FormalPara Key Points

Recent studies revealed an association between continuously administered anesthetics for treatment-refractory status epilepticus and poor outcome, an association that may be confounded, as anesthetics are mostly administered in patients with more severe status epilepticus and critical illnesses.

In this two-center cohort study using three different statistical approaches (propensity score matching, coarsened exact matching, and multivariable logistic regression model) to account for possible confounding, the use of continuously administered intravenous anesthetic drugs for the treatment of refractory status epilepticus was independently associated with death and adverse outcomes in survivors.

Randomized trials are needed to determine if these associations are biased by outcome predictors not yet identified and therefore not accounted for in this study.

1 Introduction

Status epilepticus (SE) is a life-threatening neurological emergency. Benzodiazepines are recommended as first-line medications to terminate seizures, as high-level evidence supports their superiority among antiepileptic drugs (AEDs) [1]. If SE persists, second-line treatment with phenytoin, valproate, or levetiracetam is recommended [2, 3]. Patients refractory to first- and second-line treatment are challenging, as mortality is high (40%) and further treatments are based on recommendations rather than evidence [4]. Guidelines recommend continuously administered intravenous anesthetic drugs (IVADs), such as midazolam, propofol, or pentobarbital [2] with the aim of attaining seizure suppression, a burst suppression [5], or an isoelectric pattern on electroencephalography [6]. The rationale for the use of IVADs in patients with SE refractory to first- and second-line AEDs seems clear; there are, however, concerns regarding side effects, including an increased need for mechanical ventilation with delayed weaning, and an increased risk for pneumonia and arterial hypotension [7, 8]. Recent studies of SE patients demonstrate that the administration of IVADs was associated with poor outcome [9–11]. These studies, however, could not sufficiently exclude the possibility that the IVAD recipients had clinical characteristics that led the clinician to believe that this group of patients was “more severely ill”, hence triggering the use of IVADs (i.e., confounding by indication). Randomized controlled trials regarding the risks and benefits of IVADs in SE are not registered according to the US National Institute of Health, reflecting the ethical constraints of assigning or excluding patients with treatment-refractory SE from IVADs, as they bear the inherent risk for sustained SE.

The aim of this two-center study was to reassess the associations between IVADs and outcomes in an unmatched, a propensity score-matched, and a coarsened exact-matched population of adult SE patients. Patients treated with and without IVADs were matched to compare IVAD recipients and non-recipients with similar clinical characteristics that could have triggered the use of IVADs.

2 Materials and Methods

This retrospective observational study was performed in the intensive care units of the University Hospital Basel, Switzerland and the Johns Hopkins Bayview Medical Center, USA, two tertiary academic medical care centers, with 813 and 428 beds, respectively. Patients from the Swiss cohort were partially (2005–2010) derived from an ongoing registry of patients with SE in which associations between the use of IVADs and adverse outcomes have previously been published [10]. In each hospital, the intensive care units requested consults from the same team of neurologists for the diagnosis and treatment of SE patients throughout the study period. Our database was managed and this study was implemented according to the STrengthening the Reporting of OBservational studies in Epidemiology statement [12].

The study was approved by the local ethics committees and institutional review boards of both hospitals and patients’ consent was waived.

2.1 Data Collection

From 1 January, 2005 to 1 January, 2013, data of all consecutive adult patients with SE obtained from the digital SE databases and medical charts were collected by the same investigators in both hospitals including demographics, level of consciousness at SE onset, SE duration, and acute/fatal etiology (traumatic brain injuries, cerebrovascular accidents, [meningo-]encephalitis, brain tumors, surgical brain lesions, and acute intoxication). Patients with hypoxic-ischemic encephalopathy were excluded [13]. Worst seizure types at presentation were categorized as (1) simple partial, complex partial, and absence seizures (2), generalized convulsive seizures, and (3) non-convulsive SE in coma [10, 14–16]. SE control, co-morbidities, duration of mechanical ventilation, vasopressors, infections, and intensive care unit and hospital stays were assessed. The diagnostic criteria for infections during SE are outlined elsewhere [17].

