periods, respectively, were 19 (10–35) and 18 (5–35) days (p = 0.949), and 22 (12–48) and 33 (8–62) days among 90-day survivors (p = 0.466). Similarly, ICU and hospital lengths of stay did not differ between periods (Additional file 5).

The multivariable Cox regression model identified the following patient characteristics as being significantly associated with higher 90-day mortality: being admitted to the ICU after July 1, older age and pre-ECMO SOFA cardiovascular component score ≥ 3 (Table 3). In addition, a higher pre-ECMO PCO₂ and longer ICU-admission-to-ECMO interval tended to be associated with higher 90-day mortality. Conversely, having received dexamethasone before ECMO implantation was a protective factor, whereas the SOFA renal component score, pre-ECMO driving pressure and PaO₂/FiO₂ ratio were not independently associated with higher risk of mortality. Multiple imputations of missing data yielded similar results but ICU admission after July 1 and dexamethasone treatment no longer reached statistical significance (Additional file 6).