Implementation of the seizure code in a tertiary hospital: Effects on emergency care times and clinical outcomes in severe seizures

Abstract

Objective

To assess the impact of implementing the Seizure Code (SC), a structured multidisciplinary emergency protocol for severe seizures, on care times and clinical outcomes at a tertiary hospital.

Methods

Single‐center, retrospective observational study including episodes of severe seizures in patients aged 18 years or older managed between January 2023 and December 2024, comparing pre‐SC and post‐SC periods. The primary outcome was time from seizure onset to administration of the first anti‐seizure medication (ASM). Secondary outcomes included other care intervals, process indicators, and clinical outcomes. Care intervals were analyzed with accelerated failure time models (percentage change in median time), length of stay in a competing‐risks framework, and binary outcomes with Firth‐penalized logistic regression (odds ratios, OR), all adjusted for prespecified covariates and clustered by patient.

Results

Of 1959 episodes identified, 378 (175 pre‐SC, 203 post‐SC) met SC activation criteria. Time to the first ASM, the primary outcome, did not differ significantly (27% shorter, 95% CI ‐15% to 53% for benzodiazepines; 27%, −6% to 50% for others). Time to urgent video‐EEG (VEEG) was 48% shorter (29%–61%) and the neurological assessment‐VEEG interval 63% shorter (31%–80%). Pre‐notification by Emergency Medical Services increased (22.9% vs. 37.8%; p = 0.003). Hospital stay (7.48 vs. 5.08 days; subdistribution hazard ratio for discharge alive 1.36, 1.02–1.89) and intensive care unit stay (3.20 vs. 1.48 days) were shorter. Thirty‐day seizure recurrence was lower (8.0% vs. 3.0%; adjusted OR 0.33, 0.12–0.84). Mortality and functional status at discharge did not differ.

Significance

The primary outcome, time to first anti‐seizure medication, was not significantly shorter after implementation of the structured emergency protocol. The observed benefits were concentrated in secondary process and resource‐utilization indicators: implementation of the Seizure Code was associated with faster access to urgent VEEG, shorter hospital and intensive care unit stays, and lower 30‐day seizure recurrence, without a significant effect on mortality. These findings support protocolized emergency pathways in severe seizures.

Plain Language Summary

Severe seizures are emergencies in which fast, coordinated care matters. We compared how patients were treated at one hospital before and after a new protocol, the Seizure Code, was introduced. Afterwards, patients had urgent video‐EEG (a recording of brain activity) sooner and spent fewer days in hospital and in the intensive care unit, and fewer had another seizure within 30 days. However, the time to the first anti‐seizure medication did not change significantly, and deaths and disability at discharge were unchanged.

Key points.

• Implementation of the Seizure Code was associated with shorter time to urgent VEEG and lower ICU admission and hospital/ICU length of stay.

• Time to first ASM and mortality did not differ; 30‐day recurrence was lower but rests on 20 events and disappeared when death was added.

• Structured emergency pathways may improve process indicators and intermediate clinical outcomes in severe seizures.

1. INTRODUCTION

Epilepsy affects approximately 350 000 people in Spain and more than 50 million worldwide. ¹ Seizures account for up to 1% of visits to Emergency Departments (ED), representing a frequent cause of activation of emergency care resources, accounting for 9.9% of emergency visits for neurological presentations. ² , ³

Seizure duration is an important prognostic determinant and a potentially modifiable factor. Evidence shows a consistent association between the time to administration of anti‐seizure medication (ASM) and the likelihood of seizure control, particularly in status epilepticus (SE). ⁴ , ⁵ The definition of SE according to the International League Against Epilepsy (ILAE) introduces the temporal thresholds t1 and t2, marking respectively the point beyond which the seizure is unlikely to stop spontaneously and the point at which the risk of permanent neuronal injury increases. ⁵ , ⁶ Within this framework, the principle of “time is brain” fully applies to severe seizures, where therapeutic delay translates into a worse neurological prognosis. ⁷ , ⁸ , ⁹

Despite this pathophysiological and clinical rationale, considerable heterogeneity persists in the initial management of severe seizures, both in the prehospital and in‐hospital settings. In recent years, the need to define severe seizures and structure their management through specific, coordinated protocols has been proposed, although the published evidence remains limited. ¹⁰ , ¹¹ In our setting, the definition and structured approach to severe seizures have recently been described, establishing the conceptual framework for implementing a clinical management guideline. ¹²

With the aim of standardizing clinical practice in line with the best available evidence and optimizing care times, the Seizure Code (SC) protocol was implemented in the Community of Madrid (Spain) in November 2023, inspired by urgent activation models already established for other time‐dependent conditions, such as acute stroke.

The primary objective of this study was to compare care times in the management of patients with severe seizures before and after implementation of the SC, including time to assessment by Emergency Medical Services (EMS), arrival at hospital, treatment initiation, and performance of urgent video‐EEG (VEEG), as well as prehospital pharmacological treatment. Secondary objectives were to assess differences in mortality and morbidity between both periods, and 30‐day seizure recurrence and readmission.

2. MATERIALS AND METHODS

2.1. Study design and setting

This was a single‐center, retrospective observational study conducted at a tertiary hospital (Hospital Universitario Clínico San Carlos), which included episodes of severe seizures managed between 1 January 2023 and 31 December 2024. Urgent VEEG was available 24 h a day, 7 days a week, throughout the entire study period, with no differences in availability between weekdays or weekends. Two periods were compared: pre‐SC (1 January–14 November 2023) and post‐SC (15 November 2023–31 December 2024). The study was approved by the Clinical Research Ethics Committee of Hospital Universitario Clínico San Carlos (Internal Code: 25/385‐E), which waived the requirement for informed consent owing to the retrospective nature of the study and the use of anonymized data.