2.2 Treatment

Antiepileptic drug therapy was standardized according to the international recommendations at both sites. First-line AEDs (intravenous bolus of benzodiazepines) were followed by second-line AEDs (phenytoin, levetiracetam, or valproic acid) if SE persisted. In SE refractory to first- and second-line AEDs, non-anesthetic third-line AEDs (including lacosamide, topiramate, carbamazepine, and oxcarbazepine) with or without IVADs (including continuous infusions of midazolam, propofol, and/or barbiturates) were administered. All patients were monitored with repetitive spot electroencephalograms (EEGs) at least every 12 h or with continuous EEG monitoring. Midazolam and/or propofol were administered for seizure suppression before barbiturates were considered. Barbiturates were used to induce an isoelectric EEG [18].

2.3 Duration and Severity of Status Epilepticus

Status epilepticus was defined as clinical and/or EEG evidence of seizures lasting >5 min or as series of seizures without a complete intervening clinical recovery [2, 19]. SE was defined as controlled in patients without IVADs if no further clinical and/or EEG evidence for recurring seizures was observed for at least 12 h after the last escalation of AED treatment. In patients treated with IVADs, SE was defined as controlled if no further clinical and/or EEG evidence for recurring seizures was observed throughout the weaning phase of IVADs and 12 h after cessation of IVADs. Severity of SE was assessed by the independently validated Status Epilepticus Severity Score (STESS) [14–16].

2.4 Outcomes

Primary outcomes included the Glasgow Outcome Scale (GOS) at hospital discharge in survivors, and death during hospital stay. Duration and control of SE were considered secondary outcomes. Unfavorable functional outcome in survivors was defined as a GOS of 1–3.

2.5 Statistics

Patients were categorized into two groups: with and without treatment with IVADs. Chi-square and Fisher’s exact test were used for comparisons of proportions. For continuous variables, the Shapiro–Wilk test was used to distinguish between normal and abnormal distributions. Normally distributed variables were analyzed with the Student’s t test and non-normally distributed variables with the Mann–Whitney U test.

We attempted to overcome confounding related to disease severity by applying three different statistical approaches, the rationale being that if all three yield similar results this strongly supports the robustness of our findings.

To account for imbalances in characteristics between IVAD recipients and non-recipients, matched sub-analyses were performed using two different matching methods: propensity score and coarsened exact. Propensity score methods were used to control for pretreatment outcome determinants that may differ between IVAD recipients and non-recipients (i.e., possible confounders). The probability of a patient being treated with IVADs (i.e., propensity score) was calculated with a logistic regression model based on established confounders including age, level of consciousness, and worst seizure type (integral components of the STESS that were significant in the univariable analysis), acute/fatal etiology, and need for mechanical ventilation—variables identified as independent outcome predictors in prior studies on SE [20] and associated with the use of IVADs in the present and prior cohort studies on IVADs in SE [10, 21]. The final model was checked with the Hosmer–Lemeshow goodness-of-fit test. IVAD recipients and non-recipients were matched on propensity score with use of nearest-neighbor matching with a caliper set at one quarter of the standard deviation of the logit of the propensity scores [22]. In addition, t tests and Chi-square tests were used to check for balances between the matched IVAD recipients and non-recipients with respect to each patient characteristic.

Because with a significant reduction of matched pairs after propensity score matching the results can be biased, a second matching technique was applied using the same patient characteristics representing possible confounders for coarsened exact matching. This is a matching method of the class monotonic imbalance bounding [23]. This means that reducing imbalance in the empirical distribution in one covariable has no effect on any other covariables chosen for balancing, which represents a clear advantage of this method over other matching methods [24]. Logistic regression incorporating matched weights was used to check for balances between the matched IVAD recipients and non-recipients with respect to each patient characteristic.

