This retrospective, multicenter observational cohort study was carried out in the Respiratory Intensive Care Units (RICUs) of the University Hospitals of Modena (Italy) and Bologna (Italy). The study has been approved by the regional Ethical Committee of Emilia Romagna (CE N. 453/2020 and 715/2020) and trial registered on clinicaltrial.gov (number: NCT04649658).

All patients aged 18 ≥ 80 year admitted to both RICUs for severe/critical COVID-19 pneumonia between March 1st and June 1st 2020 were selected. SARS-CoV2 infection was confirmed by PCR method with nasopharyngeal swab. COVID-19-related pneumonia was defined as severe by the presence of a respiratory rate (RR) ≥30 breaths per minute (bpm), peripheral blood oxygen saturation (SaO2) ≤93%, PaO2/FiO2 ratio <300 mmHg breathing room air, and lung infiltrates >50% of the lung filed early at admission,23,24 and critical according to criteria of either WHO25 and the National Health Commission of China24 for COVID-19-induced ARDS which aligns with the Berlin Definition.26

Patients with endotracheal intubation (ETI) performed within the first 24-h from admission, ceiling of escalation, do not intubate (DNI) order as expressed by patient’s will or upon clinical judgement, missing core data at medical record analysis (i.e. clinical characteristics at baseline, pronation data, type and time of ventilatory support, mortality, need for tracheostomy, length of RICU and hospital stay) were not considered for analysis.

Patients included were then divided into two groups: (1) those undergoing awake PP maneuvers (PP) in addition to standard care; (2) those performing standard care only (SC).

Standard of care was in agreement with the Italian Society of Infectious Diseases’ Guidelines (SIMIT)27 and started shortly after admission. This included:

• -

  oxygen supply and (NRS) to target SaO2 > 90%;

• -

  hydroxychloroquine (400 mg bis in die (BID) on day 1 followed by 200 mg BID on days 2–5 eventually adjusted for creatinine clearance estimated by a chronic kidney disease algorithm);

• -

  azithromycin (500 mg daily for 5 days) at physician’s discretion when suspecting a bacterial respiratory superinfection;

• -

  low molecular weight heparin for prophylaxis of deep vein thrombosis according to body weight and renal function unless counterndicated.

A proportion of patients received off-label treatment with Tocilizumab,28,29 a recombinant humanized monoclonal antibody of the IgG1 class directed against both the soluble and membrane-bound forms of the interleukin-6 (IL-6) receptor. NRS were adopted using an inspiratory fraction of delivered oxygen (FiO2) increased up to a target transcutaneous oxyhemoglobin saturation >90%. Settings of each NRS have been adjusted by the attending physician based on the continuous monitoring of the cardiorespiratory parameters and included:

• -

  Non-invasive ventilation (NIV) with patients connected through a conventional circuit with a sized oronasal mask (BluestarTM, KOO Medical Equipment, Shanghai, PRC) to a high-performance ventilator (GE Healthcare Engstrom CarestationTM, GE Healthcare, Finland or Mindray SynoVent E3, Mindray Medical, Italy) in pressure support pre-set mode. Positive end expiratory pressure (PEEP) was initially set at 8 cmH2O and subsequently fine-tuned according to clinical parameters and ventilator waveforms. Pressure support (PS) was set at 10 cmH2O, and then progressively modified, according to tidal volume, waveforms, and respiratory drive.

• -

  High Flow Nasal Cannulae (HFNC) with patients connected to a high flow device (Optiflow™and AIRVO™, Fisher & PaykelHealthcare Ltd, Auckland, New Zealand) deliveredoxygen through different sized nasal cannulae. Flow delivery was initially set at 60 L/min and temperature at 37 °C then adjusted according to the patient’s tolerance.

• -

  Continuous Positive Airway Pressure (CPAP) with patients were connected through a helmet interface (Helmet Starmed, Intersurgical SpA, Mirandola, Italy) designed for pandemics. PEEP was initially set at 8 cmH2O and subsequently fine-tuned according to clinical parameter. This was obtained by connecting a blender system to the available oxygen source to achieve adequate FiO2 levels.

