Evidance Health Sciences
3122-3354
Evidance Scientific Organization
944313
10.65416/ehealthsci.2026.944313
Target Trial Emulation Meta-Analysis of Benzodiazepines For Out-of-Hospital Status Epilepticus in Adults
research
Neurology & Neuroscience
Emergency Medicine
Critical Care Medicine
Mona Abdullrahman
Alromaihi
College of Medicine, Qassim University, Buraydah, Saudi Arabia
Mona.alromaihi@qu.edu.sa
Conceptualization
Methodology
Formal Analysis
Investigation
Data Curation
Visualization
Writing - Original Draft
Project Administration
Yazan J.
Alalwani
0009-0005-9429-0778
College of Medicine, Imam Abdulrahman Bin Faisal University, Khobar, Saudi Arabia
yazanalalwani518@gmail.com
Methodology
Validation
Investigation
Data Curation
Writing - Original Draft
Manal Mudhhi
Almazrui
College of Medicine, King Abdulaziz University, Jeddah, Saudi Arabia
Mhamadalmazroai@stu.kau.edu.sa
Validation
Investigation
Data Curation
Writing - Original Draft
Layan Bandar
Alzahrani
College of Medicine, Imam Abdulrahman Bin Faisal University, Khobar, Saudi Arabia
laly11069@gmail.com
Investigation
Data Curation
Writing - Original Draft
Abdulaziz A.
Alzahrani
Department of Pediatrics; Faculty of Medicine, King Abdulaziz University, Jeddah, Saudi Arabia
azizkas7007@gmail.com
Investigation
Data Curation
Writing - Original Draft
Alwaleed
Alotaibi
College of Medicine, Qassim University, Buraydah, Saudi Arabia
Alwaled22saud@gmail.com
Investigation
Data Curation
Writing - Review & Editing
Shahad Tariq
Harun
College of Medicine, University of Tabuk, Tabuk, Saudi Arabia
Shahad.h66@hotmail.com
Investigation
Data Curation
Writing - Review & Editing
Dena Nasser
Alhadlah
College of Medicine, King Saud Bin Abdulaziz University For Health Sciences, Riyadh, Saudi Arabia
denannassere@gmail.com
Investigation
Data Curation
Writing - Review & Editing
Arwa Ali
Alshehri
College of Medicine, King Khalid University, Abha, Saudi Arabia
maarar2004@gmail.com
Investigation
Writing - Review & Editing
Naif Abdullah
Alrommani
College of Medicine, Sulaiman Al Rajhi University, Al Bukayriyah, Saudi Arabia
Naif3764@gmail.com
Investigation
Writing - Review & Editing
Mohammed Salem
Barabea
College of Medicine, Fakeeh College for Medical Sciences, Jeddah, Saudi Arabia
Mohammedbarabea7@outlook.com
Investigation
Writing - Review & Editing
Ethar Ghazi
Alharbi
College of Medicine, Ibn Sina National College For Medical Studies, Jeddah, Saudi Arabia
Etharrghazi@gmail.com
Investigation
Writing - Review & Editing
Ahmed Y.
Azzam
0000-0002-4256-0159
Division of Global Health and Public Health, School of Nursing, Midwifery and Public Health, University of Suffolk, Ipswich, United Kingdom; Department of Neuroradiology, Rockefeller Neuroscience Institute, West Virginia University, Morgantown, United States
ahmedyazzam@gmail.com
Conceptualization
Methodology
Software
Formal Analysis
Validation
Resources
Visualization
Supervision
Project Administration
Writing - Review & Editing
Writing - Original Draft
Funding Acquisition
2026
04
02
16
https://ehealthsciences.com/published/EHS_2026_944313.pdf
https://ehealthsciences.com/articles/944313
Introduction: Benzodiazepines represent the first-line management for status epilepticus; however, heterogeneity in study designs has resulted in conflicting efficacy estimates. We aimed to conduct a target trial emulation meta-analysis to quantify design-induced bias and estimate the efficacy of benzodiazepines for adults in out-of-hospital settings under an ideal double-blind randomized controlled trial (RCT) setting.
Methods: Following PRISMA 2020 guidelines, we searched PubMed, Scopus, Web of Science, Cochrane Library, and Google Scholar up to November 25, 2025. Studies evaluating benzodiazepines for status epilepticus in adults were included. The T3-Meta framework was utilized to model design features as bias covariates, with the target trial effect (θ*) representing seizure cessation rates under ideal RCT conditions. Network meta-analysis was used to evaluate comparative effectiveness.
Results: Fourteen studies (2,803 adult patients) were included. Traditional random-effects pooling demonstrated a seizure cessation rate of 68.8% (95% confidence interval [CI]: 63.2–74.0%, I²=82.0%). Target trial analysis revealed significant open-label bias (β=0.757, odds ratio [OR]=2.13, P=0.048), with the bias-adjusted θ* of 64.9% (95% CI: 53.9–74.6%, I²=41.5%). Design features explained 49.4% of the between-study heterogeneity. Network meta-analysis demonstrated midazolam superiority over lorazepam (OR=1.60, P=0.001) and diazepam (OR=2.21, P=0.002). Midazolam achieved the highest P-Score (96.2%), followed by lorazepam (50.9%) and diazepam (2.9%).
Conclusions: Traditional meta-analysis was found to overestimate benzodiazepine efficacy by 3.9 percentage points due to open-label design bias. Intramuscular midazolam demonstrated superior effectiveness compared to intravenous lorazepam and diazepam for status epilepticus in adults.
Benzodiazepines
Epilepsy
Status Epilepticus
Seizures
Target Trial Emulation
CC-BY-4.0
https://ehealthsciences.com/crossmark-policy
10.65416/ehealthsci.2026.944313
Introduction
Status epilepticus represents a neurological emergency characterized by prolonged or recurrent seizure activity, affecting approximately 50 per 100,000 individuals annually worldwide. The condition carries significant morbidity and mortality, with outcomes highly dependent on rapid seizure termination. Benzodiazepines have remained the cornerstone of first-line management for status epilepticus for decades, acting through the enhancement of gamma-aminobutyric acid receptor-mediated inhibition to terminate seizure activity. Despite their established role in management guidelines, significant uncertainty persists regarding the comparative effectiveness of different benzodiazepine agents and optimal administration routes [1-3].
The evidence base for benzodiazepine treatment in status epilepticus has accumulated from studies with markedly different methodological approaches, ranging from double-blind randomized controlled trials (RCTs) to observational cohorts and open-label trials. This heterogeneity in study designs has resulted in conflicting efficacy estimates, with reported seizure cessation rates varying from 42% to 95% across published studies [4,5]. Traditional meta-analytic approaches treat this heterogeneity as statistical noise, pooling estimates without accounting for systematic biases introduced by study design features. However, methodological research has demonstrated that open-label designs, observational studies, and studies with non-blinded outcome assessment can inflate treatment effect estimates through performance bias, detection bias, and confounding [6-8].
