FormalPara Take-home message

In this cohort study that included 4244 adult patients with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection admitted to the ICU, 80% received invasive mechanical ventilation. Mortality 90 days after ICU admission was 31% in the whole cohort and 37% in the subgroup of patients who received invasive mechanical ventilation on the day of ICU admission. Among these patients with early intubation, mortality increased with the severity of ARDS at ICU admission (30%, 34%, and 50% for mild, moderate, and severe ARDS, respectively).


From March to May 2020, Europe was massively affected by the coronavirus disease 2019 (COVID-19) outbreak. In that context, the REVA network [1] designed a specific registry (COVID-ICU), to prospectively collect characteristics, management, and outcomes of patients admitted to intensive care units (ICUs) for severe COVID-19 in France, Belgium, and Switzerland. In France, as of October 1st, 2020, 395,104 patients had been tested positive for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and 32,365 deaths have been associated with the disease. On April 8, 2020, the number of COVID-19 patients hospitalized in French ICUs peaked at 7148.

A few case-series [2,3,4,5] have described baseline characteristics and short-term mortality (up to 28-days after ICU admission) ranging from 26 to more than 50% in critically ill patients with COVID-19. However, recovery from severe COVID-19 often takes several weeks and a substantial number of these patients were still in the ICU or the hospital at the time their outcome was evaluated [4, 5]. Notably, 28-day mortality was 41% in the control care group of the RECOVERY trial, which showed that dexamethasone improved the survival of patients receiving invasive mechanical ventilation or oxygen at randomization [6].

The present study reports data of 4244 patients with laboratory-confirmed SARS-CoV-2 infection admitted to the ICU and for whom day-90 status was available. We also evaluated risk factors associated with 90-day mortality in these critically ill patients.


Study Design, Patients

COVID-ICU is a multi-center, prospective cohort study conducted in 149 ICUs from 138 centers, across three countries (France, Switzerland, and Belgium). Centers were invited to participate by public announcements and by the Reseau European de recherche en Ventilation Artificielle (REVA) network (70 centers were active members of this network). We included in the present report data from participating ICUs that had enrolled at least one patient with complete data on age and 90-day vital status. COVID-ICU received approval from the ethical committee of the French Intensive Care Society (CE-SRLF 20–23) in accordance with our local regulations. All patients or close relatives were informed that their data were included in the COVID-ICU cohort.

All consecutive patients over 16 years of age admitted to the participating ICU between February 25, 2020, and May 4, 2020, with laboratory-confirmed severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection were included. Laboratory confirmation for SARS-CoV-2 was defined as a positive result of real-time reverse transcriptase-polymerase chain reaction (RT-PCR) assay from either nasal or pharyngeal swabs, or lower respiratory tract aspirates [7]. Patients without laboratory-confirmed COVID-19 were not included, even if they presented with a typical radiological pattern.

Inclusions were stopped on May 4, 2020, after enrollment of 4643 patients admitted to the ICU. Survival status up to 90 days after ICU admission was obtained for 4244 of them.

Data collection

Day 1 was defined as the first day when the patient was in ICU at 10 am. Each day, the study investigators completed a standardized electronic case report form. Baseline information collected at ICU admission were: age, sex, body mass index (BMI), active smoking, Simplified Acute Physiology Score (SAPS) II score [8], Sequential Organ Failure Assessment (SOFA) [9], comorbidities, immunodeficiency (if present), clinical frailty scale [10], date of the first symptom, dates of the hospital and ICU admissions. The case report form prompted investigators to provide a daily-expanded data set including respiratory support devices (oxygen mask, high flow nasal cannula, or non-invasive ventilation), mechanical ventilation settings (positive end-expiratory pressure (PEEP), the fraction of inspired oxygen (FiO2), respiratory rate, tidal volume, plateau pressure, arterial blood gas, standard laboratory parameters, and adjuvant therapies for acute respiratory distress syndrome (ARDS) such as the use of continuous neuromuscular blockers, nitric oxide, prone position, corticosteroids, or extracorporeal membrane oxygenation until day-90. Driving pressure was defined as plateau pressure minus PEEP and mechanical power (J/min) was calculated as follows: Mechanical power (J/min) = 0.098 × tidal volume × respiratory rate × (peak pressure − 1/2 × driving pressure) [11]. If not specified, peak pressure was considered equal to plateau pressure. Ventilatory ratio was defined as (minute ventilation × PaCO2) − (predicted bodyweight × 100 × 37.5) [12].

