Study design and setting
The study is a multicenter prospective observational case–control study, performed in 2 intensive care units located in university hospitals, and was conducted in accordance with the amended declaration of Helsinki. The study was approved by a local independent ethics committee (Comité Scientifique et Ethique des Hospices Civils de Lyon, 20_194) and complied to the STROBE criteria for observational studies [11]. Patients were enrolled between November 1, 2020, and June 16, 2021, in center#1 and between November 1, 2020, and December 31, 2021, in center#2. Consent for data utilization was sought from the patients or their representative, and follow-up lasted 90 days. The primary endpoint of the study was the amount of recruitable lung between PEEP 5 and 15 cmH2O on CT.
Patients and protocol
Eligible participants were ARDS patients [12] aged 18 years or older, under invasive mechanical ventilation, who had a COVID-19 pneumonia with a positive SARS-CoV-2 reverse transcription polymerase chain reaction, and an indication for CT.
Exclusion criteria were ARDS onset > 72 h in non-ECMO patients, ECMO onset > 72 h, contra-indication to transport to the imaging facility (ratio of oxygen arterial partial pressure to inspired oxygen fraction (PaO2/FiO2) < 60 Torr, mean arterial pressure < 65 mmHg, or intracranial hypertension), chronic obstructive pulmonary disease, pneumothorax or bronchopleural fistula, previous inclusion in the present study, the presence of intrathoracic metallic devices, pregnancy, patient under a legal protective measure, and refusal to participate by patient and/or relative.
Non-ECMO patients were ventilated with a tidal volume (VT) of 4 to 6 mL.kg−1 of predicted body weight (PBW) to keep plateau pressure (PPlat,rs) below 30 cmH2O, with recommendation to use a PEEP-FiO2 table to adjust PEEP [13]. ECMO patients were ventilated with a VT of 1 mL.kg−1 PBW, with PEEP adjusted to target a PPlat,rs approximating 20–22 cmH2O.
Respiratory measurements and arterial blood gas analysis were performed at least 1 h after adjustment of ventilatory settings. Patients were then transferred to the imaging facility using a transport ventilator (MONNAL T60—Air Liquide Medical Systems, Antony, France) with unchanged ventilatory settings. The endotracheal tube was transiently occluded with a Kocher clamp during ventilator change to avoid derecruitment.
Data analysis
Total PEEP (PEEPtot,rs) and PPlat,rs were measured at the end of 3-s end-expiratory and end-inspiratory pauses. Airway driving pressure (ΔPrs) was computed as PPlat,rs minus PEEPtot,rs. Elastance of the respiratory system was computed as ΔPrs divided by VT.
Low-dose CT acquisitions were performed in the supine position with an iCT 256 Ingenuity CT (Philips Healthcare, Eindhoven, The Netherlands), or a GE Optima CT scan (GE Medical Systems, Milwaukee, USA) using the following settings: voltage 140 kVP, slice thickness 1 mm, and matrix size 512 × 512. Field of view, pixel size, and mAs were adapted for each patient. Lung scanning was performed from apex to base during end-expiratory and end-inspiratory pauses at the PEEP level set by the clinician (CTExpi-Inspi), and during end-expiratory pauses at PEEP 15 and 5 cmH2O (CTPEEP5-15). Lack of respiratory efforts during the pauses was visually checked on the ventilator pressure–time curves. Image reconstruction was performed using a smooth filter (kernel B). The lungs were interactively segmented with a CreaTools-based software [14], excluding pleural effusions, hilar and mediastinal structures. Segmented lung volumes were analyzed using MATLAB (MathWorks, Natick, MA).
Voxel tissue and gas fraction were computed as previously described [15]. Tissue and gas volumes were computed as the product of their respective fractions times voxel volume times number of voxels in segmented lung volume, respectively.
