Early Prone Positioning in Acute Respiratory Distress Syndrome Related to COVID-19: A Propensity Score Analysis from the Multicentric Cohort COVID-ICU network. The ProneCOVID Study

Christophe LE TERRIER University Hospitals Geneva: Hopitaux Universitaires Geneve https://orcid.org/0000-0002-5455-5576 Florian Sigaud CHU de Grenoble Alpes Hopital Sud: Centre Hospitalier Universitaire Grenoble Alpes Hopital Sud Said Lebbah APHP: Assistance Publique Hopitaux de Paris Luc Desmedt CHU Nantes: Centre Hospitalier Universitaire de Nantes David Hajage AP-HP: Assistance Publique Hopitaux de Paris Claude Guérin Groupement Hospitalier Édouard Herriot: Groupement Hospitalier Edouard Herriot Jérôme Pugin Geneve University Hospitals: Hopitaux Universitaires Geneve Steve Primmaz Geneve University Hospitals: Hopitaux Universitaires Geneve Nicolas Terzi (  nterzi@chu-grenoble.fr ) Grenoble Alpes University Hospital https://orcid.org/0000-0003-4036-6245


Introduction
Since 2020, the world has been facing a global threat due to the COVID-19, overwhelming hospitals and intensive care units (ICUs) as never before. To date, the World Health Organization has reported 158 millions con rmed COVID-19 cases and more than 3 millions of deaths (1). Patients infected by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and hospitalized for a severe pneumonia may develop acute respiratory distress syndrome (ARDS), which is associated with high mortality (2)(3)(4).
Therefore, an extensive burden brought upon the intensive care units (ICUs) to provide invasive mechanical ventilation and other advanced forms of life support (5).
Before the COVID-19 pandemic, the Proseva trial (6) demonstrated an improvement in survival from prone position (PP) used as cycles of more than 16 consecutive hrs in selected ARDS patients, i.e. those with a PaO2/FIO2 ratio < 150 mmHg after 12 to 24 hrs-stabilization period. Though experts recommended PP in this setting (7), in the daily practice the rate of use of PP was lower than expected (8). Since the beginning of the COVID-19 pandemia, the surviving sepsis campaign (SSC) recommended PP in COVID-19 presenting with ARDS (9), a treatment widely adopted even though the level of evidence was similar as before the pandemia (4,10). In this recommendation, no timing to start prone position was proposed.
Owing to the very large number of COVID-19 related ARDS treated with PP it was reported that an early application of PP (11) and the response to PP in terms of oxygenation (12,13) were both possibly associated with a better outcome. Even if some studies of patients report interesting results (11)(12)(13), the impact of early PP on mortality remains unclear in COVID-19 patients in the ICU.
The objective of present ancillary study was to analyze the use of early PP in the ICU management of ARDS patient due to COVID-19 and to evaluate the impact of an early PP on survival, as well as on respiratory system mechanics and oxygenation, using a large international cohort of COVID-19 ARDS patients (4).

Study design and patients
This study was an secondary analysis of the COVID-ICU study (4). COVID-ICU was a prospective, multicenter observational cohort study of 149 ICUs from 138 hospitals conducted across three European countries (France, Belgian and Switzerland). The ethical committees of Switzerland (BASEC #: 2020 − 00704) and the French Intensive Care Society  approved this study and all patients or relatives had given their consent to be included in the COVID-ICU cohort. It recruited 4,643 patients between February and May 2020 with 80% of patients receiving invasive mechanical ventilation during their ICU stay.
All consecutive patients over 16 year-old included from February 25, 2020, to May 4, 2020 in the COVID-ICU study with an available vital status at Day-90 were eligible. Patients who met the following criteria in the rst 24 hrs after admission were included: intubated and mechanically ventilated, PaO 2 /FiO 2 < 300 mmHg with PEEP > 5 cmH 2 O, and no therapeutic limitations. Laboratory con rmation for SARS-CoV-2 was de ned as a positive result of real-time reverse transcriptase-polymerase chain reaction (RT-PCR) assay from either nasal or pharyngeal swabs, and/or lower respiratory tract aspirates. Patients without laboratory-con rmed COVID-19 were not included, even if they presented with a typical radiological pattern.
Patients were classi ed according to the fact that they had been subjected to PP at Day-1 or later. Day-1 was de ned as the rst day in ICU at 10 am following the COVID-ICU study. All patients placed in PP during their rst day in ICU constituted the early PP group. All patients placed in PP after Day-1 or nonplaced in PP during their ICU stay were categorized in the non-early prone position group. Patients placed in PP later in their ICU course were included in the non-early proning group to reduce the potential for immortal time bias and to emulate an intention-to-treat strategy of a randomized trial.

