The INTUBATIC study was a prospective, multicenter observational study conducted in 30 ICUs (29 in France and one in Spain) in both academic and non-academic hospitals. The primary endpoint was to assess rate of intubation, early and late, and factors associated with intubation practice in septic shock patients. The secondary endpoint was to assess mortality according to intubation.
We considered a time window of 8 h following vasopressor onset to define early intubation, from the results of our previous study . Two approaches were used to assess the parameters associated with early intubation. First, criteria for standard indications for early tracheal intubation were defined a priori, based on accepted recommendations for invasive ventilation initiation in neurologic or respiratory failure. Frequency of early intubation and patient characteristics were described among those with such criteria [12,13,14,15,16]. Second, a multivariate analysis entering a wide array of parameters was performed to assess the link between these parameters and early intubation among all patients. An evaluation of the goodness-of-fit of the model was performed to determine the amount of early intubation variability explained by the different covariates.
The inclusion period ran from May 2016 to October 2017. ICUs were involved gradually and had 12-months to include patients. The protocol allowed for a maximum of 60 patients enrolled per center. Physicians participating in the study were aware of its main objective, but were instructed to perform patient’s care as they usually did, and no specific hypothesis was put forward regarding optimal management of these patients. Patients, or a proxy if the patient was deemed unable to consent, received oral and written information about the study, and oral consent was obtained before inclusion. When a proxy gave the initial consent, the patient’s consent was further obtained whenever possible. This study was approved by the Angers University Hospital Ethics committee (n° 2015/96).
Patients aged 18 years and above were eligible if admitted in a participating ICU for septic shock, defined by a documented or clinical suspicion of infection and hypotension requiring vasopressor infusion despite adequate fluid loading. Patients could be included if a vasopressor was introduced in the 24 h preceding ICU admission, i.e., in another hospital or the emergency room. Patients were not eligible if vasopressor infusion was started after tracheal intubation and mechanical ventilation (non-invasive ventilation through facial mask did not prevent inclusion). Incapacitated adults, pregnant women, patients with a decision of withdrawal or withholding of care before ICU admission, patients without social health insurance and patients who refused to participate in the study were not included.
Age, sex, and main comorbidities were noted. SAPS II and SOFA score were calculated 24 h after ICU admission [18, 19]. Infection site, causal pathogen(s), and nosocomial or community-acquired subset of infection were registered. Hemodynamic, respiratory, and neurological parameters were recorded between the time of vasopressor onset (H0) and H8: oxygen administration device and PaO2/FiO2 ratio (see Additional file 1: Table S1 for FiO2 determination), respiratory rate, accessory inspiratory muscle use, paradoxical abdominal breathing, inability to cough or to clear tracheal secretions, vasopressor infusion rate (norepinephrine or epinephrine in all centers), arterial lactate concentration, serum creatinine level, urine output, and Glasgow coma score. In patients not intubated by H8, the worst values of acute neurological and respiratory severity parameters over this period were registered. In patients intubated before H8, the worst parameters between H0 and intubation were registered. Fluid loading from the first hypotension to H0 was noted. Durations of ICU stay, hospital stay, vasopressor infusion, mechanical ventilation, and renal replacement therapy were registered. ICU, hospital and 28-day mortality were recorded.
Standard criteria for theoretical immediate intubation
Criteria qualifying for theoretical immediate tracheal intubation during the H0–H8 time window were either (independently of intubation being performed or not):
Neurological failure: Glasgow coma scale < 10.
Respiratory failure, two criteria among these had to be present: oxygen saturation less than 90% during more than 5 min despite optimized oxygen administration, respiratory rate more than 35 per minute, significant accessory respiratory muscle use, respiratory acidosis defined by pH < 7.35 and pCO2 > 45 mmHg, hypoxemia with PaO2/FiO2 ratio inferior to 150, and inability to cough or clear tracheal secretions. These criteria are generally associated with strong recommendations for immediate tracheal intubation in case of neurological or respiratory failure [12,13,14,15,16].
We sought to include more than 800 patients, estimating an early intubation rate ranging from 20 to 40%. This allowed for the selection of 20 to 25 covariates to be included in a multivariable model for explaining early intubation . Continuous data were summarized as the mean and standard deviation or median with inter-quartile range as required and compared using the Kruskal–Wallis test. Categorical data were expressed as number and percentage and compared using the Fisher exact test. The cumulative hazard function of event occurrence was estimated using the Nelson–Aalen procedure.
Early intubation was studied through a mixed-effects logistic regression model, considering a list of covariates (considered as fixed effect covariates). Covariates entered in the multivariate analysis were chosen a priori based on their relevance after reaching a consensus between NL, JFH, AD, MSM, FM, SJ and CD based on published literature review and advice from experts of the topics (Prof. Pierre Asfar and Prof. Laurent Brochard). The practice variation between participating centers was also considered in this model by including “center” as a random effect covariate. Patients intubated for emergent surgery procedure were excluded for this analysis. No imputation was performed for missing data. The goodness-of-fit was assessed using a McKelvey pseudo R-squared measure, evaluating the outcome variability based on the constructed model . The weight of each covariate included in the outcome variability explanation was assessed through the percentage of McKelvey pseudo R-squared associated with this covariate .
Graphical representation of patients survival was performed with Kaplan–Meier method, and survival rates were compared with log-rank test. All the tests were two-sided considering a type I error set at 0.05. The statistical analyses were performed using Stata® 13.1.