2.2. Population

Patients aged ≥18 years with episodes of severe seizures were included — defined as those meeting at least one of the SC activation criteria (see below), according to the consensus document of the Community of Madrid ¹² — identified through the Coding Department (diagnosis of “epileptic seizure” and/or performance of VEEG) and confirmed by manual review of medical records. Each episode was considered an independent event. Patients with a baseline functional status of modified Rankin Scale (mRS) ≥4 and/or under active follow‐up by the Palliative Care Unit were excluded. To minimize selection and information bias, all consecutive episodes meeting the SC activation criteria during the study period were included, eligibility and study variables were confirmed through manual review of medical records, and multivariable models were adjusted for prespecified, clinically relevant covariates to account for potential confounding.

2.3. Intervention

The intervention evaluated was the implementation of the SC, a structured, multidisciplinary care protocol for the acute management of severe seizures in the Community of Madrid, developed from a 2020 consensus document and launched in November 2023. ⁴ , ¹² The protocol organizes care into coordinated sub‐processes spanning the entire care pathway — prehospital care, code activation, emergency department management, hospitalization, intensive care, and discharge planning — with the aim of reducing time to treatment and standardizing clinical practice. ¹²

The SC is activated in children and adults for: (1) generalized SE or focal seizure with impaired consciousness ≥5 min (SEC‐5), consistent with the t1 threshold defined above; (2) suspected non‐convulsive SE (S‐NCSE); (3) new‐onset postictal focal neurological deficit (NPFND); or (4) cluster seizures (C‐24 h). Suspected non‐convulsive SE was defined clinically, at the discretion of the treating physician, as confusion or decreased level of consciousness without a clear alternative cause. Cluster seizures were defined as more than two tonic–clonic or focal seizures with impaired consciousness within 24 h, with recovery between episodes. ¹² Operationally, EMS activate the protocol and initiate early ASM, preferably as dual therapy (benzodiazepine [BZD] and non‐benzodiazepine ASM [non‐BZD ASM], at status doses), and pre‐notify the on‐call neurologist to mobilize the hospital pathway in advance. On arrival, the patient is managed in the emergency care area, where the necessary complementary tests are completed urgently (neuroimaging, lumbar puncture if indicated, and VEEG).

2.4. Comparator

The comparator consisted of episodes managed in the period prior to protocol implementation (pre‐SC) versus those managed after its implementation (post‐SC).

2.5. Variables and outcomes

2.5.1. Primary variable

Treatment times, defined as the interval (in minutes) from seizure onset to administration of the first ASM, differentiating BZD (S‐1st BZD ASM) and non‐BZD ASM (S‐1st non‐BZD ASM). Times recorded as 0 min correspond to rescue medication administered at seizure onset.

2.5.2. Secondary variables

• Other care times: from seizure onset to assessment by EMS (S‐EMS), by neurology (S‐NEU), by the intensive care unit (ICU; S‐ICU), and to VEEG (S‐VEEG); and from pre‐notification to neurology (Pre‐notified‐NEU) and from neurology to VEEG (NEU‐VEEG).

• Process indicators: prehospital ASM administration, prehospital dual therapy, use of BZD as first line, total number of ASMs administered, and performance of urgent VEEG.

• Clinical outcomes: in‐hospital and 30‐day mortality, functional status at discharge (mRS), total hospital and ICU length of stay, need for sedation, and 30‐day seizure recurrence and readmission.

2.6. Statistical analysis

A descriptive analysis of the sample was performed. Qualitative variables were expressed as frequencies and percentages, and quantitative variables as mean ± standard deviation or median [interquartile range, IQR], according to their distribution. Between‐group comparisons were performed using the χ² test or Fisher's exact test for qualitative variables, and Student's t‐test or the Mann–Whitney U test for quantitative variables.

Care intervals were analyzed with accelerated failure time models, whose effect measure is the time ratio (TR) with its 95% confidence interval (95% CI): a TR below 1 indicates a shorter interval in the post‐SC period. Time ratios are reported as the percentage change in median time, 100 × (1 − TR). The parametric distribution (Weibull, log‐normal or log‐logistic) was selected for each outcome by the Akaike information criterion. Cox models are reported as a sensitivity analysis and the proportional‐hazards assumption was assessed with scaled Schoenfeld residuals; it was not violated for the period term in any model. The hospital arrival‐NEU interval was not modeled, as its median was 0 min in both periods.

Length of stay was measured in days from admission to hospital discharge. No admission was censored, since every episode ended in discharge or death. Because in‐hospital death precludes discharge, length of stay was analyzed within a competing‐risks framework: we report the cause‐specific hazard ratio and the subdistribution hazard ratio for discharge alive, with 95% CIs obtained via patient‐level bootstrap (1000 resamples). As neither estimand is a ratio of times, length of stay was additionally modeled directly, using a log‐link gamma model (ratio of mean days) and median regression (ratio of median days).

Binary outcomes were analyzed with logistic regression estimated by Firth's penalized likelihood, expressed as odds ratio (OR) with profile‐likelihood 95% CI, and functional status at discharge with an ordinal logistic model expressed as a common OR. Unadjusted estimates are reported alongside the adjusted ones.