The target trial emulation framework provides a structured approach to evaluate and integrate evidence from studies that deviate from an ideal RCT design. This approach defines a hypothetical target trial representing the ideal study one would conduct if resources and ethics permitted, then systematically evaluates how each study's design features introduce bias relative to this target [9-13]. By modeling design features as covariates in meta-regression, the target trial effect (θ*) can be estimated, representing the treatment effect expected under ideal conditions. This framework has been successfully applied in pharmacoepidemiology, however, its application to meta-analyses of emergency neurological interventions remains limited.
Multiple gaps remain in our understanding of benzodiazepine effectiveness for status epilepticus. First, the magnitude of design-induced bias across studies has not been quantified. Second, the true efficacy expected under ideal double-blind RCT conditions remains uncertain. Third, comparative effectiveness rankings among benzodiazepine agents require evaluation using network meta-analysis methods that account for methodological heterogeneity. Additionally, identifying whether the route of administration impacts the magnitude of design bias has implications for both clinical management and future research design.
To address these evidence gaps, we aimed to conduct a target trial emulation meta-analysis to evaluate the efficacy of benzodiazepines for status epilepticus in adults in out-of-hospital settings. Our objectives were fourfold: first, to quantify traditional pooled efficacy estimates and heterogeneity; second, to estimate the bias-adjusted target trial effect representing efficacy under ideal double-blind RCT conditions; third, to evaluate comparative effectiveness among benzodiazepine agents using network meta-analysis; and fourth, to assess the proportion of heterogeneity attributable to design features versus true variation in effects. We hypothesized that open-label and observational designs would significantly overestimate benzodiazepine efficacy compared to double-blind RCTs.
Materials and Methods
Search Strategy and Study Selection:
This systematic review and meta-analysis was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines [14]. We performed a structured literature search in PubMed, Scopus, Web of Science, Cochrane Central Register of Controlled Trials (CENTRAL), and Google Scholar from database inception up to November 25, 2025, focusing on English-language studies. The search strategy utilized a combination of Medical Subject Headings (MeSH) terms and free-text keywords designed to capture all relevant studies investigating benzodiazepine treatment for status epilepticus. The search string included the following terms: ("status epilepticus" OR "convulsive status epilepticus" OR "generalized convulsive status epilepticus" OR "prolonged seizure" OR "continuous seizure" OR "refractory seizure" OR "seizure emergency") AND ("benzodiazepine" OR "midazolam" OR "lorazepam" OR "diazepam" OR "clonazepam" OR "clobazam") AND ("treatment" OR "therapy" OR "management" OR "efficacy" OR "effectiveness" OR "cessation" OR "termination" OR "control"). Reference lists of included studies and previous systematic reviews were manually screened to identify additional eligible studies not captured by electronic searches.
We first conducted title and abstract screening according to our eligibility criteria, with discrepancies resolved through discussion. Studies were included if they met the following criteria: adult patients (≥18 years old) with established status epilepticus defined as seizure activity lasting ≥5 minutes or recurrent seizures without recovery between episodes; first-line benzodiazepine monotherapy as the intervention; intravenous or intramuscular benzodiazepines including lorazepam, diazepam, midazolam, or clonazepam; emergency settings including prehospital emergency medical services or hospital emergency departments; seizure cessation as the primary efficacy endpoint; acute treatment administered within 60 minutes of presentation; and inclusion of at least a comparator arm within prehospital settings. Studies were required to evaluate benzodiazepine treatment initiated in out-of-hospital settings, including prehospital emergency medical services (EMS), community/residential care, or field settings prior to hospital arrival. Studies conducted only in hospital emergency departments or inpatient settings were excluded, as treatment protocols, time-to-administration, drug availability, and monitoring capabilities differ between out-of-hospital and hospital environments.
Studies were also excluded if they evaluated only pediatric populations without separate reporting of adult outcomes, included mixed populations without extractable adult data, evaluated non-emergency or maintenance benzodiazepine therapy, or provided insufficient data for meta-analysis despite contact attempts with the authors.
Data Extraction and Quality Assessment:
Data extraction was performed to obtain the following information: study characteristics (first author, publication year, country, study design, and sample size); patient demographics (age, gender distribution, and status epilepticus type and definition); intervention details (benzodiazepine agent, dose, and route of administration); comparator details; and outcomes including seizure cessation events, time to cessation, recurrence rates, rescue medication requirements, and safety outcomes. Design features relevant to bias assessment were systematically extracted, including blinding status (double-blind vs. open-label vs. unblinded), randomization (RCT vs. observational), outcome adjudication method, and sample size.
The risk of bias in included studies was assessed using the Cochrane Risk of Bias tool version 2.0 (RoB 2) for RCTs and the Risk of Bias in Non-randomized Studies of Interventions (ROBINS-I) tool for observational studies. For RoB 2, domains assessed included bias arising from the randomization process, deviations from intended interventions, missing outcome data, measurement of outcomes, and selection of reported results. For ROBINS-I, domains included confounding, participant selection, classification of interventions, deviations from intended interventions, missing data, measurement of outcomes, and selection of reported results.
Target Trial Specification and T3-Meta Framework:
We utilized the T3-Meta (Target Trial-Centric Meta-Analysis) framework to conduct a bias-adjusted meta-analysis as described by Azzam (2025) [15]. The target trial was specified as follows: adults with status epilepticus (≥5 minutes duration), first-line benzodiazepine administered within 5 minutes of eligibility, and seizure cessation without recurrence within 10 to 20 minutes as the primary outcome, conducted as a double-blind RCT with adjudicated outcomes. The T3-Meta model treats each study's estimate as: θ̂ⱼ = θ* + Xⱼ'β + uⱼ + εⱼ, where θ* represents the target trial effect, Xⱼ represents the design feature deviation vector, β represents bias coefficients estimated from data, uⱼ represents residual heterogeneity following N(0, τ²), and εⱼ represents sampling error following N(0, sⱼ²). Design features modeled included open-label design (vs. double-blind), observational design (vs. RCT), route confounding, and log-transformed sample size. Estimation was performed using restricted maximum likelihood (REML) with Knapp-Hartung adjustment for confidence intervals (CI).
Statistical Analysis:
All statistical analyses were performed using random-effects meta-analysis models based on the DerSimonian-Laird method for traditional pooling and REML for meta-regression. Effect estimates for single-arm proportions were calculated using logit transformation with back-transformation for presentation. For comparative analyses, we calculated pooled odds ratios (ORs) with 95% CIs. Statistical heterogeneity was quantified using the I² statistic, with values exceeding 50% considered significant. Tau-squared values were reported to quantify absolute between-study variance. The Cochran Q test was used to assess the statistical significance of heterogeneity.