Once admitted to RICU a non-randomly subset of patients enrolled was subjected to early awake PP in addition to the standard care. In particular, PP treatment was assigned to consecutive patients in the charge of two specific physicians when they were on duty at the RICUs (namely, RT in Modena and VC in Bologna), given their experience in pronation maneuvers. Patients assigned to all the other doctors were instead submitted to standard care. PP treatment was started soon after admission. Patients eligible for PP were aged 18–80 years and had been admitted to RICU with indication for NRS therapy, whereas exclusion criteria for pronation were those previously reported by Coppo et al.19 Thos eligible were taught by RICU staff on how to achieve PP and then encouraged to maintain pronation for at least 3 -h before being helped back to supine in bed. However, they were free to resume their supine position or to maintain pronation at their own discretion. Number of daily PP sessions varied from a minimum of 1 to a maximum of 4 based on the physician’s judgment and/or the patient’s preference.

Chart review, medical record, and archived data collection wereconductedin each center. The following variables were then inserted into an electronic database: demographics, relevant comorbidities, clinical characteristics (arterial blood gases-PaO2/PaCO2/pH, PaO2/FiO2 ratio, respiratory rate-RR, blood lactate level, dyspnea grade by BORG scale, mean arterial pressure-MAP), laboratory tests (blood count, renal function, C-reactive protein-CRP, procalcitonin-PCT, D-dimer) on admission, type and duration of required NRS, rate and time of ETI, mortality, need for tracheostomy, length of RICU and hospital stay. Radiographic appearance on available computed tomography (CT) scan performed on admission was assessed by an expert radiologist blinded to the study purpose. Radiographic presentation of COVID-19 lung involvement was classified according to the number of lobes involved, the presence of bilateral abnormalities, the predominant distribution (diffuse, peripheral, patchy) and the main pattern (interstitial versus consolidative).

Primary purpose was to evaluate the impact of awake PP on ETI rate in patients with severe and critical COVID-19 pneumonia admitted to RICU to be assisted with NRS. In both hospitals the decision as to whether proceed to ETI was taken according to the best clinical practice by the attending staff. Criteria for ETI included: (a) PaO2/FiO2 ratio unchanged or worsened despite use of NRS, (b) need to protect airways due to neurological deterioration or massive secretions, (c) hemodynamic instability or major electrocardiographic abnormalities, (d) unchanged or worsened dyspnea and persistence of respiratory distress despite NRS (i.e. RR > 35 bpm, gasping for air, psychomotor agitation requiring sedation, abdominal paradox). Secondary scope was to compare time to ETI, mortality, NRS-free-days (i.e. days spent without HFNC, NIV, CPAP, or invasive mechanical ventilation at 1-month), tracheostomy, length of RICU and hospital stay between PP and SC groups. Furthermore, potential risks (among the epidemiological, clinical and radiographic factors) associated to ETI were investigated in the PP group.

Sample size calculation was performed assuming an estimated ETI rate of 70% for the study cohort1,30 and a presumed reduction by 40% in those receiving pronation (data derived from an exploratory analysis in 30 patients). Assuming α = 0.05, power 80% and an enrollment ratio of 1:2 (proportion of patients subjected to pronation or to standard care, respectively) a sample of 93 patients was considered sufficient to confidently perform analysis on the primary outcome.

To test whether baseline covariates were balanced and did not significantly affect treatment a post-hoc propensity score was allocated. We ran logistic regression using prone position as the dependent variable with all baseline features as covariates. Propensity scores were obtained by calculating the fitted value from the logistic model for each patient and then comparing, showing that allocation to treatment was not significantly affected by baseline condition (Fig. S1, supplementary material). Baseline characteristics were compared in PP and SC groups; continuous variables were expressed as median and interquartile ranges (IQR) and compared by t test and Wilcoxon-Mann–Whitney test, whereas categorical variables were reported as numbers and percentages (%) and compared by χ2 test or Fisher’s exact test.