Seven covariates were prespecified: age, sex, history of epilepsy, etiology, drug resistance, baseline mRS, and Abnormal speech, ocular Deviation, Automatisms, and Number of motor epileptic seizures (ADAN) scale score. They were used in full whenever at least ten events per variable were available, which applied to the care intervals with more than 90 observations and to total hospital length of stay. For the remaining outcomes, adjustment was restricted to three prespecified covariates: age, baseline mRS, and history of epilepsy (Table S1). Cluster‐robust (sandwich) standard errors with the patient as the clustering unit were used in all maximum‐likelihood models; penalized models used profile‐likelihood intervals and the subdistribution hazard model a patient‐level bootstrap. Status epilepticus was not used for adjustment, since it is established after seizure onset and is plausibly on the causal pathway of the intervention; it was used only to define a prespecified subgroup.

In response to a peer‐review request, a post‐hoc subgroup analysis restricted to episodes meeting the C‐24 h criterion was also performed, using the same unadjusted approach as the SE subgroup. The C‐24 h criterion was additionally entered as an interaction term with period in the main adjusted models for the whole cohort, using the same covariates and estimators described above.

Analyses were based on available cases and denominators are specified for each variable. Because the availability of recorded timestamps differed between periods, the two urgent VEEG intervals and the time to the first BZD were additionally analyzed by inverse‐probability‐of‐availability weighting and by multiple imputation (predictive mean matching, m = 50, pooled by Rubin's rules), restricted to episodes in which the procedure had actually been performed. All episodes were followed for 30 days after the index event. A p‐value <0.05 was considered statistically significant.

Interrupted time‐series models and a moving‐cutpoint placebo test, used to separate a level change at implementation from pre‐existing trends, are reported as supplementary material. Only the time to the first ASM was confirmatory; all other analyses are exploratory, and Benjamini‐Hochberg‐ and Bonferroni‐adjusted p‐values for the twenty secondary contrasts are provided as supplementary material.

This study is reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement; the completed checklist is provided as supplementary material.

All analyses were performed in R 4.5.1 (R Foundation for Statistical Computing, Vienna, Austria).

3. RESULTS

3.1. Baseline characteristics

During the study period, 1959 consecutive episodes of epileptic seizures were identified in patients aged 18 years or older with baseline mRS 3 or less. Of these, 1570 did not meet SC activation criteria. A further 11 episodes were excluded for administrative reasons (1 duplicate record and 10 with data that could not be recovered), leaving 378 episodes (175 pre‐SC and 203 post‐SC) included in the analysis (Figure 1).

FIGURE 1.

Flow of episodes through the study.

Among these, the final diagnosis was epileptic seizure in 310 episodes (82.0%), a different condition in 62 (16.4%), and the final diagnosis was indeterminate in 6 (1.6%). Among the 62 episodes with a non‐epileptic final diagnosis, the most common were non‐epileptic paroxysmal events (18/62, 29.0%; including psychogenic non‐epileptic seizures and neuromediated syncope), post‐anoxic encephalopathy (6/62, 9.7%), ischemic stroke (5/62, 8.1%), and toxic‐metabolic encephalopathy (4/62, 6.5%), yielding a positive predictive value of the activation criteria for epileptic seizure of 82.0%.

The 378 episodes occurred in 344 unique patients, 15 of whom contributed more than one episode (49 episodes, 13.0%). No significant differences were observed between periods in baseline characteristics (Table 1), but the distribution of activation criteria did differ (Table S2): new‐onset postictal focal neurological deficit was less frequent after implementation (20.6% vs. 9.9%; p = 0.005) and cluster seizures more than doubled (17.7% vs. 45.8%; p < 0.001).

TABLE 1.

Baseline characteristics of the included episodes according to the period of SC implementation.

Variable	Pre‐SC (n = 175)	Post‐SC (n = 203)	p value
Age, median [IQR]	62.0 [38.5–78.5]	60.0 [40.5–78.0]	0.699
Male sex, n (%)	93 (53.1)	107 (52.7)	1.000
Known epilepsy, n (%)	69 (39.4)	94 (46.3)	0.214
Drug–resistant epilepsy, n (%)	23 (13.3)	35 (17.2)	0.361
Baseline mRS, n (%)			0.241
0	110 (62.9)	124 (61.1)
1	31 (17.7)	41 (20.2)
2	23 (13.1)	17 (8.4)
3	11 (6.3)	21 (10.3)
ADAN scale score, median [IQR]	1.00 [1.00–2.00]	1.00 [1.00–2.00]	0.697
Alteration in level of consciousness, n (%)	43 (24.7)	82 (40.6)	0.002

Note: Values are presented as median [IQR] or n (%), as appropriate. Percentages were calculated on the number of episodes with available data for each variable.

Abbreviations: ADAN, Abnormal speech, ocular Deviation, Automatisms, and Number of motor epileptic seizures scale; mRS, modified Rankin Scale.

3.2. Clinical presentation and etiology

Seizures with impaired level of consciousness were more frequent after SC implementation (24.7% vs. 40.6%; p = 0.002). The distribution of the ADAN scale score differed significantly between periods (p = 0.028), with no difference in the median total score (p = 0.698).

The structural etiology group was the most frequent in both periods (44.6% pre‐SC and 36.9% post‐SC), within which ischemic lesion was the predominant individual cause.

Etiology was unknown or classified as other in 41.1% and 39.4% of episodes, metabolic, toxic or systemic infectious causes accounted for 10.3% and 16.3%, and etiology was not recorded in 4.0% and 7.4%. The distribution across these groups did not differ significantly between periods (p = 0.15; Table S3).