Network meta-analysis was performed to compare multiple benzodiazepine agents simultaneously using a frequentist framework with random-effects models. Treatment rankings were calculated using P-scores representing the probability of each treatment being superior to competitors, with Monte Carlo simulation (n = 100,000) generating probability distributions for ranking positions. Network consistency was evaluated using the node-splitting approach comparing direct versus indirect estimates.
Subgroup analyses were conducted by stratifying studies by design (double-blind RCT vs. open-label RCT vs. observational), route of administration, and geographic region. Meta-regression modeling investigated dose-response relationships. Publication bias was assessed through visual inspection of funnel plots and quantified using Egger's regression test and Begg's rank correlation test. Trim-and-fill adjustment was performed when asymmetry was detected. Sensitivity analyses evaluated the impact of individual studies through leave-one-out analysis. Statistical significance was defined as two-tailed P-values less than 0.05. All analyses were conducted using RStudio statistical software (R version 4.4.2) and the Python programming language (version 3.11) for the T3-Meta package (https://github.com/drazzam/t3meta/).
Results
Study Selection and Characteristics:
The literature search identified 418 records from electronic databases, with no additional records identified from registers or other sources (Figure 1). After removing 84 duplicate records and 36 records marked as ineligible by automation tools, 298 unique records underwent title and abstract screening. A total of 257 records were excluded during screening, leaving 41 reports for full-text retrieval. Of these, two reports could not be retrieved, and 39 reports underwent full-text assessment. After full-text review, 25 reports were excluded, resulting in 14 studies included in our study for quantitative synthesis.
Figure 1: PRISMA Flow Diagram.
Edit Remove
Study characteristics, patient demographics, and intervention details are presented in Table 1. The included studies were published between 1983 and 2025, enrolling a total of 2,803 adult patients across 23 study arms. Mean patient age ranged from 29 to 72 year-old across studies, with male representation ranging from 31.1% to 72.1%. Status epilepticus definitions varied, with most studies utilizing a threshold of ≥five minutes of continuous seizure activity.
Table 1: Baseline Demographics and Characteristics of Included Studies.
Study
Country
Design
Number of Individuals
Age, years
Male, n (%)
SE Type
Intervention
Comparator
Primary Outcome
Zitek et al. 2025 [16]
USA
Retrospective Cohort
688 (556 adults)
36.7 (21.2) vs 34.6 (20.8)
275/598 (46.0%) vs 28/90 (31.1%)
CSE
MDZ + Ketamine 100mg
MDZ alone
Cessation prior to hospital
Guterman et al. 2022 [17]
USA
Retrospective Cohort
7,597
46 (17–18)
NR
SE
MDZ IN/IM/IV; 3–5mg
Route comparison
Rescue therapy needed
Halliday et al. 2022 [18]
Australia
Prospective Cohort
53
52.8 (17.2) vs 47.3 (19.9)
NR
OHSE
Pre-hospital BZD
No pre-hospital BZD
Time to cessation
Guterman et al. 2021 [19]
USA
Research Letter
357
46 (18)
NR
SE
BZD per guidelines
Non-concordant
Guideline adherence
Maier et al. 2021 [20]
Germany
Retrospective Cohort
273
72 (55–81)†
NR
SE
BZD mixed
No BZD
Neurological deficit
Guterman et al. 2020 [21]
USA
Retrospective Cohort
1,537
52.9 (19.6) vs 53.4 (18.9)
NR
SE
MDZ ≥5mg
MDZ <5mg
Rescue therapy needed
Sairanen et al. 2019 [22]
Finland
Retrospective Cohort
121
NR
NR
SE (>5 min)
Pre-hospital Tx
In-hospital Tx
Treatment delay
Semmlack et al. 2017 [23]
Switzerland
Retrospective Cohort
150
NR
NR
OHSE
Recognized SE
Missed SE
Functional recovery
Navarro et al. 2016 [24]
France
RCT
136
58 (18) vs 55 (18)
49/68 (72.1%) vs 45/68 (66.2%)
GCSE (>5 min)
CLZ 1mg + LEV 2500mg IV
CLZ 1mg + Placebo IV
Cessation within 15 min
Silbergleit et al. 2012 [25]
USA
RCT
893
43 (22) vs 44 (22)
250/448 (55.8%) vs 238/445 (53.5%)
CSE
MDZ 10mg IM
LZP 4mg IV
Cessation at ED arrival
Nakken et al. 2011 [26]
Norway
Prospective Cohort
22 (80 events)
42.4 (25–68)*
12/22 (54.5%)
Serial SE
MDZ 10–20mg Buccal
DZP 10–30mg Rectal
Cessation <10 min
De Haan et al. 2010 [27]
The Netherlands
Prospective Crossover
21 (124 events)
40.2
13/21 (61.9%)
Refractory SE
MDZ 10mg IN
DZP 10mg Rectal
Cessation <15 min
Fitzgerald et al. 2003 [28]
USA
Retrospective Cohort
6 (57 events)
29 (20–39)†
NR
GCSE (>10 min)
DZP 15–20mg Rectal
LZP 4mg IV
Cessation <10 min
Alldredge et al. 2001 [29]
USA
RCT
205
50.1 (19.1)
129/205 (62.9%)
GCSE
LZP 2mg/DZP 5mg IV
Placebo IV
Cessation at ED arrival
Abbreviations: BZD, benzodiazepine; CLZ, clonazepam; CSE, convulsive status epilepticus; DZP, diazepam; ED, emergency department; GCSE, generalized convulsive status epilepticus; IM, intramuscular; IN, intranasal; IV, intravenous; LEV, levetiracetam; LZP, lorazepam; MDZ, midazolam; NR, not reported; OHSE, out-of-hospital status epilepticus; RCT, randomized controlled trial; SE, status epilepticus; Tx, treatment. Notes: *Mean (range). †Median (range or IQR). Age presented as Mean (SD) unless otherwise noted. For multi-arm studies, values are presented as Intervention vs Comparator. Sample sizes reflect total enrolled; parenthetical values indicate events when studies used event-based analysis.
Edit Remove
Single-Arm Seizure Cessation Rates:
Single-arm seizure cessation rates by benzodiazepine drug are presented in Table 2. Midazolam was evaluated across four study arms (N=1,036), demonstrating a pooled seizure cessation rate of 77.6% (95% CI: 72.1–82.7%) with significant heterogeneity (I²=66.3%, τ²=0.0096, P-value= 0.031). Individual study rates ranged from 73.4% for intramuscular administration in the Rapid Anticonvulsant Medication Prior to Arrival Trial (RAMPART) trial to 82.0% for intranasal administration. Lorazepam was evaluated across three study arms (N=532), with a pooled rate of 71.2% (95% CI: 54.7–85.2%) and significant heterogeneity (I²=85.1%, τ²=0.0748, P-value= 0.001). Diazepam was evaluated across four study arms (N=204), demonstrating a pooled rate of 74.7% (95% CI: 50.7–92.6%) with high heterogeneity (I²=92.4%, τ²=0.2426, P-value<0.001). Clonazepam was evaluated in a single study (N=68), with a cessation rate of 83.8% (95% CI: 73.3–90.7%).