ARDS severity, complications, and outcomes

ARDS was graded based on the Berlin definition for patients undergoing mechanical ventilation (invasive or non-invasive) on ICU day 1 [13]. Patients on nasal, mask or high-flow oxygen therapy were not included in this group. However, their day-1 PaO2/FiO2 was calculated by converting O2 flow to estimated FiO2 (see conversion tables in the supplement) [14]. ICU-complications and organ dysfunction included acute kidney failure requiring renal replacement therapy, thromboembolic complications (distal venous thrombosis or proven pulmonary embolism by either pulmonary CT angiography or cardiac echography), ventilator-associated pneumonia, and cardiac arrest. Clinical suspicion of ventilator-associated pneumonia was confirmed before antibiotics either by quantitative distal bronchoalveolar lavage cultures growing ≥ 104 cfu/mL, blind protected specimen brush distal growing ≥ 103 cfu/mL, or endotracheal aspirates growing ≥ 106 cfu/mL.

Patient outcomes included the date of liberation from mechanical ventilation, dates of ICU and hospital discharge, vital status at ICU and hospital discharge, and 28, 60, and 90 days after ICU admission.

Statistical analyses

Characteristics of patients were described as frequencies and percentages for categorical variables and as means and standard deviations or medians and interquartile ranges for continuous variables. Categorical variables were compared by Chi-square or Fisher’s exact test, and continuous variables were compared by Student’s t test or Wilcoxon’s rank-sum test. Kaplan–Meier overall survival curves until Day 90 were computed, and were compared using log-rank tests. The median length of stay in ICU and in hospital were also estimated using a Kaplan–Meier estimator to take into account patients that may be still in ICU at the time of the analysis.

Baseline risk factors of death at Day 90 were assessed within the whole cohort using univariate and multivariate cox regression. Baseline variables (i.e., obtained during the first 24 h in the ICU) included in the multivariate model were defined a priori, and no variable selection was performed (see the description of the statistical analysis plan in the Supplement). ICU admission dates were split into four calendar periods (i.e., before March, 15; from March 16 to 31; from April 1 to 15; and after April 16). Proportional hazard assumption was assessed by inspecting the scaled Shoenfeld residuals and Harrel’s test [15] (Table S4). Multiple imputations were used to replace missing values when appropriate (Figure S1–S2). Ten copies of the dataset were created with the missing values replaced by imputed values, based on observed data including outcomes and baseline characteristics of participants. Each dataset was then analyzed and the results from each dataset were pooled into a final result using Rubin’s rule [16]. Lastly, a sensitivity analysis using a Cox model stratified on the center variable was also performed. Hazard ratios and their 95% confidence interval were estimated. A p value < 0.05 was considered statistically significant. Statistical analyzes were conducted with R v3.5.1.


Participating ICUs and Patients Enrolled

Patients were included in 149 ICUs (71 [48%] university, 66 [44%] public regional, and 12 [8%] private, semi-private, or military hospitals, respectively) from 138 centers in three countries. The median (interquartile) number of ICU beds in these centers and these ICUs were 26 (18-55) and 20 (14-28), respectively. Fifty-six percent of the patients were recruited in Paris and the surrounding area (see Table S1-S3 in the Supplement for an extensive description of ICUs and center characteristics). Ninety-four percent of the centers reported having extended the number of ICU beds during the COVID-19 outbreak.

Of the 4643 patients enrolled on May 4, 2020, 399 were lost to follow-up at Day-90. Thereafter we describe the characteristics of the remaining 4244 patients with available day-90 vital status (Fig. 1).

Fig. 1
figure 1

Flowchart of patients screening and inclusion. ICU intensive care unit

There were 1085/4244 (26%) female patients (Table 1). At ICU admission, their median (interquartile) age, SAPS II, and SOFA scores were 63 (54–71) years, 37 (28–50), and 5 (3–8), respectively. The rate of obese (BMI ≥ 30 kg/m2) patients was 1607/3935 (41%). The most frequent comorbidities were hypertension 2018/4197 (48%), known diabetes 1167/4196 (28%), and immunocompromised status 314/4192 (7%). Median (IQR) time between first symptoms and ICU admission was 9 (6–12) days. Of note, only 176/4124 (4%) patients were active smokers and only 208/4116 (5%) had concomitant bacterial pneumonia at ICU admission.

