Out-of-hospital emergency medical care inFrance is a two-step process in which the patient receives basic life support from fire-fighter squadron paramedics, who typically arrive on scene within 5 to 10 minutes of the call (25-27); then advanced lifesupport from physicians in mobile medical teams from hospitals, who usually arrive on-scene within 15 minutes of the call, stabilize the patient, and take the patient directly to the ICU. In-hospital emergencies are managed by both emergency-room and ICU physicians who usually arrive on scene almost immediately and provide advanced lifesupport before ICU admission (28) .
Efforts to control the seizures and to identify their cause were started on-scene. Anticonvulsant drugs were given on-scene according to current recommendations (29-31). Intravenous benzodiazepines (clonazepam 1 mg or diazepam 10 mg) were given either before ICU admission at the discretion of the attending physician or in the ICU to patients with persistent seizure activity at ICU admission, that could be repeated up to three time, if necessary. Then, all patients with ongoing seizure activity received standard complementary intravenous anticonvulsant drugs (phenobarbital 10 to 15 mg/kg, phenytoin 18 mg/kg or equivalent dose of fosphenytoin). Patients with refractory status epilepticus received midazolam, propofol, or thiopental in titrated doses until remission of the clinical seizure activity. When the EEG revealed electrical status epilepticus, these anesthetic drugs were also given in titrated doses to induce EEG burst suppression then followed by continuous infusion during 12 hours at least.
Mechanical ventilation was used according to standardized criteria. Patients whose Glasgow Coma Scale (GCS) score remained less than 8 despite first-line anticonvulsant therapy received endotracheal mechanical ventilation. Some patients required on-scene intubation because of deep coma related to a post-ictal state or to electrographic status epilepticus, and others required mechanical ventilation because of progression to refractory status epilepticus requiring treatment with anesthetic agents. Finally, patients with aspiration pneumonia and respiratory failure or shock received intubation with rapid-sequence induction using various combinations of etomidate propofol or thiopental and succinylcholine. Because neuromuscular blockade can mask persistent seizure activity, patients were carefully examined within a few minutes after intubation.
Laboratory tests were obtained routinely in all patients. Plasma anticonvulsant drug assays and qualitative tests for toxic substances or medications associated with seizures were performed at the discretion of the attending physicians. Cerebral imaging and intermittent EEG monitoring were performed routinely. EEG was particularly performed in cases of patients with a depth coma, long lasting post-ictal state and/or evolution toward refractory status epilepticus. Lumbar puncture was performed when there was a fever or clinical suspicion of meningitis and when deemed appropriate by the attending physicians. Patients without meningeal involvement after extensive investigations but who presented with CSF abnormalities were defined as having postictal pleocytosis (32).
A standardized form was used to collect the variables listed in Tables 1 to 4. The clinical examination on scene and at ICU admission routinely included a thorough neurological evaluation with determination of the Glasgow Coma Scale (GCS) score (33) and evaluation for focal neurological signs. Focal neurological signs were defined as symptoms or signs consistent with damage to, or dysfunction of, a specific anatomic site in the central nervous system (34). Signs were unifocal or multifocal, and transient or persistent. Focal neurological signs were assessed as absent or present. Neurological failure was defined by any neurological disorder of central origin among impairment of consciousness, focal sign, encephalopathy, and meningeal symptoms, which required intensive care management for monitoring or life support management. Status epilepticus characteristics (Table 2) were collected on scene and at ICU admission. Begin of seizure activity was determined prospectively based on data in the pre-hospital notes, emergency room chart, ICU chart, and witness interview reports. All investigators were aware of the need to clock seizure duration.
The primary cause of status epilepticus was classified as cerebral insult when a structural cause was identified (e.g., stroke, central nervous system infection, cancer-related, traumatic brain injury) (14, 35). Severity and organ dysfunction at ICU admission were assessed using the Simplified Acute Physiology Score II (SAPS-II) (36) and the Logistic Organ Dysfunction (LOD) system score (37).
Quantitative parameters are reported as median (interquartile range [IQR]) and qualitative parameters as number [%]). Categorical variables were compared using Fisher’s exact tests and continuous variables using Wilcoxon rank-sum tests.
Patients whose 90-day GOS score was not determined were excluded from the analysis. To identify associations between patient characteristics and having a 90-day GOS score of 5, we used a logistic mixed-effect model to take into account a potential centre effect. Variables yielding p-values smaller than 0.20 in the univariate analysis and clinically relevant were considered for the multivariable model. Notably, only one severity scores (LOD) was kept in the initial multivariate model. Final multivariate was built using a stepwise backward selection procedure based on Akaike’s criterion. Variable selection was checked using bootstrap procedures. The bootstrap model selection involved the following steps: 1000 bootstrap samples were drawn with replacement, a backward stepwise variable selection algorithm was performed on each bootstrap sample (stopping rule based on Akaike’s criterion), variables selected in more than 50% of selections were kept for the final model. Log-linear effects of continuous covariates were tested, and the validity of the final logistic regression model was checked using the Le Cessie and van Houwelingen goodness-of-fit test. According to the fact that the log-linearity of the seizure duration was excluded (p=0.03), seizure duration was split into several categories to create an ordinal variable using a general additive model with natural splines (38, 39). The choice of categories was driven by the knots of the natural splines. Possible loss of information was evaluated using the likelihood ratio test. Odds ratios (ORs) and their 95% confidence intervals (95% CIs) were computed. Impact of missing 90 day GOS was evaluated by comparing the baseline variables of patients without 90 day GOS available to that of patients with 90 day GOS available. Moreover a sensitivity analysis was performed by taking into account missing data using a pooled analysis of five samples created by Multivariate Imputations by Chained Equations (MICE) to re-estimate ORs in the multivariate model (38, 40).
All tests were two-sided at the 0.05 significance level. Analyses were performed using the R statistical package (online at http://www.R-project.org).