Table 2: Single-Arm Seizure Cessation Rates By Benzodiazepine Drug.
Drug
Study
Route
Dose
Events (n)
Total (N)
Rate (%)
95% CI (%)
Weight (%)
I² (%)
τ²
Q (p-value)
Prediction Interval (%)
Midazolam
Silbergleit et al. 2012
IM
10mg
329
448
73.4
69.2–77.3
35.3
—
—
—
—
Zitek et al. 2025
Mixed
Variable
393
484
81.2
77.5–84.4
35.8
—
—
—
—
De Haan et al. 2010
IN
10mg
50
61
82.0
70.5–89.6
16.2
—
—
—
—
Nakken et al. 2011
Buccal
10–20mg
32
43
74.4
59.8–85.1
12.8
—
—
—
—
Pooled (k=4, N=1036)
—
—
804
1036
77.6
72.1–82.7
100.0
66.3
0.0096
8.91 (p=0.031)
60.5–91.0
Diazepam
De Haan et al. 2010
Rectal
10mg
56
63
88.9
78.8–94.5
25.5
—
—
—
—
Nakken et al. 2011
Rectal
10–30mg
30
37
81.1
65.8–90.5
24.5
—
—
—
—
Fitzgerald et al. 2003
Rectal
15–20mg
30
36
83.3
68.1–92.1
24.4
—
—
—
—
Alldredge et al. 2001
IV
5mg (×2)
29
68
42.6
31.6–54.5
25.6
—
—
—
—
Pooled (k=4, N=204)
—
—
145
204
74.7
50.7–92.6
100.0
92.4
0.2426
39.48 (p<0.001)
2.5–87.8
Lorazepam
Silbergleit et al. 2012
IV
4mg
282
445
63.4
58.8–67.7
40.1
—
—
—
—
Alldredge et al. 2001
IV
2mg (×2)
39
66
59.1
47.0–70.1
34.4
—
—
—
—
Fitzgerald et al. 2003
IV
4mg
20
21
95.2
77.3–99.2
25.5
—
—
—
—
Pooled (k=3, N=532)
—
—
341
532
71.2
54.7–85.2
100.0
85.1
0.0748
13.39 (p=0.001)
9.0–98.2
Clonazepam
Navarro et al. 2016
IV
1mg
57
68
83.8
73.3–90.7
100.0
N/A
N/A
N/A
N/A
Abbreviations: CI, confidence interval; I², heterogeneity statistic; IM, intramuscular; IN, intranasal; IV, intravenous; k, number of studies; N, total sample size; N/A, not applicable (single study); PI, prediction interval; Q, Cochran's Q statistic; τ², between-study variance.
Edit Remove
Component-Based Meta-Analysis:
Component-based meta-analysis evaluating drug, route, and dose effects is presented in Table 3. Regarding drug effects compared to placebo reference, midazolam have demonstrated the highest efficacy (OR=12.76, 95% CI: 7.14–22.81, P-value<0.001), followed by clonazepam (OR=19.35, 95% CI: 8.30–45.11, P-value<0.001), diazepam (OR=7.90, 95% CI: 4.11–15.17, P-value<0.001), and lorazepam (OR=6.49, 95% CI: 3.60–11.69, P-value<0.001). Route analysis with intravenous as reference have demonstrated that rectal administration was associated with significantly higher efficacy (OR=3.34, 95% CI: 2.04–5.44, P-value<0.001), followed by intranasal (OR=2.69, 95% CI: 1.39–5.20, P-value=0.003), mixed routes (OR=2.55, 95% CI: 1.93–3.37, P-value<0.001), and intramuscular (OR=1.63, 95% CI: 1.25–2.13, P-value<0.001). Dose-response analysis have demonstrated a significant inverse relationship, with each 5mg midazolam-equivalent increase associated with reduced efficacy (OR=0.78, 95% CI: 0.63–0.96, P-value=0.019). Model statistics revealed residual heterogeneity of I²=88.7% with R²=5.8% variance explained by component effects.
Table 3: Component-Based Meta-Analysis For Drug, Route, and Dose Effects on Seizure Cessation.
Section
Component
Contributing Arms
Total Patients (N)
Coefficient (β)
SE
Odds Ratio
95% CI
P-value
Drug Effect
Placebo (Reference)
1
71
0.00
—
1.00
—
—
Midazolam
4
1036
2.547
0.296
12.76
7.14–22.81
<0.001
Diazepam
4
204
2.067
0.333
7.90
4.11–15.17
<0.001
Lorazepam
3
532
1.870
0.301
6.49
3.60–11.69
<0.001
Clonazepam
1
68
2.962
0.432
19.35
8.30–45.11
<0.001
Route Effect
IV (Reference)
5
668
0.00
—
1.00
—
—
IM
1
448
0.491
0.135
1.63
1.25–2.13
<0.001
IN
1
61
0.988
0.337
2.69
1.39–5.20
0.003
Buccal
1
43
0.542
0.353
1.72
0.86–3.43
0.125
Rectal
3
136
1.205
0.250
3.34
2.04–5.44
<0.001
Mixed
1
484
0.937
0.142
2.55
1.93–3.37
<0.001
Dose Effect
Per 5mg MDZ-equivalent increase
12
1840
−0.248
0.106
0.78
0.63–0.96
0.019
Model Statistics
Between-study variance (τ²)
—
—
0.2881
—
—
—
—
Residual heterogeneity (I²)
—
—
88.7%
—
—
—
—
Residual Q-statistic
—
—
88.80
df=10
—
—
<0.001
Variance explained (R²)
—
—
5.8%
—
—
—
—
Abbreviations: β, regression coefficient (log-odds scale); CI, confidence interval; df, degrees of freedom; I², heterogeneity statistic; IM, intramuscular; IN, intranasal; IV, intravenous; MDZ-eq, midazolam equivalents; N, total sample size; OR, odds ratio; R², coefficient of determination; SE, standard error; τ², between-study variance.
Edit Remove
Pairwise and Network Comparisons:
Pairwise drug comparisons with direct, indirect, and combined estimates are presented in Table 4. Direct evidence from Alldredge et al. 2001 have demonstrated lorazepam superiority over placebo (OR=5.39, 95% CI: 2.54–11.44) and diazepam superiority over placebo (OR=2.78, 95% CI: 1.32–5.85). The combined estimate for lorazepam versus diazepam was OR=1.94 (95% CI: 1.09–3.46) with no significant inconsistency (P-value= 1.000). Midazolam versus lorazepam from Silbergleit et al. 2012 have demonstrated significant midazolam superiority (OR=1.60, 95% CI: 1.20–2.12, P-value= 0.001). Midazolam versus diazepam showed significant inconsistency between direct and indirect estimates (P-value= 0.003), with direct evidence suggesting OR=0.62 and indirect evidence via lorazepam suggesting OR=3.10, resulting in a combined OR=1.38 (95% CI: 0.82–2.33). Indirect comparison demonstrated any benzodiazepine superior to placebo (OR=3.85, 95% CI: 1.98–7.46).