Data collection
A standardized electronic case report form was completed each day at 10 am by the study investigators.
Baseline characteristics were collected at ICU admission: age, sex, body mass index (BMI), active smoking, Simpli ed Acute Physiology Score (SAPS) II score, Sequential Organ Failure Assessment (SOFA), treated hypertension, diabetes, long term corticosteroids, immunode ciency, Clinical Frailty Scale, the date of the rst symptom, and dates of the hospital and ICU admissions. All investigators were asked to provide the lowest arterial partial pressure of oxygen (PaO 2 ) during last 24 hrs and fraction of oxygen inspired (FiO 2 ) to calculate PaO 2 /FiO 2 ratio, and categorized according to the ARDS Berlin de nition (14).
Static compliance was de ned by dividing the tidal volume by the driving pressure. The driving pressure was calculated by subtracting plateau pressure from positive end-expiratory pressure (PEEP). All biological data were collected at ICU admission. Proved concurrent bacterial pneumonia was de ned by a positive bacterial culture at ICU admission in either a bronchoalveolar lavage sample, or in a blind protected specimen brush distal, or in endotracheal aspirates. The main outcome was Day-60 survival.
Secondary outcomes included Day-28 and Day-90 mortality, ventilator free-days until Day-28, extracorporeal membrane oxygenation (ECMO) requirement, extracorporeal CO 2 removal (ECCO 2 R) requirement, and inhaled nitric oxide. The ventilatory free days were computed as the number of days that a patient was alive and free of invasive ventilation, calculated from ICU admission until Day-28. Patients who died before Day-28 or received invasive ventilation for more than 28 days were considered to have 0 ventilator-free days (15). The static compliance, the SOFA score and the PaO 2 /FiO 2 ratio were also evaluated at Day-3, Day-5, Day-7 as secondary outcomes.

Statistical analysis
Characteristics of patients were described as counts and percentages for categorical variables, and as mean and standard deviation or median and interquartile range for quantitative variables. Categorical variables were compared by Chi-square or Fisher's exact test, and quantitative variables were compared by Student's t-test or Wilcoxon's rank-sum test. Kaplan-Meier overall survival curves until Day-28, Day-60 and Day-90 were computed.
The primary endpoint was the Day-60 survival according to prone positioning at Day-1 of ICU stay. To assess prone positioning at Day-1 effect on Day-60 survival, we used a Cox proportional hazard model weighted on inverse probability of treatment weighting (IPTW) using propensity score (PS) de ned as the predictive probability of prone positioning conditional on measured baseline covariates (16). The variables used to estimate propensity score were: age, gender, clinical frailty scale, SOFA cardiovascular, SOFA renal, SOFA coagulation, SAPS II score, immunodepression, long-term corticosteroids, treated hypertension, diabetes, BMI, delay between rst symptoms and ICU admission, bacterial coinfection, ICU admission period (March 29 or after vs. March 28 or before), PaO2/FiO2 ratio and static compliance. A multivariate logistic regression model was performed to estimate the PS for each patient. To assess the balance of measured covariates between treatment groups, we used the standardized mean differences before and after PS weighting (17). Then, a Cox proportional hazard model weighted on IPTW was performed to estimate the average treatment effect in the entire eligible population (16). Hazard ratio and its 95% con dence interval were then estimated for the Day-60 mortality associated with prone positioning at Day-1. This analysis was performed on the complete cases data set, and a sensitivity analysis was performed using multiple imputations due to missing data. Imputation method, missing data were realized according to Vesin et al. (18). Proportional hazard assumption was assessed by inspecting the scaled Schoenfeld residuals and Harrel's test (19). Multicolinearity was checked using variance in ation factor.
All analyses were performed at a two-sided α level of 5% and conducted with R version 3.5.1 (R Foundation for Statistical Computing, Vienna, Austria).

