Design and ethics
This was a retrospective study performed in three ICUs of Assistance Publique—Hôpitaux de Paris, University Hospitals (Louis Mourier, Lariboisière and Kremlin-Bicêtre) between March 2009 and March 2014 designed to compare the risk of surgical complications that could be a priori induced or worsened by prone position between patients who had at least one prone position session (prone group) and those who remained supine (supine group) after abdominal surgery. Admission of abdominal emergencies in these three ICUs is part of their routine activity. In case of ARDS, decision to prone patients was taken by the ICU physicians and the context of abdominal surgery was not considered as a contraindication.
The study was approved by the Ethics Committee of the French Intensive Care Society (project no. 14-31). We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement guidelines for observational cohort studies [15].
Study population
Two independent searches on the ICU’s electronic database were performed over the study period, one with the search label “ARDS” (ICD label J80) and the other with “acute respiratory failure” (ICD label J960). The two lists of patients were cross-checked to ensure exhaustibility and verify the final diagnosis of ARDS. Once extracted, medical records were reviewed. Patients were retained in the final analysis if they had an ARDS consistent with Berlin definition [16] (oxygenation criteria: PaO2/FiO2 <300 mmHg with PEEP or CPAP ≥5 cmH2O) in a context of recent (less than 7 days) abdominal surgery. We did not include in the analysis patients who had just had a laparoscopy or who died in the next 48 h following surgery.
The day of inclusion (D0) in the analysis was defined as the day when ARDS occurred.
Main characteristics of protocol use for the prone positioning placement
During the study period, medical and paramedical teams followed protocol for prone positioning placement. A minimum of four persons were required for the procedure; one of them was placed at the patient’s head to secure the endotracheal tube. Rotation to the left or to the right depended on the location of invasive arterial pressure and central venous lines. The upper limbs were placed alongside the body. Potential pressure points were protected using adhesive pads.
A circular pillow was used to ensure appropriate position of the head and the endotracheal tube. Pillows were placed under the thorax and pelvis in order to limit abdominal pressure.
Data collection
The data recorded from the files were the following:
Epidemiological data: age, sex, weight, body mass index (BMI), chronic obstructive pulmonary disease, ischemic heart disease, systemic hypertension and diabetes.
Characteristics of ICU severity: SAPS II [17], septic shock (at D0) defined by Bone’s criteria [18] and catecholamine infusion (at D0).
Characteristics of ARDS and mechanical ventilation: lowest PaO2/FiO2 ratio at D0, highest plateau pressure (Pplat) at D0, lowest tidal volume at D0, highest PEEP at D0, use of adjunctive therapies (including neuromuscular blocking agents, inhaled nitric oxide, prone positioning) and duration of mechanical ventilation. For patients who had at least one prone position session (prone group), data collection included: time between surgery and first prone position session, number and duration of prone position session, PaO2/FiO2 before (measure on the last arterial blood gas before first prone session) and after (measure on the first arterial blood during the first prone position session) and hemodynamic changes after first prone session. To address this hemodynamic issue, we defined three categories depending on the changes in catecholamine dosage during the first 2 h of the first prone session: i/hemodynamic worsening (defined as increase in catecholamines), ii/hemodynamic improvement (decrease) and iii/hemodynamic stability (no change).
Characteristics of abdominal surgery: planned or emergent surgery, delay between surgery and ICU hospitalization, presence of peritonitis (defined according to the International Sepsis Forum Consensus Conference on Definitions of Infection in the Intensive Care Unit [19], type of surgical procedure, number and type of stoma. Not being a routine procedure, intra-abdominal pressure was not systematically measured.
Postoperative surgical complications: We defined a priori these complications: scar dehiscence, abdominal compartment syndrome (define as intra-abdominal hypertension >20 mmHg with new organ dysfunction or failure) [20], stoma leakage, stoma necrosis, scar necrosis, wound infection, displacement of a drainage system, removal of a gastro- or jejunostomy feeding tube and digestive fistula. The Clavien–Dindo classification for surgical complications was assessed. However, it was not discriminant since all the patients were de facto in the ICU and under invasive MV (≥IVa) [21].
Primary endpoint
The primary endpoint was the number of patients who had at least one surgical complication defined a priori (see above) that could be induced or worsened by prone position.
Secondary endpoints
Secondary endpoints were the number of revision operations due to complication induced or worsened by prone position. Other secondary endpoints were effect of prone position on oxygenation duration of mechanical ventilation, ICU mortality and ICU length of stay.
Statistical analysis
Statistical analysis was performed with GraphPad Prism 5 (GraphPad Software, San Diego, USA) and R version 3.1.2 (R Foundation for Statistical Computing, Vienna, Austria.). Categorical variables are described by their numbers and proportions and compared by the Fisher’s exact test. The normality of continuous variables was tested by the Kolmogorov–Smirnov test. Continuous variables of normal distribution are described by mean and standard deviation and compared by Student’s t test. Continuous variables of non-normal distribution are described by median and interquartile [25, 75%] range and compared by the Mann–Whitney test.
Primary endpoint was compared between the prone and supine groups using propensity score weighting to balance patient characteristics between the two groups. It was conducted in two stages. In the first stage, we performed a multivariate logistic regression to predict the probability of being in the prone group (i.e., the estimated propensity score (PS)), controlling for all the pre specified covariates (see above). In the second stage, we constructed logistic regression model to compare the risk of complication between prone and supine groups, using the inverse of the propensity score as a weight, targeting the average treatment effect in the whole population [22]. More precisely, a logistic model regressing the outcome with exposure (i.e., prone or supine group) as the only covariate was fitted, each subject being weighted according to its PS value, with a stabilized weight W equal to: W = pT/PS if subject is in the prone group, and (1 − pT)/(1 − PS) if subject is in the supine group, where pT is the is the overall probability of being in the prone group in the sample. Robust standard errors were used.
Variables considered for propensity score estimation were chosen based on empirical knowledge and included: age, weight, SAPS II, diabetes status, presence of a colonic stoma, of a small bowel stoma and of jejunostomy, use of catecholamines and delay from surgery to ICU hospitalization. No variable selection was performed. Balance on covariates between prone and supine groups was assessed and reported using absolute standardized differences (ASD) [23], and a sensitivity analysis with additional adjustment for covariates with ASD >10% after weighting was performed.