3.3. Care activation and coordination

After SC implementation, the proportion of episodes pre‐notified by EMS increased from 22.9% (40/175) to 37.8% (76/201; p = 0.003). Of the 76 pre‐notified post‐SC episodes, 50.0% (n = 38) were channeled through a specific SC activation, 48.7% (n = 37) via the Stroke Code protocol and 1.3% (n = 1) as a direct call to the ICU. In the pre‐SC period, the 40 pre‐notified episodes had been guided through the Stroke Code pathway (n = 28), a direct call to the ICU (n = 11) or an early SC activation (n = 1). Episodes not pre‐notified through any of these pathways were transferred directly to the ICU owing to a low level of consciousness.

3.4. Care times and processes

Median care times decreased across most intervals (Table 2). In the adjusted analysis, time from seizure onset to urgent VEEG was 48% shorter (95% CI 29%–61%; median 255 vs. 170 min; p < 0.001) and the in‐hospital interval between neurological assessment and VEEG was 63% shorter (31% to 80%; 74 vs. 32 min; p = 0.002). The interval from seizure onset to neurological assessment, which was not significant in the unadjusted comparison, was 31% shorter after adjustment (4% to 51%; 165 vs. 120 min; p = 0.027).

TABLE 2.

Care intervals according to the period of SC implementation.

Interval	Pre‐SC median [IQR], min	Post‐SC median [IQR], min	n pre/post	Unadjusted p	Time ratio (95% CI)	Reduction in median time	Adjusted p
S‐EMS	66 [39–118]	45 [25–90]	46/73	0.122	0.65 (0.41 to 1.04)	35% (−4 to 59)	0.074
S‐1st BZD ASM	103 [1–231]	55 [23–164]	71/115	0.845	0.73 (0.47 to 1.15)	27% (−15 to 53)	0.182
S‐1st non‐BZD ASM	124 [57–260]	76 [40–194]	74/116	0.182	0.73 (0.50 to 1.06)	27% (−6 to 50)	0.099
Pre‐notified‐NEU	30 [22–40]	32 [20–51]	18/56	0.905	0.90 (0.58 to 1.41)	10% (−41 to 42)	0.649
S‐NEU	165 [82–280]	120 [66–249]	61/122	0.136	0.69 (0.49 to 0.96)	31% (4 to 51)	0.027
S‐ICU	90 [64–327]	86 [47–132]	15/20	0.473	0.27 (0.09 to 0.80)	73% (20 to 91)	0.018
S‐VEEG	255 [172–705]	170 [101–318]	61/114	<0.001	0.52 (0.39 to 0.71)	48% (29 to 61)	<0.001
NEU‐VEEG	74 [16–240]	32 [10–64]	47/110	0.018	0.37 (0.20 to 0.69)	63% (31 to 80)	0.002

Note: Unadjusted p from the Mann–Whitney U test. Time ratios from accelerated failure time models with cluster‐robust standard errors by patient; a time ratio below 1 indicates a shorter interval after implementation. The hospital arrival‐NEU interval is not shown, as its median was 0 min in both periods. Model n, number of parameters, events per variable and the covariate set used for each interval are given in Supplementary Table S1.

Abbreviations: ASM, anti‐seizure medication; BZD, benzodiazepine; EMS, emergency medical services; ICU, intensive care unit; NEU, neurologist; S, seizure onset; VEEG, video‐EEG.

No significant difference was found in the primary outcome: time to the first BZD was 27% shorter (−15% to 53%; 103 vs. 55 min; p = 0.182) and time to the first non‐BZD ASM 27% shorter (−6% to 50%; 124 vs. 76 min; p = 0.099). With the available sample, only reductions of 51% and 44% in these two intervals would have been detectable with 80% power. The difference in time to EMS assessment did not persist once the prespecified covariates and clustering by patient were applied (35% shorter, −4% to 59%; p = 0.074), nor did the pre‐notified‐NEU interval (10% shorter, −41% to 42%; p = 0.649). The interval from seizure onset to ICU assessment was 73% shorter (20% to 91%; p = 0.018), but this estimate is based on 35 episodes and is not supported by the unadjusted comparison (90 vs. 86 min; p = 0.47), and should be regarded as exploratory. The hospital arrival‐NEU interval was not modeled, as its median was 0 min in both periods (102 of 171 episodes were assessed immediately). The proportion of episodes with urgent VEEG increased from 67.4% to 81.8% (adjusted OR 2.59, 1.57–4.35; p < 0.001) (Figure 2).

FIGURE 2.

Adjusted change in care intervals after implementation of the Seizure Code. Points are time ratios from accelerated failure time models with cluster‐robust standard errors by patient, adjusted for the seven prespecified covariates; the dagger marks the two intervals with fewer than ten events per variable, adjusted for age, baseline mRS and prior epilepsy only. A time ratio below 1 indicates a shorter interval after implementation: A time ratio of 0.52 corresponds to a median interval 48% shorter. Medians are computed on all episodes with a documented time. The hospital arrival‐NEU interval is not shown, as its median was 0 min in both periods.

3.5. Prehospital treatment

After SC implementation, a trend towards greater use of prehospital ASM by EMS was observed (25.5% vs. 35.1%; p = 0.077). Among treated patients, BZDs were used as first‐line ASM more frequently (72.2% vs. 82.7%; p = 0.025), with a significant change in the BZD used (p < 0.001), driven by greater use of midazolam (19.4% vs. 37.8%) over lorazepam (41.1% vs. 25.6%). Prehospital dual therapy was also more frequent, without reaching significance (12.3% vs. 19.4%; p = 0.108; Table S4). Taken together, these findings reflect an earlier therapeutic approach in line with the protocol, although several of the changes did not reach statistical significance and prehospital ASM administration was already increasing before implementation.