Propensity score matching and coarsened exact matching were performed within each hospital. Patients without an eligible match were excluded from additional analyses to reduce the risk of bias from non-exchangeable subjects. A logistic regression model was performed in the unmatched and matched cohorts to determine odds ratios (ORs) for death and unfavorable functional outcome in survivors. To account for confounding by important pretreatment outcome determinants, additional adjustments for status duration (the most important outcome predictor not known at SE onset) were performed in the unmatched and both matched cohorts. Two-sided p values ≤0.05 were considered significant. Statistical analysis was performed with STATA® version 12.0 (Stata Corp., College Station, TX, USA).

3 Results

3.1 Demographic, Clinical, and Treatment Characteristics

Overall, 406 consecutive adult patients were identified with SE (275 [68%] patients from the Swiss hospital and 131 [32%] from the US hospital). Table 1 presents univariable comparisons of demographic, clinical, and treatment characteristics in IVAD recipients and non-recipients. Both the propensity score- and the coarsened exact-matched cohorts were balanced for important pre-defined clinical confounders (Table 2; Fig. 1). The box plots of the median propensity scores for patients treated with and without IVADs are presented in Fig. 2. While in the original unmatched cohort, patients treated with and without IVADs differed regarding potential confounders, such as age, acute/fatal etiology, seizure type, history of seizures, level of consciousness, SE severity, and the need for mechanical ventilation, both the propensity score- and coarsened exact-matched cohorts were balanced regarding these characteristics (Table 2).

Table 1 Demographics and clinical characteristics of patients with and without IVADs
Full size table
Table 2 Propensity score matching and coarsened exact matching for potential confounders
Full size table
Fig. 1
figure 1

Flow chart. IVADs intravenous anesthetic drugs

Full size image
Fig. 2
figure 2

Propensity score analysis. Box plots of median propensity scores for patients with and without IVADs. IVADs intravenous anesthetic drugs

Full size image

3.2 Course and Outcome

Univariable comparisons in the matched cohorts revealed that IVAD recipients had severe hypotension more  frequently requiring hemodynamic support by continuous administration of at least one vasopressor. While infections emerged more frequently during SE in IVAD recipients of the unmatched cohort (most infections being respiratory tract infections followed by urinary tract infections and bacteremia) (Table 1), infections no longer differed between IVAD recipients and non-recipients in the matched cohorts (Table 3).

Table 3 Univariable comparisons of course and outcome in patients with and without IVADs in the matched cohort
Full size table

Similarly, SE duration no longer differed between patients with and without IVADs in the matched cohorts. Unfavorable outcome in survivors and mortality were increased in patients treated with IVADs in both the unmatched and the matched cohort (Table 3). In the unmatched cohort, patients treated with IVADs died more often from uncontrolled SE or from care withdrawal, a difference that lost significance in the matched cohort.

Logistic regression in the unmatched cohort adjusted for potential confounders such as age, level of consciousness, worst seizure type, SE acute/fatal etiology, and the need for mechanical ventilation revealed increased ORs for unfavorable outcome in survivors and death (Table 4). Logistic regression performed using the propensity score-matched cohort, and the coarsened exact-matched cohort revealed similar ORs for the association between IVADs and unfavorable outcome in survivors and death (Table 4).

Table 4 Logistic regression analyses in the unmatched, propensity score-matched, and coarsened exact-matched cohorts for the effect of IVADs on primary outcome
Full size table

Effect estimates and significance levels did not change substantially after additional adjustment for seizure duration in both the unmatched (OR 3.09, 95% CI 1.43–6.66, p = 0.004 for death and OR 2.48, 95% CI 1.35–4.54, p = 0.003 for GOS 1–3 in survivors), the propensity score-matched (OR 3.95, 95% CI 1.51–10.33, p = 0.005 for death and OR 2.27, 95% CI 1.02–5.06, p = 0.045 for GOS 1–3 in survivors), and the coarsened exact-matched (OR 2.33, 95% CI 1.31–4.16, p = 0.005 for death and OR 3.86, 95% CI 2.07–7.20, p < 0.001 for GOS 1–3 in survivors) cohorts.