Table 1 Demographic, clinical, and ventilatory support characteristics of 4244 patients according to their 90-day survival status

Ventilatory support, adjunctive therapies, and ARDS severity

On day-1, standard oxygen therapy, high flow oxygen, and non-invasive ventilation were applied to 1219/4157 (29%), 786/4096 (19%), and 230/4109 (6%) patients, respectively. The use of these modalities tended to increase over time (Table S5 and Figure S3). 2635/4175 (63%) were placed on invasive mechanical ventilation during the first 24 h, whereas in total 3376/4209 (80%) were intubated during their ICU stay. On the first day in ICU, median tidal volume, PEEP, plateau, driving pressures, and mechanical power were 6.1 (5.8–6.7) mL/kg, 12 (10–14) cmH2O, 24 (21–27) cmH2O, 13 (10–17) cmH2O, and 26.5 (18.6–34.9) J/min, respectively (Table 1). 1841/2560 (72%) patients required a FiO2 ≥ 50%, while 1371 (54%) received a PEEP ≥ 12 cmH2O.

Mild, moderate, and severe ARDS was reported in 539/2233 (24%), 1154/2233 (52%), and 540/2233 (24%) patients on mechanical ventilation (invasive or non-invasive) on ICU day 1, respectively (Table 2). Continuous neuromuscular blockade and prone position were used in 1966/2224 (88%), and 1556/2223 (70%) in these patients. Moderate and severe ARDS patients were more likely to receive these adjunct therapies, with a median number of 3 (IQR 2–6) prone positioning sessions per patient. Of note, 888/2224 (41%) of them received corticosteroids for a median of 5 (IQR 2–8) days. Lastly, 321/4187 (8%) patients were placed on extracorporeal membrane oxygenation (ECMO). Table S6 provides the use of adjunct therapies in the whole cohort of 4224 patients.

Table 2 Use of adjunct measures, organ dysfunction and major complications according to acute respiratory distress syndrome severity for patients on mechanical ventilation (invasive or non-invasive) on ICU day 1

ICU complications and organ support in patients intubated on ICU-day 1

Ventilator-associated pneumonia was diagnosed in 1209/2101 (58%) patients who were intubated on ICU day 1, whereas 623/2227 (28%) patients had an acute kidney failure requiring renal replacement therapy (Table 2). A venous thromboembolic complication was diagnosed in 373/2226 (17%) patients, of whom 207/2226 (9%) had proven pulmonary embolism.

Patient outcomes and predictors of 90-day mortality

Overall 90-day mortality was 31%. Within the first 7 days after ICU admission, 64 (12%, 95 confidence interval [CI], 9–15%) patients with mild and 183 (16%, 95 CI 14–18%) with moderate ARDS progressed to severe ARDS. In patients on mechanical ventilation (invasive or non-invasive) at ICU day one, 90-day mortality was 820/2233 (37%), and increased with the severity of ARDS at ICU admission (30%, 34% and 50% in mild, moderate, and severe ARDS patients, respectively) (Table 3 and Fig. 2). 90-day mortality was 292/1219 (24%), 202/786 (26%), and 96/230 (42%) in patients who received standard oxygen therapy, high flow oxygen, or non-invasive ventilation at day-1. Noticeably, 90-day mortality declined over time from 42 to 25% (p < 0.001) in the first and the last period, respectively (Table S5 and Figure S3). Of note, 90-day mortality was 36% in patients intubated during their ICU stay and 11% for those not intubated (see Table S7). The overall median durations of mechanical ventilation, ICU, and hospital stay for 90-day survivors were 13 (8–18), 21 (13–36), and 30 (20–48) days, respectively. Of note, these durations increased with the severity of the ARDS (Table 3).