The time to ETI analysis was performed with participants’ follow-up accrued from the date of admission until ETI. Time to ETI was compared using unweighted Kaplan–Meier curves and analyzed through a cumulative incidence function model using Fine-Grey competing risk model31 with baseline fixed covariates considering mortality as competing risk. The effect of pronation on ETI was shown by means of unadjusted and adjusted hazard ratio (HR) with 95%CI. Age, PaO2/FIO2 ratio, pH value and respiratory rate were identified as 4 key confounders, and then used for adjustment. In order to test the hypothesis that the difference between groups might vary according to the type of NRS, we formally included an interaction term in the Fine-Grey regression model. Results were then showed after categorizing the population into two strata using categorical separation. The association of the two different treatments with pre-specified secondary outcomes was further carried out through Fisher’s exact test and Wilcoxon–Mann–Whitney test.

In patients undergoing PP univariate and multivariate logistic regression were then performed to detect predictors of ETI among all of the available factors recorded at admission.

The time course of PaO2/FiO2 ratio before and after pronation according to ETI within the first 7 days from RICU admission was assessed through ANOVA analysis. Then a post-hoc Bonferroni-Dunn’s multiple test was used to perform the pairwise comparison of means for each group.

A two-sided test of less than 0.05 was considered statistically significant. Statistics were performed using SPSS version 25.0 (IBM Corp.New York, NY, USA) and Graphpad prism version 8.0 (Graphpad Software, Inc. La Jolla, Ca, USA) unless otherwise indicated.

One-hundred-fourteen patients were included 1:2 in the analysis (38 PP: 76 SC) among all those patients diagnosed with severe and critical COVID-19 and referred to the two RICUs over the period considered. Study chart is shown in Fig. S2 (Supplementary material).

Epidemiological, clinical and respiratory characteristics are presented in Table 1. The vast majority of patients were male (70%) and more than a third of them presented ARDS (37%), while the median PaO2/FIO2 ratio was 149 (78–232) mmHg. Time from disease onset to admission was comparable between the two groups (median 8 [4–12] days for PP and 9 [4–13] for SC, p = 0.5).

Patients in the PP group came out significantly younger than in SC (61 VS 70 years old, p = 0.03), while no differences were observed in terms of severity scores, comorbidities and biochemical markers. PP patients showed a worse PaO2/FIO2 ratio, higher RR and pH value at baseline. No inter-groups difference was observed in the received standard treatment, in either drugs or NRS.

No adverse events were reported when proning maneuvers were applied to these patients.

Overall ETI rate was 32.5%; 7 (18%) and 30 (39.5%) patients were subjected to MV in PP and SC groups respectively. Time to ETI did not differ between groups (5 [3–5] days and 4 [3–5] days for PP and SC, p = 0.7). Unadjusted Kaplan–Meier estimates (supplementary Figure 3, panel A) and Cox regression analysis showed the beneficial effect of PP compared with SC on ETI (HR = 0.45 95CI [0.2−0.92], p = 0.02). After adjusting for the key confounders, results again confirmed the group difference (HR = 0.59 95CI [0.3−0.94], p = 0.04); however, the stratified analyses showed that this difference varied according to the use of NRS even after adjusting confounders (Table 2).In particular, awake PP significantly reduced the risk of ETI in patients undergoing HFNC, but not NIV or CPAP.

Tracheostomy and mortality rate were similar (p = 0.4 and p = 0.4) between PP and SC; Kaplan–Meier curve analysis did not show any difference in 30-day survival between groups (Fig. S3, panel B). Number of days free from NRS was higher (20 and 15 days respectively, p = 0.03), and length of stay in RICU (10 vs 15 days, p = 0.02) and in hospital (20 vs 24 days, p = 0.03) were shorter in PP than in SC group at 30-day. See the overall results in Table 3.

Table 4 reports the results derived from the univariate and the multivariate analysis in patients proned. In the univariate analysis, those patients proned and intubated were older, had a higher prevalence of ARDS, maintained awake proning for less time/day, showed greater perceived dyspnea, mainly received NIV, and displayed a diffuse distribution of lung CT scan abnormalities. In the multivariate analysis, less time spent when proning, presence of ARDS, use of NIV, and diffuse pattern at chest CT scan were factors independently associated with ETI.

Interestingly, the daily time course of average PaO2/FIO2 ratio before and after proning was considerably different in PP patients intubated compared with those who were not (see Fig. S4).