4 Discussion

We reveal independent associations between the use of IVADs and poor outcome in SE by using different statistical approaches, thus underscoring the robustness of these findings and adding further credence to the limited body of evidence that the use of IVADs in SE is strongly associated with unfavorable outcome.

Concerns regarding the use of IVADs in SE patients are increased by our results, as the association between IVADs and unfavorable outcome in survivors persists after the exclusion of significant confounding effects of variables identified as independent outcome predictors and associated with the use of IVADs [10, 21], such as age, level of consciousness, etiology, worst seizure type, need for mechanical ventilation. The association of IVADs and poor outcome in SE patients is believed to be mediated by different systemic adverse effects of IVADs, such as severe arterial hypotension and an increase of infections possibly caused by prolonged mechanical ventilation [9–11, 17]. Episodes of severe arterial hypotension requiring treatment with at least one continuously administered vasopressor were more frequent with IVADs. This is in line with prior studies reporting that 16–57% of SE patients receiving IVADs require vasopressors [9, 10]. Severe hypotension from thiopental has been described in patients during anesthesia for surgery [25] and a systematic review regarding the effectiveness of IVADs to terminate treatment-refractory SE revealed high mortality (48%) and a larger proportion of severe arterial hypotension with pentobarbital [26]. The use of vasopressors has been linked to an increase of pulmonary hypertension, potentially worsening respiratory compromise caused by IVADs [27–29], and increased cerebral blood flow, possibly leading to elevated intracranial pressure and worsened edema [30].

In our patients with similar clinical characteristics likely to trigger the use of IVADs such as age, level of consciousness, worst seizure type, acute/fatal etiology, and the need for mechanical ventilation (variables used to calculate the propensity score and to balance with the coarsened exact matching), infections were not seen more frequently in IVAD recipients compared with non-recipients, indicating that infections are possibly not a major mediator of poor outcome in IVAD recipients. We could not identify additional mediators for unfavorable outcome in IVAD recipients, possibly explained by an accumulation of minor effects of a number of different mediators (e.g., infections, vasopressors, and side effects from AEDs and IVADs) on outcome.

In our unmatched cohort, patients treated with IVADs died more often from uncontrolled SE or after care withdrawal, a finding similar to the results from a prior study of new-onset refractory SE [31]. However, in our matched cohorts, death from uncontrolled SE or after care withdrawal did not differ significantly between IVAD recipients and non-recipients; hence, the link between the use of IVADs and uncontrolled SE or care withdrawal does not explain the association between treatment with IVADs and unfavorable outcome. In our cohort, 26 patients treated with IVADs were not mechanically ventilated, leading to the assumption that these patients received only low doses of IVADs (i.e., inappropriate treatment) and were therefore not in need of mechanical ventilation. In fact, in only two of these 26 patients, SE could not be controlled because of care withdrawal and the short-term low-dose treatment with IVADs in the remaining 24 patients without the need for mechanical ventilation is explained by a rapid response of SE to IVADs after starting IVADs and before the dosage was increased.

Midazolam was used more frequently than propofol, most likely because benzodiazepines have less suppressive effects on hemodynamics and the evidence for the efficacy of benzodiazepines regarding SE termination is stronger compared to propofol [1]. SE control usually needs high doses of propofol with the risk of arterial hypotension and the maintenance of continuous-burst suppression is often difficult [32]. In addition, barbiturates were the only IVADs used to induce an isoelectric EEG after midazolam and/or propofol failed to terminate SE. Hence, it remains questionable if barbiturates are similarly associated with death or unfavorable outcome in survivors when used for seizure control without aiming for an isoelectric EEG. However, barbiturates were also linked to cardiotoxic effects, severe arterial hypotension [26], delayed recovery from coma, prolonged mechanical ventilation, and intensive care stay [8, 26, 33]. Of note, in two prior studies there was no significant difference concerning the association between specific anesthetics and death [10, 11].