Table 3 Outcome of patients on mechanical ventilation (invasive or non-invasive) according to Acute Respiratory Distress Syndrome Severity assessed the first day in the ICU
Fig. 2
figure 2

Kaplan–Meier survival estimates during the 90 days following ICU admission, according to A) Acute Respiratory Distress Syndrome Severity in patients on invasive mechanical ventilation or non-invasive ventilation at Day-1; B) age categories; C) ARDS severity progression within 7 days in patients with mild ARDS at Day-1*; D) ARDS severity progression within 7 days in patients with moderate ARDS at Day-1. *Only patients alive at day-7 were included in this analysis. ICU intensive care unit

Results of the multivariable analysis are reported in Table 4. After inspection of the proportional hazard assumption, a time-varying effect was introduced in the multivariate Cox model for four variables: body mass index, active smoking, renal component of the SOFA score, and lymphopenia. Thus, for these variables, two types of hazard ratio are reported, indicating the early effect (before 14 days of follow-up) or the late effect (after 15 days of follow-up) of the corresponding baseline characteristic on the risk of death, respectively.

Table 4 Predictive patient factors associated with 90-day mortality in critically ill adults with COVID-19

Non-survivors were older, and more frequently diabetic or immunocompromised than survivors. At ICU admission, they had a higher renal and hemodynamic SOFA component scores and lower PaO2/FiO2 ratio. Interestingly, they also had a shorter time since the onset of the first symptoms. Day-1 patients’ characteristics significantly associated with higher 90-day mortality identified by the Cox regression model were older age, known diabetes, class 2 and extreme obesity, immunodeficiency, higher renal and cardiovascular components of the SOFA score, lower PaO2/FiO2, lower pH, and a shorter time between first symptoms and ICU admission (Table 4). The same analysis re-run after multiple imputations of missing data (Table S8), and a sensitivity analysis introducing the center variable as a stratification variable in the multivariable model yielded similar results (Table S9). Kaplan–Meier survival estimates according to age, ICU admission period, the renal component of the SOFA score, the delay between the first symptoms and ICU admission, immunocompromised status, diabetes, severe lymphopenia, and static pulmonary compliance categories at day-1 are provided in Figs. 2b, S4-S10. Lastly, outcomes of patients who progressed from mild to moderate or severe ARDS and those who progressed from moderate to severe ARDS within the first week of mechanical ventilation are reported in Fig. 2c, d.


We report herein one of the largest prospective case-series of COVID-19 patients who required intensive care admission, with detailed information on their baseline characteristics, ARDS severity, and 90-day outcomes. Overall 90-day mortality was 31% and was higher in older and obese patients, diabetics, immunocompromised patients, and those who had multiple organ dysfunction at ICU admission. 90-day mortality rates were 30%, 34%, and 50%, in patients with mild, moderate, and severe ARDS who were on mechanical ventilation (invasive or non-invasive) on ICU day-1, respectively. Noticeably, mortality rates decreased over time during the study period, while ICU and hospital length of stay were substantially longer than in other cohorts of ARDS patients [17].

Acute respiratory failure was the main indication for ICU admission, with 80% of our COVID-19 patients requiring invasive mechanical ventilation which is consistent with the experience in Lombardy, Italy [2], where 88% of ICU patients were intubated. However, lower rates of intubation in ICU patients were reported in Wuhan, China by Wang et al. (47%), and Yang et al. (42%) [18, 19], and in Washington state, USA (71%) [20]. While early single-center reports in small groups of COVID patients reported well-preserved lung mechanics despite the severity of hypoxemia [21], more recent data [22] and our observations suggested that lung compliance and driving pressure were close to those of reported in classical ARDS. Mechanical power which is the energy delivered to the respiratory system over time during mechanical ventilation was very high in our patients with ARDS, reaching 26.5 (18.6–34.9) J/min, while a higher mortality risk has been suggested for patients with ARDS whose value exceeded 17 J/min [23]. In that context, the application of evidence-based ARDS care, such as lung-protective mechanical ventilation and proning are both warranted [24]. ECMO, which was used in 15% of severe ARDS in our cohort should be considered when these measures have failed [25].