Table 4: Pairwise Drug Comparisons with Direct, Indirect, and Combined Estimates.
Comparison
Direct Evidence
Direct OR (95% CI)
Indirect Pathway
Indirect OR (95% CI)
Combined OR (95% CI)
Consistency P-value
Lorazepam vs Placebo
Alldredge et al. 2001
5.39 (2.54–11.44)
—
—
5.39 (2.54–11.44)
—
Diazepam vs Placebo
Alldredge et al. 2001
2.78 (1.32–5.85)
—
—
2.78 (1.32–5.85)
—
Lorazepam vs Diazepam
Alldredge et al. 2001
1.94 (0.98–3.86)
Via Placebo
1.94 (0.67–5.60)
1.94 (1.09–3.46)
1.000
Midazolam vs Lorazepam
Silbergleit et al. 2012
1.60 (1.20–2.12)
—
—
1.60 (1.20–2.12)
—
Midazolam vs Diazepam
De Haan et al. 2010; Nakken et al. 2011
0.62 (0.30–1.29)
Via Lorazepam
3.10 (1.48–6.53)
1.38 (0.82–2.33)
0.003
Midazolam vs Placebo
—
—
Via Lorazepam
8.62 (3.86–19.26)
8.62 (3.86–19.26)
—
Clonazepam vs Placebo
—
—
Single-arm comparison
19.35 (8.18–45.76)
19.35 (8.18–45.76)
—
Any BZD vs Placebo
Alldredge et al. 2001
3.85 (1.98–7.46)
—
—
3.85 (1.98–7.46)
—
Abbreviations: BZD, benzodiazepine; CI, confidence interval; OR, odds ratio.
Edit Remove
Safety Outcomes:
Pooled safety outcomes by drug and route are presented in Table 5. Respiratory complications requiring intubation occurred in 6.1% of midazolam-treated patients (95% CI: 0.0–24.4%, I²=98.7%), 9.7% of lorazepam-treated patients (95% CI: 3.7–18.1%, I²=70.8%), 4.6% of diazepam-treated patients (95% CI: 0.1–20.0%, I²=83.6%), and 2.9% of clonazepam-treated patients (95% CI: 0.8–10.1%). Hypotension occurred in 3.5% of all active benzodiazepine-treated patients (95% CI: 2.2–5.2%, I²=30.1%). Excessive sedation was reported in 44.0% of midazolam-treated patients (95% CI: 6.3–86.6%, I²=95.7%) and 52.5% of diazepam-treated patients (95% CI: 42.7–62.2%, I²=0.0%). Mortality occurred in 6.4% of benzodiazepine-treated patients (95% CI: 3.9–9.4%, I²=0.0%). Comparative safety analysis demonstrated that active benzodiazepine treatment was protective compared to placebo for both respiratory complications (OR=0.40, 95% CI: 0.18–0.88, P-value= 0.023) and mortality (OR=0.35, 95% CI: 0.13–0.91, P-value= 0.031).
Table 5: Pooled Safety Outcomes by Drug and Route.
Section
Study
Drug
Route
Events (n)
Total (N)
Rate (%)
95% CI (%)
Weight (%)
I² (%)
Definition
Respiratory Complications/Intubation
Alldredge et al. 2001
Lorazepam
IV
7
66
10.6
5.2–20.3
—
—
Respiratory intervention
Alldredge et al. 2001
Diazepam
IV
7
68
10.3
5.1–19.8
—
—
Respiratory intervention
Silbergleit et al. 2012
Midazolam
IM
63
448
14.1
11.1–17.6
—
—
Intubation
Silbergleit et al. 2012
Lorazepam
IV
64
445
14.4
11.4–17.9
—
—
Intubation
Navarro et al. 2016
Clonazepam
IV
2
68
2.9
0.8–10.1
—
—
Respiratory disorder SAE
Navarro et al. 2016
Clonazepam+LEV
IV
3
68
4.4
1.5–12.2
—
—
Respiratory disorder SAE
Zitek et al. 2025
Midazolam
Mixed
7
598
1.2
0.6–2.4
—
—
Intubation
Fitzgerald et al. 2003
Diazepam
Rectal
0
36
0.0
0.0–9.6
—
—
Respiratory difficulty
Fitzgerald et al. 2003
Lorazepam
IV
0
21
0.0
0.0–15.5
—
—
Respiratory difficulty
Pooled
Midazolam
—
70
1046
6.1
0.0–24.4
100.0
98.7
—
Pooled
Lorazepam
—
71
532
9.7
3.7–18.1
100.0
70.8
—
Pooled
Diazepam
—
7
104
4.6
0.1–20.0
100.0
83.6
—
Pooled
Clonazepam
—
2
68
2.9
0.8–10.1
100.0
N/A
—
Hypotension
Silbergleit et al. 2012
Midazolam
IM
12
448
2.7
1.5–4.6
—
—
Not specified
Silbergleit et al. 2012
Lorazepam
IV
13
445
2.9
1.7–4.9
—
—
Not specified
Navarro et al. 2016
Clonazepam
IV
3
68
4.4
1.5–12.2
—
—
Circulatory failure
Navarro et al. 2016
Clonazepam+LEV
IV
5
64
7.8
3.4–17.0
—
—
Circulatory failure
Pooled
All Active BZD
—
33
1025
3.5
2.2–5.2
100.0
30.1
—
Excessive Sedation
De Haan et al. 2010
Midazolam
IN
40
59
67.8
55.1–78.3
—
—
Reported sedation
De Haan et al. 2010
Diazepam
Rectal
34
62
54.8
42.5–66.6
—
—
Reported sedation
Nakken et al. 2011
Midazolam
Buccal
9
43
20.9
11.4–35.2
—
—
Sedation >2h
Nakken et al. 2011
Diazepam
Rectal
18
37
48.6
33.4–64.1
—
—
Sedation >2h
Pooled
Midazolam
—
49
102
44.0
6.3–86.6
100.0
95.7
—
Pooled
Diazepam
—
52
99
52.5
42.7–62.2
100.0
0.0
—
Mortality
Alldredge et al. 2001
Lorazepam
IV
5
65
7.7
3.3–16.8
—
—
In-hospital
Alldredge et al. 2001
Diazepam
IV
3
67
4.5
1.5–12.4
—
—
In-hospital
Alldredge et al. 2001
Placebo
IV
11
70
15.7
9.0–26.0
—
—
In-hospital
Navarro et al. 2016
Clonazepam
IV
4
65
6.2
2.4–14.8
—
—
In-hospital
Navarro et al. 2016
Clonazepam+LEV
IV
3
66
4.5
1.6–12.5
—
—
In-hospital
Halliday et al. 2022
BZD
Mixed
2
35
5.7
1.6–18.6
—
—
In-hospital
Halliday et al. 2022
No BZD
—
0
18
0.0
0.0–17.6
—
—
In-hospital
Pooled
All Active BZD
—
17
298
6.4
3.9–9.4
100.0
0.0
—
Comparative Safety (Active BZD vs Placebo)
Alldredge et al. 2001
Respiratory
—
—
—
OR = 0.40
0.18–0.88
—
—
BZD protective (p=0.023)
Alldredge et al. 2001
Mortality
—
—
—
OR = 0.35
0.13–0.91
—
—
BZD protective (p=0.031)
Abbreviations: BZD, benzodiazepine; CI, confidence interval; I², heterogeneity statistic; IM, intramuscular; IN, intranasal; IV, intravenous; KET, ketamine; LEV, levetiracetam; N, total sample size; n, number of events; N/A, not applicable (single study); NNT, number needed to treat; OR, odds ratio; SAE, serious adverse event.