The limitations of this study include the observational design. Therefore, analysis can only provide associations and inference regarding causality cannot be drawn. We attempted to overcome confounding related to SE severity by controlling for (among other known confounders) SE severity using the integral components of the STESS instead of treatment refractory SE (RSE) for the following reasons. First, the STESS is an independently validated scoring system for the prediction of mortality in SE, which is the primary outcome in our study. Second, there is an ongoing debate on how to define RSE; many authors require failure of two AEDs before deeming SE refractory, producing unacceptable delays in using definitive therapies, and trials revealed that there is an unacceptably small likelihood that a second conventional agent would succeed [1], leading to the conclusion that failure of any additional drug should constitute RSE [34]. As by definition, any SE episode treated with IVADs is categorized as refractory, hence leading to a strong collinearity impeding further correction for treatment refractoriness in the present cohorts, only a randomized clinical trial comparing third-line AEDs with IVADs can finally address the impact of IVADs on outcome in patients with RSE.

Despite our attempt to overcome confounding by using three different statistical approaches, unmeasured residual confounding may have occurred. The balance of important variables determining SE severity and patient’s characteristics in the matched cohorts is, however, reassuring. As not all patients had continuous EEG monitoring and some had spot EEGs at least every 12 h, SE duration represents an estimation. As SE duration was not an outcome measure, it is unlikely that this approximation had a significant impact our findings. Even in prospective studies with continuous EEG monitoring, SE duration may be underestimated, especially if onset was unwitnessed as often is the case for non-convulsive SE. The fact that SE was defined as controlled, if no further evidence for seizures was observed throughout and 12 h following the weaning phase of IVADs may have hampered the judgment on the effectiveness of IVADs in patients receiving barbiturates, owing to the long biological half-life. However, only 16 patients received barbiturates in our study.

The correlation between dosage and duration of IVAD administration and outcome could not be analyzed, as they were influenced by the individual EEG responses and not by the actual serum levels. Individual IVAD doses were adapted according to interacting co-medication and individual factors such as induced liver enzymes. Hence, serum levels did not reflect the administered doses in many patients.

The two-center design, analyses restricted to patients with eligible matches reducing the risk of bias from non-exchangeable subjects, and the fact that several baseline characteristics in our cohort are similar to those of prior studies of SE represent strengths of our study. In particular, median age [35, 36], the proportions of acute/fatal etiologies, non-convulsive SE [35, 36], infections during SE [17, 37], and mortality [38] are close to prior studies of SE.

As our study included patients from 2005 to 2013, SE was defined as clinical and/or EEG evidence of seizures lasting >5 min or as a series of seizures without a complete intervening clinical recovery. However, the revised definition of SE by the International League Against Epilepsy suggests a seizure duration of >10 min for complex partial SE [39], a discrepancy that cannot be overcome with the current study design.

Although recent studies have identified that prolonged RSE is associated with sequelae reflected by increased brain atrophy [40] and mortality [4] calling for rapid treatment and termination of SE, there is growing uncertainty regarding the benefits of IVADs in SE vs. their potential harm [41]. Should rapid intubation and high-dose IVADs be favored against less aggressive management with the risk of delayed seizure control and neuronal damage [42]? In specific subgroups, SE-related impairment of consciousness with loss of airway protection or durable generalized convulsive seizure leading to neuronal injury may direct the risk-benefit scale in favor of IVADs. Closer and intensified EEG monitoring of SE patients treated with IVADs may help to reduce the use of anesthetics, as the time of SE termination and, hence, weaning of IVADs could be determined earlier. It remains to be determined, if specific conditions and types of SE are at a higher risk for adverse effects from IVADs and if the risks from IVADs can be outweighed by the potential injury resulting from RSE in specific patients.

5 Conclusions

This two-center study using three different statistical approaches reveals independent associations between the use of continuously administered IVADs and an increased risk of death and poor functional outcome in survivors. Randomized trials are needed to determine if these associations are biased by outcome predictors not yet identified and therefore not accounted for in this study.