Lung parenchyma was classified into four compartments, according to CT number: non-inflated (density between + 100 and − 100 Hounsfield units (HU)), poorly inflated (density between − 101 and − 500 HU), normally inflated (density between − 501 and − 900 HU), and hyperinflated tissue (density ≤ − 901 HU).
Total lung weight and weight of each compartment were estimated using lung tissue volume, assuming a tissue density of 1 g.mL−1 [16].
VT was assessed on CT (VTCT) by subtracting the volume of gas at end-inspiration and at end-expiration in segmented lungs.
The amount of recruitable lung between PEEP 5 and 15 cmH2O (∆PEEP5-15-induced recruitment) was computed as the weight of the non-inflated compartment at PEEP 5 cmH2O minus its weight at PEEP 15 cmH2O, and standardized to total lung weight.
Tidal recruitment of the non-inflated compartment was defined as the weight of the non-inflated compartment at end-expiration minus its weight at end-inspiration [17], and standardized to total lung weight.
Change in lung aerated volume induced by PEEP change from 5 to 15 cmH2O (PEEPvolume) was computed as the difference in the total volume of gas within the lungs between PEEP 15 and 5 cmH2O.
The hyperinflation-to-recruitment ratio was computed as the difference between hyperinflated compartment total volume at PEEP 15 minus its value at PEEP 5 cmH2O, over the difference between non-inflated compartment total volume at PEEP 5 minus its value at PEEP 15 cmH2O [8].
Tidal hyperinflation was computed as the volume of the hyperinflated compartment at end-inspiration minus its volume at end-expiration [17], and standardized to predicted body weight.
The total superimposed pressure in the most dorsal parts of the lung was computed as previously described [18].
The lung inhomogeneity extent was measured by comparing the inflation of neighboring lung regions as previously described [19, 20] and was defined as the percentage of lung volume presenting an inflation ratio of neighboring regions greater than 1.61 (i.e., the 95th percentile of a control population) [19].
We finally developed a method to estimate elastic properties of the already aerated lung at PEEP 5 cmH2O with CT (CBABY LUNG, Additional file 1). Classical computation of compliance between PEEP 5 and 15 cmH2O (i.e., change in lung aerated volume divided by change in PEEP) overestimates CBABY LUNG as recruited alveoli account partly for the change in aeration. As recruitment assessed by CT is computed as the difference in non-aerated lung compartment weight between PEEP levels, a computation of recruited aerated volume (RecAer vol) from recruited lung weight was performed using the methodology proposed by Paula and coworkers [21], assuming that recruitable alveoli would remain aerated at PEEP 5 cmH2O and have equilibrated to a level of expansion equivalent to that of other already open alveoli at PEEP 5 [21]. CBABY LUNG between PEEP 5 and 15 cmH2O was finally computed as: (PEEPvolume—RecAer vol)/∆PEEP (i.e., 10 cmH2O).
A quality control was performed on both couples of CT images (CTExpi-Inspi, CTPEEP5-15). Images couples with segmented lung weight differing by more than 5% were excluded. CTExpi-Inspi, in which VTCT differed from VT set on the ventilator by more than 60 mL were also excluded (Additional file 2).
Statistical analysis
Statistical analysis was performed using R version 4.1.1 [22] with packages multcomp [23], lme4 [24], lmerTest [25], and interactions [26]. A p value ≤ 0.05 was chosen for statistical significance.
Data were expressed as count (percentage) or median [interquartile range] and compared between groups with the Fisher’s exact test for categorical variables and ANOVA for continuous variables. Multiple comparisons between groups were made using the Holm–Sidak procedure. Comparisons of variables involving the same individual were made with linear mixed models, using patient as a random effect.
Multivariate analyses were performed using linear models, by incorporating variables with p values < 0.2 in univariate analysis and stepwise backward selection.
Estimation of sample size was not computed as the study is exploratory, and data collection stopped with the control of COVID-19 fifth wave in our geographic area.
Missing data were not imputed owing to the low rate of missingness for variables included in the multivariate models (Additional file 3).