Edit Remove
Time-to-Cessation and Secondary Outcomes:
Time-to-cessation and secondary outcomes are presented in Table 6. Median time to seizure cessation was 3.3 minutes (interquartile range [IQR]: 1.7–7.5) for intramuscular midazolam compared to 1.6 minutes (IQR: 0.9–4.7) for intravenous lorazepam in the RAMPART trial. Mean time to cessation for intranasal midazolam was 4.6±3.4 minutes compared to 4.3±3.4 minutes for rectal diazepam in De Haan et al. 2010. Seizure recurrence rates ranged from 15.4% to 31.6% across studies. Rescue medication was required in 26.6% of intramuscular midazolam-treated patients compared to 36.6% of intravenous lorazepam-treated patients in the RAMPART trial (P-value<0.001). Intensive care unit admission (after admission to the hospital if needed) occurred in 36.8% of midazolam-treated patients compared to 35.1% of lorazepam-treated patients, with median hospital length of stay (if admitted to hospital) of two days in both groups.
Table 6: Time-to-Cessation and Secondary Outcomes.
Section
Study
Drug
Route
N
Outcome
Value
95% CI
Notes
Time to Seizure Cessation (minutes)
Silbergleit et al. 2012
Midazolam
IM
329
Median (IQR)
3.3 (1.7–7.5)
—
Time from study drug
Silbergleit et al. 2012
Lorazepam
IV
282
Median (IQR)
1.6 (0.9–4.7)
—
Time from study drug
De Haan et al. 2010
Midazolam
IN
50
Mean ± SD
4.6 ± 3.4
3.7–5.5
Time to cessation
De Haan et al. 2010
Diazepam
Rectal
56
Mean ± SD
4.3 ± 3.4
3.4–5.2
Time to cessation
Nakken et al. 2011
Midazolam
Buccal
32
Mean (range)
5.0 (1–15)
—
Among responders
Nakken et al. 2011
Diazepam
Rectal
30
Mean (range)
5.5 (1–15)
—
Among responders
Alldredge et al. 2001
Lorazepam
IV
39
Mean ± SD
2.0 ± 1.5
1.5–2.5
Responders only
Alldredge et al. 2001
Diazepam
IV
29
Mean ± SD
2.5 ± 2.0
1.8–3.2
Responders only
Seizure Recurrence
Alldredge et al. 2001
Lorazepam
IV
39
Events/N (%)
6/39 (15.4%)
7.2–29.7%
Recurrence <60 min
Alldredge et al. 2001
Diazepam
IV
29
Events/N (%)
8/29 (27.6%)
14.7–45.7%
Recurrence <60 min
Alldredge et al. 2001
Placebo
IV
15
Events/N (%)
3/15 (20.0%)
7.0–45.2%
Recurrence <60 min
Silbergleit et al. 2012
Midazolam
IM
329
Events/N (%)
57/329 (17.3%)
13.6–21.8%
Additional Tx required
Silbergleit et al. 2012
Lorazepam
IV
282
Events/N (%)
52/282 (18.4%)
14.3–23.4%
Additional Tx required
Navarro et al. 2016
Clonazepam
IV
57
Events/N (%)
18/57 (31.6%)
21.0–44.5%
SE recurrence at 1h
Navarro et al. 2016
Clonazepam+LEV
IV
54
Events/N (%)
14/54 (25.9%)
16.1–38.9%
SE recurrence at 1h
Rescue Medication Required
Silbergleit et al. 2012
Midazolam
IM
448
Events/N (%)
119/448 (26.6%)
22.7–30.8%
Second-line AED
Silbergleit et al. 2012
Lorazepam
IV
445
Events/N (%)
163/445 (36.6%)
32.3–41.2%
Second-line AED
De Haan et al. 2010
Midazolam
IN
61
Events/N (%)
11/61 (18.0%)
10.4–29.5%
Additional Tx
De Haan et al. 2010
Diazepam
Rectal
63
Events/N (%)
7/63 (11.1%)
5.5–21.2%
Additional Tx
Nakken et al. 2011
Midazolam
Buccal
43
Events/N (%)
11/43 (25.6%)
14.9–40.2%
Additional Tx
Nakken et al. 2011
Diazepam
Rectal
37
Events/N (%)
7/37 (18.9%)
9.5–34.2%
Additional Tx
Alldredge et al. 2001
Lorazepam
IV
66
Events/N (%)
27/66 (40.9%)
29.9–53.0%
Second BZD dose
Alldredge et al. 2001
Diazepam
IV
68
Events/N (%)
39/68 (57.4%)
45.5–68.4%
Second BZD dose
Alldredge et al. 2001
Placebo
IV
71
Events/N (%)
56/71 (78.9%)
68.0–86.8%
Second BZD dose
Hospital Outcomes (ICU Admission)
Silbergleit et al. 2012
Midazolam
IM
448
Events/N (%)
165/448 (36.8%)
32.5–41.4%
—
Silbergleit et al. 2012
Lorazepam
IV
445
Events/N (%)
156/445 (35.1%)
30.8–39.6%
—
Halliday et al. 2022
BZD
Mixed
35
Events/N (%)
25/35 (71.4%)
54.9–83.7%
—
Halliday et al. 2022
No BZD
—
18
Events/N (%)
11/18 (61.1%)
38.6–79.7%
—
Hospital Outcomes (Length of Stay [days])
Silbergleit et al. 2012
Midazolam
IM
448
Median (IQR)
2 (1–5)
—
Days
Silbergleit et al. 2012
Lorazepam
IV
445
Median (IQR)
2 (1–5)
—
Days
Comparative Time Analysis
Silbergleit et al. 2012
MDZ vs LZP
IM vs IV
611
Median difference
+1.7 min
—
MDZ slower post-drug
De Haan et al. 2010
MDZ vs DZP
IN vs Rec
106
Mean difference
+0.3 min
−1.0 to +1.6
NS (p=0.65)
Nakken et al. 2011
MDZ vs DZP
Buc vs Rec
62
Mean difference
−0.5 min
−2.2 to +1.2
NS (p=0.58)
Alldredge et al. 2001
LZP vs DZP
IV vs IV
68
Mean difference
−0.5 min
−1.4 to +0.4
NS (p=0.26)
Abbreviations: AED, antiepileptic drug; BZD, benzodiazepine; Buc, buccal; CI, confidence interval; DZP, diazepam; ICU, intensive care unit; IM, intramuscular; IN, intranasal; IQR, interquartile range; IV, intravenous; LEV, levetiracetam; LZP, lorazepam; MDZ, midazolam; N, total sample size; NS, not significant; Rec, rectal; SD, standard deviation; SE, status epilepticus; Tx, treatment.