Outcomes
In unadjusted analysis, mortality at Day-28, Day-60 and Day-90 were 30.5%, 35.4% and 35.9% respectively in the complete cohort study. Mortality was signi catively lower in the non-early PP group compared to the early PP group as shown in Table 2. More patients needed adjunctive therapies (ECMO, ECCO 2 R, inhaled nitric oxide) in the early PP group. The static compliance, the PaO 2 /FiO 2 ratio and the SOFA score at Day-3, Day-5 and Day-7 were worse in the early PP group. In the whole cohort, ventilatory parameters did not improve during the rst seven days after ICU admission. After propensity score adjustment, results were analyzed in both complete case analysis including 944 patients and in multiple imputation analysis with all baseline population of 2137 patients. Baseline characteristics before and after weighted-propensity score analysis are provided in additional  Fig. 3. Mortality at Day-28 and Day-90 were also similar between the two study groups after weighted-propensity score analysis.
In the subgroups of ARDS patients according to their PaO 2 /FiO 2 more or less than 150 at Day-1, mortality was higher in patients with PaO 2 /FiO 2 less than 150 mmHg (Table 3). Among the 1,504 patients who received prone positioning during their ICU stay, an early PP was not associated with a reduction of mortality nor an increase in ventilator-free-days up to Day-28. Among patients who subjected to PP at Day-1, a duration of PP session greater than 16 hrs was associated with a reduction of mortality at Day-28, 60 and 90.

Discussion
In this secondary analysis of a multicenter cohort study, our results show that PP was widely used across European ICUs during the COVID-19 pandemic, with 70% of patients intubated at ICU admission placed in prone position during their ICU stay. This rate contrasts with the results of the Lung Safe study and Apronet studies published before this pandemic, reporting less than 15% use of PP in ARDS of all-causes worldwide (8,20). Interestingly, our study highlights that prone positioning was not always used according to international guidelines (7,21). As a result, a large proportion of patients (37%) was placed in PP despite a PaO 2 /FiO 2 ratio higher than 150 mmHg. In addition, approximately 50% of patients were not placed in PP at Day-1 despite PaO 2 /FiO 2 ratio lower than 150 mmHg. Those ndings are consistent with results of previous studies (11,12). In a recent observational study, Mathews et al. reported that 44% of intubated patients with a PaO 2 /FiO 2 ratio less than 100 mmHg were not placed in PP during the rst 2 days, and only 30% of patients experienced proning during their ICU stay (11). In a large cohort study of more than 1000 patients, 21% of patients were not placed in PP despite a PaO 2 /FiO 2 ratio of less than 100 mmHg (12). Those results highlight the di culty to recognize ARDS and to properly apply international guidelines. Higher number of ICU beds, higher number of patients per physician or per nurse have previously been associated with a more di cult ARDS recognition and a lower use of prone positioning (20). The intervention of prone positioning in intubated patient requiring experimented staff to do it safely. Work overload, the deterioration of work conditions and the hiring of unexperimented staff associated with this pandemic (22,23) may have contributed to an under-recognition of ARDS and may explain why patients had not been placed in PP or placed in PP disregarding international guidelines.
Our study failed to demonstrate an improvement of survival in intubated patients receiving an early PP at Day-1 compared to non-early PP. Our ndings therefore contrast to those reported in another study in mechanically ventilated patients, in which early prone positioning in the rst two days of ICU admission was associated with a survival bene t in COVID-19 related ARDS (11). Several reasons may explain these discrepancies. First, de nition of treatment group was different between studies. In our study, treatment groups were de ned according to their PP status at Day-1 and not according to their PP status in the rst 48h after admission. In order to respect the validity of the propensity score using, our study was designed to analyze a potential survival bene t of prone positioning during the rst 24 hrs of ICU admission.
Although, the median delay between ICU admission and the rst prone positioning in the non-early PP group was 3 days, we could have failed to demonstrate a bene t because approximately 25 % of patients in this group had been nally placed in PP during Day-2. Those patients would have been referred as PP group in Mathews et al. study (11). Consequently, our results suggest no additional outcomes' improvement supporting very early PP during the rst 24 hrs of ICU admission. Second, our study enrolled all intubated ARDS patients and more than a third of patients placed in PP had a PaO 2 /FiO 2 ratio higher than 150 mmHg. The Proseva trial showed survival bene t with PP in moderate to severe patients with a PaO 2 /FiO 2 ratio less than 150 mmHg (6). Even if PP is supposed to limit the extent of lung injuries induced by ventilation in ARDS patients with various degrees of severity, the potential survival bene t in patients with PaO 2 /FiO 2 ratio higher than 150 mmHg has not been demonstrated and remains unclear mainly due to under-powered previous studies (24). Third, a large proportion of patients in the early PP group were placed in PP for less than 16 hrs in contrast to the Proseva trial showing a bene t in patient placed two time in prone position for at least 16 hrs during the rst two days (6). Similar to previous studies (25,26), the short duration of PP session could also explain the absence of bene t of PP observed in the early PP group. Finally, 660 patients were proned after 48 hrs of ICU admission, representing 43.8% of all proned patients in our cohort, and Guerin et al. found a survival bene t when using prone positioning early after endotracheal intubation (within 48 hrs) (6). In Mathews et al.'s study a smaller proportion of patients (19.5%) was initiated on proning after 48 hrs of ICU admission (11), which might have contributed to greater difference in patient's care between groups and thus impact mortality.
However, impact of timing of prone sessions initiation after endotracheal intubation has not been speci cally studied yet and is scarcely described in other randomized control trials assessing proning in ARDS (27)(28)(29).
Prone position has been shown to improve blood oxygenation by homogenizing the distribution of pulmonary ventilation/perfusion ratios; preventing ventilator induced lung injury by homogenizing the strain to lung tissue associated with mechanical ventilation on in amed alveoli, and preserving systemic hemodynamics, particularly right ventricular function (30). However, the clear response to the prone position has remained non-de ned. Our results show that patients placed in PP at Day-1 did not improve their ventilatory parameters, including the static compliance and oxygenation during their ICU stay at least until Day-7. In a large cohort of intubated COVID-19 patients, Langer et al. found that prone positioning was associated with immediate oxygenation improvement without any increase of respiratory system compliance (12). The lack of oxygenation improvement in our study could be due to the timing of assessment of oxygenation. Indeed, we recorded blood gases results daily independently of patients proning status at that time and did not study blood gases evolution during and just after proning. This could be in line with results reported by Langer et al. showing a trend toward worsening of oxygenation after re-supination (12). Our results considering the lack of improvement of static compliance are consistent with those of Langer et al. contrasting data on non-COVID-19 related ARDS which showed a reduction of driving pressure and plateau pressure when placed in prone position, suggesting better static compliance (30). This difference of effect of PP on respiratory mechanics between COVID-19 and non-COVID-19 related ARDS possibly highlight different pathophysiologies (31). Those lack of ventilatory parameters improvements could explain why the median duration of invasive mechanical ventilation in ARDS COVID-19 patients is approximately 12-13 days, longer that previously reported in all-causes ARDS patients included in Lung safe study (4,19). It might therefore also be possible that the follow-up of seven days in our study did not allow us to show a potential ventilatory parameters bene ts of prone position due to the short time of the follow-up.
This study has some limitations. First, it is not a randomized controlled study. We used however a propensity score adjusting on potential confounders. Second, despite of the propensity score weighting adjustment, it might be possible that patients in the early PP group were more severe at ICU admission and required a prone positioning earlier than patients in the non-early PP group, leading to confusion bias. Third, our study design did not allow us to analyze outcomes in patients respecting the PP status in the rst 48 hrs and after stabilization according the Proseva trial protocol, but only depending on the PP status at Day-1, in order to carefully respect the methodology of propensity score analysis. Fourth, some patients required up to 20 prone sessions leading to potential complications. Unfortunately, those data were not collected in this study.