28-day mortality was 39% in 257 critically-ill COVID-19 patients in New-York city, of whom 203 (79%) received invasive mechanical ventilation [4], 41% in patients on invasive mechanical ventilation included in the usual care group of the RECOVERY randomized trial [6] and > 50% in 733 Chinese patients admitted in the ICU [3]. Despite similar severity at baseline, day-28 mortality were 26% in the whole cohort and 30% in our patients intubated at day-1, a rate close to that reported in the large LUNG-SAFE study [17]. Different characteristics of patients admitted to ICUs and different degrees of stress on healthcare systems could explain these discrepancies [26]. Besides, we report a progressive decrease in 90-day mortality over the study period with a higher proportion of patients on high flow oxygen and non-invasive ventilation and a lower rate of intubation on ICU day-1 in the last period of the study (Table S5 and Figure S3). Similar findings have been reported by other groups [27] and might reflect better knowledge of the pathophysiology of the disease over time and less reluctance to use non-invasive oxygenation strategies. It should however be noticed that duration of invasive mechanical ventilation and length of ICU and hospital stays were substantially longer than those of in ARDS unrelated to COVID-19. For instance, ICU length of stay in patients surviving severe ARDS was 26 (13–43) days, compared to 14 (7–23) days in the LUNG-SAFE study [17]. These patients rapidly overwhelmed ICU’s capacity, forcing a major reorganization of ICU beds during the crisis [28].

Identifying the determinants of outcomes of critically ill patients with severe COVID-19 is crucial to optimize the use of ICU care and other hospital resources. Older age, obesity, diabetes, being immunocompromised, lower PaO2/FiO2 and higher hemodynamic and renal SOFA score at ICU admission were independently associated with 90-day mortality, highlighting the dismal impact of premorbid conditions and multiorgan damage on the outcomes of patients with the most severe forms of COVID-19 [3, 29]. The rate of patients with a BMI ≥ 30 kg/m2) was 41% in our cohort and unusually high compared to the prevalence of obesity in the French population [30]. More severe COVID-19 in obese patients may be explained by impairments in the adaptive immune response [31], cardiometabolic and thrombotic derangements [32], and alterations in lung function [33]. Obesity may also be a marker of poorer baseline health conditions since it is frequently associated with a lower socio-economic status [34]. As previously reported [35], a shorter time between first symptoms and ICU admission was also independently associated with increased mortality. Lastly, we and others [36, 37] observed an unusually high rate of thromboembolic complications, with 9% of proven pulmonary embolism. This rate may likely be higher since pulmonary CT angiography was not systematically performed in all patients. Diffuse vascular endothelium injury and intense activation of the coagulation cascade may explain this increased risk of venous thrombosis [38].

The major strength of this study is the detailed report of physiological, clinical features, ventilatory management, and 90-day outcomes of a large, multicenter series of 4244 critically ill COVID-19 patients. We acknowledge several limitations to our study. First, we conducted this cohort at a time where the national health system was extremely pressured with a need for a large number of ICU beds in some regions. Then, we cannot rule out that ICU admission policies and patients’ management were similar in all centers, although a sensitivity analysis introducing the center variable in the multivariable model found similar results. Second, testing was not standardized across sites, which might have led to misclassification. Third, some variables have missing data (as reported in the tables) due to a large number of patients included in a short period and intense clinical activity during the crisis. Indeed, our multivariable model included only 51% of the whole cohort of patients because of these missing data, which may explain, together with other residual confounders, the unanticipated lower mortality before 14 days of follow-up associated with baseline lymphopenia [39, 40]. However, this association was no longer statistically significant in the model with multiple imputations. Fourth, Grasselli et al. recently reported that high D-dimer concentration was significantly associated with mortality in COVID-19-related ARDS patients when associated with low values of static respiratory system compliance [22]. Unfortunately, we were unable to confirm that result in our multivariable model due to inconsistent collection of this data at ICU admission.


In this case series of 4244 critically ill patients with laboratory-confirmed COVID-19 admitted to our ICUs, overall 90-day mortality was 31% and decreased over time during the study period. Mortality was higher in older patients, immunocompromised, extreme obese, diabetics, those with a shorter delay between first symptoms and ICU admission, and those with extra-pulmonary organ dysfunction at ICU admission. Ninety-day mortality increased with the severity of ARDS from 30% in mild to 50% in severe ARDS. These information, together with the very long durations of mechanical ventilation and of ICU stay, which have contributed to the swamping of our ICU’s capacity, will be critical for the management of the second wave of the epidemic. Lastly, long-term follow-up is warranted to provide a complete description of the outcomes and potential sequelae associated with the most severe forms of COVID-19 requiring ICU treatment.