The investigators conducted a multicentre, retrospective, observational study from April 2013 to February 2019 in the surgical ICUs of 5 level 1 Trauma centers in France. The investigators collected data from the "regional intensive care network of the great west of france" (AtlanRéa) database. The AtlanRéa database registers prospectively and consecutively numerous informations about brain injury and trauma patients hospitalized in ICUs, in order to provide epidemiologic informations for this population. Data were collected by clinical research assistants in each participating ICU, using an electronic case report form. The investigators assessed data regarding outcomes after discharge from ICUs through phone interview led by dedicated clinical research assistants.
The following data were entered in the database: Age, gender and body mass index, previous medical history, and more specifically the presence of cardiac insufficiency, chronic renal failure defined as an estimated glomerular filtration rate less than 60 ml/min.1.73 m-1, chronic respiratory illness, neurologic background, diabetes, neoplasia history, previous TBI, active smoking and chronic alcoholism. Mechanism of injury (domestic accident, road traffic crash, fall from heights or others). The Glasgow Coma Score (GCS) score determinates in the prehospital setting or at admission at hospital before intubation and/or sedation, and the presence of at least one nonreactive and dilated pupil at the initial management. Severity of illness according to the Simplified Acute Physiology Score II (SAPS II) the Sequential Organ Failure Assessment (SOFA) score, and the Injury Severity Score (ISS). The investigators also specified the severity of TBI by the abbreviated Injury Score (AIS) and reported the associated injuries from 5 territories (face, chest, abdomen, extremity (including pelvis) and external). Initial CT-scan was classified according to the Marshall classification in six categories (Diffuse injury I, II, III, IV, evacuated mass (V) and non-evacuated mass lesion (VI)).
The following data during the patients' hospitalization are also entered in the database: use of an intracranial pressure (ICP) catheter, occurrence of intracranial hypertension (defined as an ICP above 20 mmHg in absence of confounding factors), and use of barbiturates and/or osmotherapy, decompressive craniectomy, and other neurosurgical procedure. The occurrence of intercurrent events occurring during ICU stay, including infections, Acute Respiratory Distress Syndrome (ARDS), need of vasopressor drugs, acute renal failure, thrombophlebitis, pulmonary embolism, hemorrhage, infection and tracheostomy. The investigators specified the need for intubation and durations of mechanical ventilation, central venous catheterism and vasopressor. The investigators also recorded the decision in ICU to withdraw or withhold life support.
The length of stay in ICU, ICU and 3-month mortalities and patient outcome were assessed using the Glasgow Outcome Scale Extended (GOSE) score at 3 months. The investigators dichotomized the GOSE score between the four lower values (corresponding to unfavorable outcome- GOSE 1 to 4) and the four upper values (corresponding to favorable outcome- GOSE 5 to 8).
Statistical Analysis All statistical analysis will be performed using R software 3.3.1 (package pROC) and SAS 9.1 Statistical Software (SAS Institute, Cary, NC, USA). For quantitative continuous variables, position and dispersion parameters (mean, standard deviation, median, interquartile range) will be calculated. For qualitative variables, proportion rates will be calculated. For groups' comparison, we used independent-samples t-tests for normally distributed continuous variables, and Mann-Whitney U-tests for non-normally distributed continuous variables. Χ² tests will be performed for categorical variables. To build the model for multivariate analysis, the investigators selected among the variables with a p ≤ 0.20 according to the univariate analysis. The investigators performed logistic regression model to identify factors associated with in-ICU mortality and dichotomized GOSE score at 3 months. The Odds Ratios (ORs) and 95% confidence intervals (CI) will be calculated. For continuous variables, diagnosis of log linear behavior was established, and if not confirmed, the variable will be divided into categories and treated as categorical variables. Model's fit will be assessed by checking residual plots and Cook's distance. The calibration of the model will be tested by Goodness of Fit Hosmer-Lemeshow test. A p ≤0.05 will be considered statistically significant for all the comparisons.
