This retrospective cohort study was based on an already existing cohort that was previously used to study the associations of magnetic resonance imaging (MRI), electroencephalography (EEG), and new neurological deficits with long-term outcome [9, 16, 17]. In brief, all adult patients (≥ 18 years old) with first-time, nonanoxic SE (as defined in [1, 18]) who received a diagnosis and were treated at the University Hospital of Odense between January 2008 and December 2017 were retrospectively identified based on referrals for acute EEGs, from the International Classification of Diseases, Tenth Edition codes (DG40.3) at discharge and/or from clinical information by reviewing medical records. To further delineate the cohort relevant to this study examining the prognostic impact of postictal encephalopathy on long-term survival, patients who remained in SE at discharge were excluded. Further, patients without EEG and without unequivocal postictal clinical improvement after SE were excluded from analysis due to missing data (Supplementary Fig. 1). Patients with withdrawal of care during ongoing SE were excluded, whereas patients with withdrawal of care after successful treatment of SE remained in the study. The study complies with ethical standards defined by Danish legislation. Permission to handle medical records was acquired from the Danish Data Protection Agency (18/58576) and the Danish Health Authority (3-3013-2661/1). Data reporting followed the Strengthening the Reporting of Observational Studies in Epidemiology guideline [19].

For each patient, the postictal state was retrospectively estimated daily for 14 consecutive days following SE-cessation or until the patient returned to the habitual state and/or was discharged from the hospital. The time point of cessation of SE was identified using a combination of clinical and EEG-verified seizure freedom and was essentially based on the treating neurologists’ evaluation; new neurological deficits were determined by using the estimated National Institute of Health Stroke Scale score before admission and at discharge as described in [9].

The West Haven Criteria (WHC) for hepatic encephalopathy were used to quantify the severity of the clinical manifestations seen in the postictal state during the measured period [20, 21]. It comprises the following stages:

Grade 0: Asymptomatic patients.

Grade 1: Mild lack of awareness, euphoria or anxiety, shortened attention span, impaired performance of addition.

Grade 2: Lethargy or apathy, minimal disorientation for time or place, subtle personality changes, inappropriate behavior, impaired performance of subtraction.

Grade 3: Somnolence to semistupor but response to verbal stimuli, confusion, gross disorientation.

Grade 4: Coma.

Grade 5: Is not an official part the WHC but added as part of this study to account for the patients’ death when calculating the cumulative WHC grade after SE.

The evaluation was based on electronic medical records including all notes from nurses and therapist that were available for all patients. If a patient’s state varied during the day, the worst state was graded according to the WHC. All estimates of the WHC grade were performed by a single rater (C.M.B.) to avoid interrater bias. After the assessments were completed, the rater, anonymized to the results of the first assessment, evaluated the postictal state once again for the first 15 patients, which confirmed the initial results, and no changes of the initial assessment had to be made. The cumulative postictal WHC grade during the first 14 days was defined as the sum of the WHC grades from day 1 to 14. If a patient died before day 14, the patient received a score of 5 for the remaining days. If a patient was discharged before day 14, the patient was scored with a fictive score of 0 for the days after discharge. We defined a cumulative postictal WHC grade of 0–20 as “mild,” 21–40 as “moderate,” and values > 40 as “severe/prolonged” postictal encephalopathy.

In patients with persistent postictal impairment of consciousness, a 20–30-min spot EEG was routinely performed to exclude persistent nonconvulsive SE (NCSE).

Survival data were available for all patients due to the linkage of the electronic medical records and the Danish Central Person Register [6, 9, 16, 17].

Clinical characteristics of the cohort were assessed as described previously [9] and were available for this study. SE was classified as described by Trinka et al. [1].The following parameters were assessed based on the patients’ electronic medical records: sex, age, weight, Charlson Comorbidity Index (CCI) score (without age criterion) [22], SE duration, modified Rankin Scale before admission, etiology as classified in [9], diagnosis of possible NCSE based on the Salzburg criteria [16], worst seizure type [23], and history of epilepsy [23].

Data on ictal MRI changes were previously published in [17]. In brief, postictal MRI changes were identified retrospectively in patients with SE and available standard 3T MRI (diffusion-weighted images) taken under SE or within 1 week after cessation of SE (C.D.C.). All available ictal EEGs were analyzed and classified retrospectively by L.E.R., O.M., and T.K. based on the Salzburg criteria [24]. Data were previously published in Monsson et al. [16].

Cumulative doses of midazolam and propofol were obtained retrospectively using the archive function within Cambio Clinical Information System (version 4.9.0.1) by S.B.K., H.T.O., and P.T. The cumulative doses were calculated from the start of the continuous infusion of a given sedative as part of SE treatment until stop of sedation (excluding opioids). If detailed information on infusion was not available, doses were estimated using the electronic medical records, administration lists, and the average rate of actual administered sedatives in the ICU.

Data were stored using RedCap (Vanderbilt University) [25]. Statistical analyses were performed using IBM SPSS 29. Percentages and frequencies, medians/means and interquartile ranges (IQRs) were given for descriptive analysis. Long-term survival was addressed using Kaplan–Meier estimator, and log-rank test was used to determine statistical significance. p values < 0.05 were considered significant without correction for multiple testing.

Univariable comparisons of sex, age group, etiology, diagnosis of possible NCSE, worst seizure type, history of epilepsy, and the primary outcome of these groups were performed with the χ2 for categorical variables and the Kruskal–Wallis test for numeric data.

Automatic linear modeling was used for the identification of major contributor of postictal encephalopathy. The cumulative WHC was used as continuous variable and end point; age, CCI, etiological groups, cumulative midazolam and propofol doses, modified Rankin Scale before admission, duration of SE, worst seizure type, and duration of sedation were putative predictors in the model. Variables were chosen based on assumed clinical relevance and low risk of collinearity. The identified significant predictors were then used as covariates in a linear regression model (method: enter) with cumulative postictal WHC as end point. The model did not account for heteroscedasticity. Significance of the model was assessed using F-statistic and goodness-of-fit was assessed with R-square. Effect sizes were measured with coefficients β and corresponding 95% confidence intervals.