Conclusions
Our results suggest that ICUs across European countries have largely adopted prone positioning in ARDS patients due to COVID-19 regardless of their severity. In this study, early prone positioning initiated during the rst day of ICU admission did not confer a survival bene t for patients requiring invasive mechanical ventilation, but prone sessions of more than 16 hrs seemed to be associated with better outcome. Further studies are needed to identify subgroups of patients with COVID-19 related ARDS who might bene t from early prone positioning. Authors' contributions: CLT and NT had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Concept and design: CLT, SP and NT. All authors interpreted the data and critically revised the manuscript for important intellectual content and gave approval for the nal version to be published. All authors agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Abbreviations
The manuscript's guarantors (CLT and NT) a rm that the manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned have been explained.
Funding: This study was funded by the AP-HP Foundation and its donators though the program "Alliance Tous Unis Contre le Virus", by the Clinical Research and Development Department, by the French Ministry of Health and by the foundation of the University hospitals of Geneva, Geneva, Switzerland.
The Reseau European de recherche en Ventilation Arti cielle (REVA) network received a 75,000 € research grant from Air Liquide Healthcare.
Sponsor: The sponsor was Assistance Publique Hôpitaux de Paris (AP-HP). Role of the funder: The funder had no role in the design and conduct of the study, collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Availability of data and materials: The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations
Ethics approval and consent to participate: All patients or close relatives were informed that their data were included in the COVID-ICU cohort. Human research ethics committee approval for the study were the ethical committee of Geneva (BASEC #: 2020-00704) and the ethical committee of the French Intensive Care Society (CE-SRLF 20-23) following our local regulations.
Consent for publication: All patients or close relatives were informed that their data might be published.
Competing interests: All authors declare no con icts of interest.