Materials and Methods
Cohort Definition
We included consecutive adult patients (more than 16 yr old) with SE, admitted to our center between April 1, 2006, and July 30, 2013 (88 mo), who were prospectively enrolled in our registry that received full approval from our Ethic commission. Cases with postanoxic etiologies were not enrolled in our registry. SE was defined as the occurrence of continuous seizures or repetitive seizures between which there is incomplete recovery of baseline clinical conditions for greater than or equal to 30 minutes (until 2008) and for greater than or equal to 5 minutes (since 2008). SE episodes were clinically diagnosed by neurology consultants and confirmed with electroencephalogram (EEG) studies, which was mandatory for nonconvulsive forms. Further details on this registry have been published previously. This study considered only incident cases of SE which lasted greater than or equal to 30 minutes, in order to avoid overrepresentation of certain individuals and include only episodes with consistent prognostic implications.
Variables
Age, gender, and history of previous seizures were prospectively recorded on admission. Etiology leading to death if not specifically treated was categorized as "potentially fatal" as previously detailed. Level of consciousness before treatment was categorized as alert, confused or somnolent (arousable toward a clear clinical contact), stuporous (arousable, but without contact), and comatose; the latter two were classified as "severe impairment of consciousness." Type of SE was defined by the worst clinical seizure in the given episode and classified, in increasing order of severity, as simple partial (focal without consciousness impairment), absence, myoclonic (related to genetic generalized epilepsy), complex partial (focal with consciousness impairment), generalized convulsive SE (GCSE), or nonconvulsive SE in coma (NCSEC). GCSE episodes were further dichotomized into "proper" GCSE (those which presented prolonged generalized convulsive seizures up to the SE diagnosis and treatment) and "GCSE then focal" (those with focal seizures without coma on diagnosis, but having presented an earlier generalized convulsion during the same episode), as we hypothesized that these two forms might have different prognostic implications. The SE severity score (STESS), a validated clinical scoring system considering age, worst seizures type, level of consciousness impairment, and history of previous seizures, was calculated for each patient at admission.
We prospectively recorded use of TC for SE treatment, specific anesthetic agents administered in each episode, and treatment latency (representing the best estimated time from SE onset to administration of the first medication, and dichotomized at 1 hr following the beginning of the SE episode). TC was a clinically driven endpoint, monitored by EEG (with seizure suppression, or burst suppression, as target), and it was accomplished using anesthetic drugs as continuous IV drips. Our hospital is a third-level center including a multidisciplinary ICU; patients with SE are mostly treated in the neurology ward (intermediate care unit) and admitted to the ICU if they need mechanical ventilation. The main clinical outcome, prospectively assessed at hospital discharge, was categorized into three groups: return to baseline, new disability (defined as new neurological impairment, as compared to the situation before the incident SE episode), or death.
Statistical Analysis
The association between potential predictors and clinical outcome was analyzed using univariable multinomial logistic regressions. Multinomial (polytomous) logistic regression fits maximum likelihood models with discrete dependent variables, when the dependent variable takes on more than two outcomes and the outcomes have no natural ordering, as in the present study (supplemental data, Supplemental Digital Content 1, http://links.lww.com/CCM/B200). Outcome prognosticators with a p value less than 0.05 were used in a backward procedure to fit a multivariable multinomial model. Results were described with relative risk ratios and 95% CIs. We conducted analyses in the complete cohort as well as in patients with GCSE "proper" or NCSEC versus other SE forms.
Prevalence of infectious complications was analyzed in a nested case-control assessment comparing all patients managed with TC and a control group of the same number of subjects treated without TC, matched for outcome, potentially fatal etiology, STESS, and CCI; results are given in odds ratio (OR) and 95% CI. The same approach was used to analyze the length of hospital stay using a Wilcoxon test; patients who died were excluded from this analysis in order to avoid bias of shorter hospitalization. We chose to adjust for the aforementioned variables in order to specifically address the role of TC (exposure) on infections, respectively duration of in-hospital stay (outcomes), as clinical outcome may confound these relationships. The use of specific anesthetic drugs for coma induction was investigated with a Fisher exact test comparing the three outcome groups. Analysis was performed using the Stata software version 12 (StataCorp, College Station, TX); significance was considered at p value less than 0.05.