Edit Remove
Target Trial Effect Analysis:
Bias-adjusted target trial effect analysis is presented in Table 7 and Figure 2. Traditional random-effects pooling across 23 study arms have demonstrated a seizure cessation rate of 72.3% (95% CI: 68.1–76.1%, τ²=0.111, I²=73.3%). The target trial effect (θ*), representing the bias-adjusted estimate for an ideal double-blind RCT, was 64.6% (95% CI: 57.6–71.1%, τ²=0.065, I²=28.4%). Additional adjustments for route confounding and sample size resulted in final estimates of 64.7% and 65.8%, respectively. Bias coefficient analysis demonstrated significant open-label design bias (β=0.757, 95% CI: 0.01–1.51, P-value=0.048), indicating that open-label studies overestimate efficacy by OR×2.13 compared to double-blind RCTs. Observational design (β=−0.104, P-value=0.812), route confounding (β=−0.191, P-value=0.549), and sample size (β=−0.061, P-value=0.646) were not statistically significant.
Figure 2: Bias-Adjusted Forest Plot For Benzodiazepine Efficacy.
Edit Remove
Heterogeneity decomposition have demonstrated that design features explained 51.0% of total I², reducing residual heterogeneity to 22.3%. The R² of 69.6% indicated that design features accounted for nearly 70% of between-study variance. Drug-specific target trial effects were 73.4% for midazolam (bias: +7.8% points), 64.6% for lorazepam (bias: +2.0% points), and 59.3% for diazepam (bias: +11.4% points). Validation analysis have demonstrated that double-blind RCTs only resulted in a pooled rate of 65.0% (95% CI: 55.8–73.1%), closely approximating the target trial effect θ* (difference: +0.3%). The 95% prediction interval for a new ideal double-blind RCT was 47.9–80.1%. Figure 3 demonstrates the sequential decomposition of bias through the waterfall plot, showing the progression from traditional pooled estimate (68.8%) through open-label bias adjustment (−7.7% points) and observational bias offset (+3.5% points) to the final target trial effect (64.7%).
Table 7: Bias-Adjusted Target Trial Effect Analysis for Benzodiazepine Efficacy in Status Epilepticus.
Analysis / Parameter
k
Estimate
95% CI
τ²
I²
Interpretation
PRIMARY ANALYSIS: POOLED EFFICACY ESTIMATES
Traditional random-effects
23
72.3%
68.1–76.1
0.111
73.3%
Unadjusted pooled seizure cessation rate
Target trial effect (θ)*
23
64.6%
57.6–71.1
0.065
28.4%
Bias-adjusted estimate for ideal double-blind RCT
+ Route confounding adjustment
23
64.7%
57.5–71.2
0.067
29.0%
Additional adjustment for non-IV route comparisons
+ Sample size adjustment
23
65.8%
56.8–73.7
0.091
22.3%
Full model with small-study effect adjustment
BIAS COEFFICIENTS (β): QUANTIFIED METHODOLOGICAL BIASES
Open-label design (vs double-blind)
—
β = 0.757*
0.01–1.51
—
—
Overestimates efficacy by OR × 2.13 (P = 0.048)
Observational design (vs RCT)
—
β = −0.104
−1.01–0.81
—
—
Non-significant (OR × 0.90, P = 0.812)
Route confounding present
—
β = −0.191
−0.85–0.47
—
—
Non-significant (OR × 0.83, P = 0.549)
Log sample size (centered)
—
β = −0.061
−0.33–0.21
—
—
OR × 0.94 per log-unit N (P = 0.646)
HETEROGENEITY DECOMPOSITION:
Total I² (unadjusted)
—
73.3%
—
—
—
Overall heterogeneity across all included studies
Explained I² (by design features)
—
51.0%
—
—
—
Attributable to methodological differences
Residual I² (true clinical variation)
—
22.3%
—
—
—
Remaining heterogeneity after bias adjustment
R² (variance explained)
—
69.6%
—
—
—
Proportion of τ² explained by design features
DRUG-SPECIFIC TARGET TRIAL EFFECTS:
Midazolam
6
73.4%
66.2–79.5
0.171
0.0%
Traditional: 81.2%; Bias: +7.8 pp
Lorazepam
5
64.6%
42.8–81.7
0.570
0.0%
Traditional: 66.6%; Bias: +2.0 pp
Diazepam
9
59.3%
40.8–75.5
0.218
47.7%
Traditional: 70.7%; Bias: +11.4 pp
Clonazepam
1
—
—
—
—
Insufficient studies for bias-adjusted analysis
VALIDATION: COMPARISON WITH STUDY DESIGN SUBSETS
Double-blind RCTs only
8
65.0%
55.8–73.1
0.153
82.1%
Closely approximates θ* (difference: +0.3%)
All RCTs (excluding observational)
20
72.5%
67.0–77.3
0.108
73.4%
Difference from θ*: +7.8 pp
Observational studies only
3
72.8%
56.1–84.8
0.156
80.8%
Difference from θ*: +8.1 pp
95% Prediction interval for new ideal RCT
—
—
47.9–80.1
—
—
Expected range for a future ideal double-blind RCT
Abbreviations: β = bias coefficient (log-odds ratio scale); CI = confidence interval; I² = heterogeneity statistic; k = number of study arms; OR = odds ratio; pp = percentage points; RCT = randomized controlled trial; θ* = target trial effect (bias-adjusted estimate); τ² = between-study variance. Notes: *P < 0.05. Positive β values indicate the design feature leads to overestimation of efficacy compared to an ideal double-blind RCT. Target trial specification: Adults with status epilepticus (≥5 min), first-line benzodiazepine within 5 min, seizure cessation without recurrence within 10–20 min, double-blind RCT with adjudicated outcomes. Model: θ̂ⱼ = θ* + Xⱼ'β + uⱼ + εⱼ, where θ* = target trial effect, Xⱼ = design feature deviations, β = bias coefficients. Estimation via REML with Knapp-Hartung adjustment.