3.6. Clinical and prognostic outcomes

No differences were observed in in‐hospital mortality between periods (5.8% vs. 6.4%; adjusted OR 1.17 [0.50–2.80]; p = 0.719) nor in 30‐day mortality (5.8% vs. 7.4%; adjusted OR 1.37 [0.60–3.23]; p = 0.454). Functional status at discharge also did not differ after adjustment (common OR 0.88 [0.58–1.35]; p = 0.570; mRS 0–2 in 77.3% vs. 80.3%) (Figure 3).

FIGURE 3.

Distribution of the modified Rankin Scale at discharge, by period. Numbers inside the bars are episode counts. Three pre‐SC episodes had no mRS recorded at discharge and are excluded from the denominator.

The post‐SC period was associated with a lower probability of 30‐day seizure recurrence (14/175, 8.0% vs. 6/203, 3.0%; adjusted OR 0.33 [0.12–0.84]; p = 0.019). The difference was also present in the unadjusted comparison (OR 0.37 [0.13–0.91]; p = 0.031), but it did not persist when death was added to the endpoint (recurrence or death within 30 days: 24/175, 13.7% vs. 21/203, 10.3%; adjusted OR 0.74 [0.39–1.37]; p = 0.330). Of the 20 episodes with 30‐day seizure recurrence, 4 (20.0%) occurred during the index hospitalization (in‐hospital recurrence: 4 pre‐SC and 0 post‐SC) and 16 (80.0%) after discharge (post‐discharge recurrence, via ED visit or readmission: 10 pre‐SC and 6 post‐SC).

Total hospital length of stay was shorter after implementation (median 7.48 vs. 5.08 days): the cause‐specific hazard of discharge alive was 1.55 (1.15–2.09; p = 0.004) and the subdistribution hazard 1.36 (1.02–1.89; p = 0.040); modeled directly, the mean stay was 38% shorter (15% to 55%; p = 0.003) and the median 29% shorter (6% to 47%; p = 0.017). ICU length of stay was also shorter (3.20 vs. 1.48 days; cause‐specific hazard 1.76 [1.04–2.98], p = 0.034; subdistribution hazard 1.79 [1.02–3.42], p = 0.042; median 49% shorter [4% to 73%], p = 0.036), although the ratio of mean days did not reach significance (0.63 [0.37–1.09]; p = 0.100). The need for ICU admission was lower (26.3% vs. 15.8%; adjusted OR 0.54 [0.32–0.90]; p = 0.018), with no difference in the requirement for pharmacological sedation (76.1% vs. 68.8%; p = 0.646).

The results of the clinical outcome analysis are summarized in Tables 3 and 4.

TABLE 3.

Binary clinical outcomes according to the period of SC implementation.

Outcome	Pre‐SC, n (%)	Post‐SC, n (%)	Unadjusted OR (95% CI)	Adjusted OR (95% CI)	Adjusted p	Events	Events per variable
In‐hospital mortality	10 (5.8)	13 (6.4)	1.10 (0.48–2.58)	1.17 (0.50–2.80)	0.719	23	5.8
30‐day mortality	10 (5.8)	15 (7.4)	1.28 (0.57–2.96)	1.37 (0.60–3.23)	0.454	25	6.2
30‐day seizure recurrence	14 (8.0)	6 (3.0)	0.37 (0.13–0.91)	0.33 (0.12–0.84)	0.019	20	5.0
30‐day ED readmission	31 (18.9)	24 (12.8)	0.63 (0.35–1.12)	0.65 (0.36–1.16)	0.144	55	13.8
ICU admission	46 (26.3)	32 (15.8)	0.53 (0.32–0.87)	0.54 (0.32–0.90)	0.018	78	19.5
Urgent VEEG performed	118 (67.4)	166 (81.8)	2.15 (1.35–3.48)	2.59 (1.57–4.35)	<0.001	94	23.5

Note: Odds ratios from logistic regression estimated by Firth's penalized likelihood with profile‐likelihood confidence intervals, adjusted for age, baseline mRS and history of epilepsy. Functional status at discharge: common OR 0.88 (0.58–1.35), p = 0.570 (ordinal logistic model adjusted for the seven prespecified covariates); mRS 0–2 in 77.3% pre‐SC vs 80.3% post‐SC.

TABLE 4.

Length of stay: Competing‐risks and direct models.

Outcome	Pre‐SC median [IQR], days	Post‐SC median [IQR], days	n pre/post	Deaths	Cause‐specific HR of discharge alive (95% CI)	p	Fine–Gray sHR (95% CI)	p
Total hospital length of stay	7.48 [3.93–13.01]	5.08 [2.93–8.20]	98/111	23	1.55 (1.15 to 2.09)	0.004	1.36 (1.02–1.89)	0.040
ICU length of stay	3.20 [1.41–6.54]	1.48 [0.80–4.03]	44/32	13	1.76 (1.04 to 2.98)	0.034	1.79 (1.02–3.42)	0.042

Outcome	Ratio of mean days (95% CI)	p	Ratio of median days (95% CI)	p	Model n	Adjustment
Total hospital length of stay	0.62 (0.45–0.85)	0.003	0.71 (0.53–0.94)	0.017	198	7 covariates
ICU length of stay	0.63 (0.37–1.09)	0.100	0.51 (0.27–0.96)	0.036	76	3 covariates

Note: Hazard ratios above 1 indicate faster discharge; ratios of days below 1 indicate a shorter stay. In‐hospital death was treated as a competing event. 95% CIs for the Fine–Gray model from a patient‐level bootstrap with 1000 resamples. The ICU length‐of‐stay analysis uses in‐hospital death as the competing event, since death occurring specifically in the ICU was not recorded.

The monthly interrupted time‐series analysis is shown in Figure S1. The proportion of urgent VEEG had no pre‐existing trend (+0.16 percentage points per month; p = 0.906) and increased progressively after implementation (+1.69 points per month), and neither mortality nor length of stay showed a significant pre‐existing trend. In contrast, pre‐notification by EMS (+1.57 points per month; p = 0.020) and prehospital ASM administration (+1.87 points per month; p = 0.003) were already rising beforehand, and the time from seizure onset to VEEG was already falling (−0.40 h per month; p = 0.049). No level change at implementation was robust to excluding the transition month or a three‐month run‐in period, and a moving‐cutpoint placebo analysis did not localize the change at the implementation date (Figure S2).

3.7. Subgroup analyses

3.7.1. Clinical subgroup: Status epilepticus

In the subgroup of episodes meeting criteria for status epilepticus (56 pre‐SC and 65 post‐SC), care times moved in the same direction as in the whole cohort but no comparison reached significance: NEU‐VEEG 92.0 vs. 32.5 min (p = 0.074), S‐EMS 65.5 vs. 34.5 min (p = 0.053) and S‐VEEG 290 vs. 168 min (p = 0.229). No differences were observed either in 30‐day seizure recurrence (6.2% vs. 3.4%; p = 0.655), in‐hospital mortality (13.2% vs. 9.2%; p = 0.562), total hospital length of stay (6.78 vs. 6.28 days; p = 0.762) or ICU length of stay (2.12 vs. 2.92 days; p = 1.000). The proportion of urgent VEEG (67.9% vs. 78.5%; p = 0.218) and of prehospital dual therapy (22.6% vs. 22.8%; p = 1.000) did not differ between periods (Table S5). With this number of episodes and events the subgroup cannot support the models used in the whole cohort, so the comparison is reported unadjusted and should be regarded as descriptive and exploratory.

3.7.2. Implementation subgroup: Pre‐notified seizure code

A descriptive analysis was performed on episodes with formal pre‐notified activation as SC after implementation (n = 38), comparing them with the pre‐SC episodes that had also been pre‐notified (n = 40), since pre‐notification is not random. Care times were shorter among formally activated episodes: S‐EMS 93.0 vs. 37.0 min (p < 0.001), S‐VEEG 237 vs. 147 min (p = 0.002) and NEU‐VEEG 54.0 vs. 27.0 min (p = 0.026). The proportion of urgent VEEG (75.0% vs. 94.7%; p = 0.026) and of prehospital dual therapy (33.3% vs. 65.7%; p = 0.010) was also higher. With this comparator, however, no differences were observed in ICU length of stay (1.43 vs. 1.24 days; p = 0.648), total hospital length of stay (6.90 vs. 5.74 days; p = 0.375), 30‐day seizure recurrence (2.6% vs. 0%; p = 1.000) or in‐hospital mortality (2.7% vs. 2.6%; p = 1.000) (Table S6).

3.7.3. Clinical subgroup: Cluster seizures (C‐24 h)

Before SC implementation, 29 of the 31 cluster episodes also met another activation criterion; after implementation, this was the case in 47 of 93 episodes, with the remainder meeting the C‐24 h criterion alone. SE was less frequent among cluster episodes post‐SC (58.1% vs. 38.7%; p = 0.093). Baseline characteristics did not differ between periods. Care times moved in the same direction as in the whole cohort but did not reach significance, the largest change being in S‐VEEG (290 vs. 161 min; p = 0.052). VEEG performance (71.0% vs. 83.9%; p = 0.124) and EMS pre‐notification (19.4% vs. 37.6%; p = 0.078) did not differ significantly. ICU admission was lower post‐SC (35.5% vs. 18.3%; p = 0.080), with no differences in total hospital stay (6.28 vs. 5.78 days; p = 0.430) or ICU stay (1.96 vs. 1.41 days; p = 0.353). 30‐day seizure recurrence was 13.3% vs. 3.4% (p = 0.070), 30‐day ED readmission 23.3% vs. 12.6% (p = 0.238), and in‐hospital mortality 3.4% vs. 6.5% (p = 1.000) (Table S7). In the whole cohort, none of the cluster‐by‐period interaction terms reached significance, for either care times or binary outcomes (all p ≥ 0.23; Table S8).

4. DISCUSSION

After the SC was introduced, a consistent reduction was observed across multiple care intervals in severe seizures, most clearly in access to urgent VEEG monitoring. A decrease in total hospital and ICU length of stay was also observed, together with lower 30‐day seizure recurrence. No differences were found in in‐hospital mortality or functional status at discharge. In addition, a trend towards an increase in patients receiving early prehospital treatment was observed, a factor that has been shown to be a determinant of patient prognosis.

The primary outcome, time to first ASM, did not reach statistical significance, despite an observed reduction of approximately 27% for both BZD and non‐BZD ASM. Given the available sample size, only reductions of approximately 51% and 44%, respectively, could have been detected with 80% power. The absence of statistical significance should therefore not be interpreted as evidence that the SC had no effect on time to treatment; it may instead reflect insufficient statistical power to detect an effect of this magnitude. However, a reduction of this order—from 103 to 55 min for BZD and from 124 to 76 min for non‐BZD ASM—may be clinically relevant in a time‐dependent condition such as SE, in which shorter treatment delays have been linked to shorter seizure duration, lower risk of progression to refractory status, and better functional outcome. This effect appeared more pronounced among formally pre‐notified episodes (see Subgroup analyses). Contamination of the pre‐SC period by anticipatory practice changes, discussed further below, offers an additional possible explanation for this non‐significant result (see Limitations).

This rationale is consistent with the ILAE's operational definition of SE, based on the t1 and t2 temporal thresholds, ⁶ and with previous studies conducted at our center showing that therapeutic delay was associated with longer episode duration and worse clinical outcome. ⁵ In this context, the reduction observed in the S‐EMS, S‐VEEG, and NEU‐VEEG intervals is consistent with the pathophysiological hypothesis that early intervention may limit seizure perpetuation. Although statistical significance was reached only for certain key intervals, the remaining care times showed a consistent trend in the same direction after the SC was implemented.

Beyond its effect on care times, earlier access to urgent VEEG carries direct clinical implications. It facilitates the identification of non‐convulsive status epilepticus in patients with persistent altered consciousness, allows earlier optimization or de‐escalation of ASM therapy, and may help avoid unnecessary pharmacological sedation or ICU admission when electroclinical monitoring shows resolution of ictal activity. Earlier VEEG also supports a more precise etiological work‐up, guiding ancillary testing and specific treatment when needed. These mechanisms are consistent with the reductions in ICU admission and length of stay observed with the SC, and reinforce the clinical relevance of the observed reduction in VEEG delays beyond its value as a process indicator.

Our results are in line with international initiatives aimed at structuring emergency epilepsy care. The Finnish model emphasizes care coordination and organizational care protocols, ¹¹ while the “Seizure Emergency Code” strategy implemented in Bogotá ¹⁰ demonstrated reductions in treatment times and hospital‐level improvements. Unlike these studies, our study incorporates a detailed, multistage analysis of several care intervals, allowing assessment of the protocol's impact at different levels along the care pathway, in both prehospital and in‐hospital settings.

The reduction in 30‐day seizure recurrence is one of the most clinically appealing findings, but also one of the least robust, and should be regarded as exploratory and hypothesis‐generating rather than confirmatory. This effect could be related to more effective control of the index episode and to earlier electroclinical characterization through urgent VEEG, facilitating treatment adjustments before discharge. It rests, however, on only 20 events, and it disappears when death is incorporated into the endpoint; accordingly, this finding should be confirmed in prospective, multicenter studies before it can inform clinical practice.

In addition, the reduction in hospital and ICU length of stay may have further clinical implications, since prolonged stays are associated with a higher risk of medical complications, nosocomial infections, and functional decline. ¹³ The reduction in ICU length of stay, in the absence of differences in the need for pharmacological sedation, suggests an optimization of the clinical course without an increase in invasive interventions, although only in‐hospital death and not death occurring in the ICU was recorded.

Beyond the reduction in length of stay, the decrease in ICU admission itself (26.3% vs. 15.8%; adjusted OR 0.54, 95% CI 0.32–0.90; p = 0.018) is arguably one of the most clinically relevant findings of this study, given its implications for critical‐care resource utilization and its persistence after adjustment for prespecified covariates. This finding is biologically plausible in light of the improvements observed in care coordination, earlier neurological assessment, and faster access to urgent VEEG. As with the other before‐after comparisons in this study, however, residual confounding cannot be excluded.

Regarding mortality, we observed no significant differences, but our figures are lower than those reported in the literature in recent years, ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ including studies conducted at our center in which in‐hospital mortality exceeded 20%. ⁵ Although the cohorts are not directly comparable, mortality observed in our SE subgroup was lower in both periods, which may reflect evolving changes in overall clinical practice and therapeutic support. Because mortality may be an outcome that is less responsive to short‐term organizational changes, the shorter hospital and ICU stays observed with the SC could plausibly translate into a longer‐term mortality benefit, insofar as they reflect less frequent progression to more severe clinical courses. Nevertheless, mortality in severe seizures depends largely on the underlying etiology and the patient's baseline status, factors that may limit the impact of an organizational protocol on vital outcomes.

Regarding functional status at discharge, the distribution of the mRS did not differ between periods after adjustment (common OR 0.88 [0.58–1.35]), and the graphical representation should not be interpreted as suggesting an effect of the protocol.

The subgroup analysis showed that, in status epilepticus, care times maintained a favorable direction following introduction of the SC, although with 56 and 65 episodes no comparison reached statistical significance.

Episodes with formal SC activation showed markedly shorter care intervals and greater use of urgent VEEG and of prehospital dual therapy than pre‐notified episodes managed before implementation. Once the comparator was restricted to pre‐notified episodes, however, the differences in ICU length of stay and in mortality disappeared, indicating that the apparent outcome advantage of formal activation largely reflects the selection of pre‐notified patients. These findings still emphasize the importance of correct activation of the prehospital pathway and of optimizing treatment during the early prehospital phase, but the analysis of this subgroup was descriptive, involved few events, and should be considered exploratory. Collectively, these observations suggest that the benefit associated with the SC may depend not merely on the existence of a written protocol, but on the effectiveness of its earliest components (prompt recognition of the event, prehospital notification, and timely coordination of the receiving team).

Overall, our findings suggest that structured emergency care protocols in severe seizures may improve process indicators and certain intermediate clinical outcomes, reinforcing the applicability of the “time is brain” concept in the field of SE.

Among the limitations, the retrospective design affected the reconstruction of care times unevenly across periods: the availability of recorded timestamps improved after the SC was introduced, so the analyses of care intervals are therefore based on a subset of episodes that was not randomly selected. Additionally, 18.0% of SC activations did not ultimately correspond to a confirmed epileptic seizure (16.4% non‐epileptic, 1.6% indeterminate), reflecting the inherent trade‐off of a time‐critical activation protocol designed for high sensitivity. In addition, as this is a recently implemented protocol, a learning‐curve phase with non‐systematic activation of prehospital pre‐notification was likely present during the first months, which could have attenuated the observed effect. Finally, the single‐center nature of the study limits the generalizability of the results.

In addition, the marked shift in activation criteria between periods—C‐24 h more than doubling while NPFND declined by half—cannot be fully explained by our data; in the authors' interpretation, this may partly reflect increased EMS familiarity with the C‐24 h criterion following training initiatives conducted during SC implementation, rather than a genuine change in the underlying population of severe seizures, though this explanation remains speculative.

The before‐after design also cannot differentiate the effect of the protocol from concurrent secular trends: pre‐notification by EMS and prehospital ASM administration were already increasing, and the time to VEEG was already decreasing before implementation, and a moving‐cutpoint placebo test did not localize the change at the implementation date. A plausible explanation is that, as a pioneer site, informal practice changes anticipating the SC were already under way before its official launch on 14 November 2023, so the pre‐SC period may not represent a genuinely unexposed baseline; this would bias estimates of the primary outcome towards the null. In addition, the primary outcome was underpowered, the models for the rarest outcomes rest on 23 to 25 events and are therefore imprecise; twenty secondary contrasts were tested without prespecified multiplicity control, and the proportion of cluster seizures rose from 17.7% to 45.8% between periods.

5. CONCLUSION

The primary outcome of this study, time to ASM, did not reach statistical significance, although the study was powered to detect only large reductions and, as discussed above, such a reduction in time to ASM is probably clinically relevant in a time‐dependent condition such as SE. Implementation of the SC at our center was, nonetheless, associated with reduced times across several secondary care intervals and process indicators for severe seizures — most marked in access to urgent VEEG — as well as fewer ICU admissions, shorter hospital and ICU length of stay, and a lower 30‐day seizure recurrence, with no demonstrated impact on mortality or disability at discharge.

These findings support the value of structured care protocols in the emergency management of epileptic seizures and suggest, on an exploratory basis, that optimizing prehospital activation could enhance its effect. The reduction in 30‐day seizure recurrence, in particular, should be regarded as hypothesis‐generating pending confirmation. Prospective, multicenter studies with contemporaneous control groups and standardized data collection will be needed to confirm these results, assess their long‐term impact, and further characterize clinically meaningful outcomes such as ICU admission, ICU length of stay, and urgent VEEG utilization.

AUTHOR CONTRIBUTIONS

María Malaret‐Segurado: Conceptualization, Data curation, Investigation, Methodology, Visualization, Writing – original draft. Javier Ortega‐Macho: Conceptualization, Data curation, Investigation, Methodology, Visualization, Writing – original draft. Adrián Valls‐Carbó: Formal analysis, Methodology, Software, Validation, Writing – review & editing. Beatriz Parejo‐Carbonell: Data curation, Investigation, Resources, Writing – review & editing. María Romeral‐Jiménez: Data curation, Investigation, Resources, Writing – review & editing. Clara Lastras‐Fernández‐Escandón: Data curation, Investigation, Resources, Writing – review & editing. Pablo Mayo‐Rodríguez: Data curation, Investigation, Resources, Writing – review & editing. Álvaro Gutiérrez‐Viedma: Data curation, Investigation, Resources, Writing – review & editing. Irene García‐Morales: Conceptualization, Formal analysis, Methodology, Project administration, Supervision, Writing – review & editing. All authors contributed to data interpretation, critically revised the manuscript for important intellectual content, approved the final version, and agree to be accountable for all aspects of the work.

FUNDING INFORMATION

This research was funded by the Instituto de Salud Carlos III (ISCIII) through project PI24/01162 and co‐funded by the European Union.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

ETHICAL PUBLICATION STATEMENT

We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.

Supporting information

Appendix S1. STROBE checklist for observational studies.

Table S1. Model specification, sample size and events per variable.

Table S2 Seizure Code activation criteria, by period.

Table S3 Etiology, by period.

Table S4 Process indicators according to the period of Seizure Code implementation.

Table S5 Status epilepticus subgroup: unadjusted comparison between periods.

Table S6 Pre‐notified episodes: formal Seizure Code activation versus pre‐notified pre‐SC episodes.

Table S7 Seizure cluster subgroup: unadjusted comparison between periods.

Table S8 Effect modification by cluster seizures in the whole cohort.

Figure S1. Monthly interrupted time‐series analysis.

Figure S2. Moving‐cutpoint placebo analysis.

Figure S3. Cumulative incidence functions for discharge alive and in‐hospital death.

Figure S4. Adjustment for multiplicity across the outcome families.

DATA AVAILABILITY STATEMENT

The data supporting the findings of this study are available from the corresponding author upon reasonable request. The data are not publicly available because they contain information that could compromise participant privacy.