Edit Remove
Risk of Bias Assessment:
Risk of bias assessment is presented in Supplementary Table 1. Among double-blind RCTs assessed using RoB 2, Silbergleit et al. 2012 and Alldredge et al. 2001 were judged as low overall risk of bias, while Treiman et al. 1998 and Leppik et al. 1983 showed some concerns mainly due to outcome measurement domains. Open-label RCTs showed some concerns due to deviations from intended interventions and outcome measurement, as expected given the lack of blinding. Observational studies assessed using ROBINS-I demonstrated serious overall risk of bias, mainly due to confounding and selection of participants domains. The presence of this bias gradient supported our hypothesis that design features impact effect estimates.
Figure 3: Bias-Adjustment Waterfall Plot.
Edit Remove
Dose-Response Meta-Analysis by Drug:
Dose-response meta-analysis by drug is presented in Supplementary Table 2. Equipotent dose conversions were utilized (midazolam 5mg = lorazepam 2mg = diazepam 10mg = clonazepam 0.5mg) to standardize doses across studies. Dose categories were defined as low (≤5mg midazolam-equivalent, k=6, N=304, 65.5%), medium (5–10mg midazolam-equivalent, k=8, N=1,271, 71.4%), and high (>10mg midazolam-equivalent, k=1, N=104, 63.5%). Drug-specific dose-response analyses have demonstrated no significant linear relationships, with midazolam showing OR=0.68 per 5mg increase (95% CI: 0.11–4.23, P=0.701, R²=4.1%) and lorazepam showing OR=0.89 per 5mg increase (95% CI: 0.30–2.65, P=0.854, R²=2.1%). These findings suggested that within the therapeutic dose ranges evaluated, higher doses did not significantly improve seizure cessation rates.
Subgroup and Sensitivity Analyses:
Subgroup analyses are presented in Supplementary Table 3. Study design stratification have demonstrated significantly higher efficacy in open-label RCTs (79.0%, 95% CI: 74.8–82.6%, I²=0.0%) compared to double-blind RCTs (65.0%, 95% CI: 57.5–71.9%, I²=82.3%) and observational studies (72.9%, 95% CI: 65.7–79.0%, I²=81.1%), with significant interaction (P-value= 0.011). Route stratification have demonstrated highest efficacy for intranasal administration (87.9%, 95% CI: 81.2–92.4%) compared to intravenous (68.3%, 95% CI: 61.4–74.5%), with significant interaction (P-value= 0.024).
Sensitivity analyses are presented in Supplementary Table 4. Leave-one-out analysis have demonstrated significant findings, with pooled estimates ranging from 71.1% (excluding De Haan et al. 2010) to 73.9% (excluding Alldredge et al. 2001), indicating no single study exerted undue influence on pooled results. Restriction to low risk of bias studies (k=6) resulted in a pooled rate of 68.1% (95% CI: 60.2–75.1%), while restriction to RCTs only (k=20) resulted in 72.5% (95% CI: 67.0–77.3%).
Network Consistency and Comparative Effectiveness:
Network consistency and indirect comparison validation are presented in Supplementary Table 5. The network structure included four nodes (midazolam, lorazepam, diazepam, placebo) connected by five direct edges forming two closed loops. Node-splitting analysis have demonstrated no statistically significant inconsistency in either loop: Loop 1 (lorazepam-diazepam-placebo) showed direct/indirect ratio of 0.76 (P-value= 0.630), and Loop 2 (midazolam-lorazepam-diazepam) showed ratio of 0.58 (P-value= 0.167). These findings supported the validity of indirect comparisons for estimating midazolam versus diazepam effects via the lorazepam pathway.
Advanced network meta-analysis results including probability rankings are presented in Supplementary Table 6 and Figure 4. Monte Carlo simulation (n=100,000) have demonstrated that midazolam achieved the highest P-Score of 96.2% with expected rank of 1.08, indicating 92.4% probability of ranking first, 7.5% probability of ranking second, and 0.1% probability of ranking third. Lorazepam achieved a P-Score of 50.9% with expected rank of 1.98, indicating 7.6% probability of ranking first, 86.7% probability of ranking second, and 5.7% probability of ranking third. Diazepam achieved a P-Score of 2.9% with expected rank of 2.94, indicating near-zero probability of ranking first or second and 94.2% probability of ranking third.
Figure 4: Comparative Effectiveness Rankogram.
Edit Remove
Route-Stratified Target Trial Analysis:
Route-stratified target trial effect analysis is presented in Supplementary Table 7. For intravenous route specifically (k=8, N=918), traditional random-effects pooling resulted in 66.1% (95% CI: 56.0–75.0%, I²=79.4%), while target trial analysis resulted in θ*=63.0% (95% CI: 50.3–74.2%, I²=39.5%), representing a bias magnitude of +3.1% points. For intramuscular route (k=2, N=509), traditional pooling resulted in 81.9% while target trial analysis resulted in θ*=66.2%, representing significant bias adjustment. The open-label bias coefficient for intravenous studies was β=1.075 (OR×2.93), however this did not reach statistical significance (P-value= 0.101) likely due to limited study numbers.
Publication Bias Assessment:
Publication bias assessment is presented in Supplementary Table 8 and Figure 5. Egger's regression test has demonstrated borderline asymmetry (t=1.99, intercept=1.24, P-value= 0.060), while Begg's rank correlation showed significant asymmetry (Kendall's τ=0.371, P-value= 0.014). Peters' test for binary outcomes was not significant (P-value= 0.210). Funnel plot metrics have demonstrated asymmetry ratio of 2.29 (16 studies above pooled estimate versus seven below), with small studies showing higher efficacy (78.7%) compared to large studies (70.1%), representing a difference of +8.5% points. Trim-and-fill imputed one missing study, adjusting the pooled estimate from 68.1% to 67.8% (Δ=+0.3% points), indicating minimal impact of possible publication bias on overall conclusions.
Figure 5: Funnel Plot with Trim-and-Fill Adjustment For Publication Bias Assessment.
Edit Remove
Discussion
Our study represents the first target trial emulation meta-analysis investigating benzodiazepine efficacy for status epilepticus in adults in out-of-hospital settings, utilizing a novel framework that quantifies and adjusts for design-induced bias across heterogeneous study designs. The findings have demonstrated that traditional meta-analytic methods overestimate seizure cessation rates by around 3.9% points compared to bias-adjusted estimates, with open-label study design identified as the primary source of this inflation. Network meta-analysis has confirmed intramuscular midazolam as the most effective first-line agent, with a P-Score of 96.2%, followed by lorazepam and diazepam.
The bias-adjusted pooled cessation rate of 64.6% observed in our analysis aligns closely with estimates from double-blind RCTs. The landmark RAMPART trial reported seizure cessation rates of 73.4% for intramuscular midazolam and 63.4% for intravenous lorazepam [30,25], while the Prehospital Treatment of Status Epilepticus (PHTSE) study documented rates of 59.1% for lorazepam and 42.6% for diazepam [31,29]. When our analysis was restricted to double-blind RCTs only, the pooled estimate of 65.0% showed minimal deviation (+0.3% points) from the target trial-adjusted θ* of 64.6%, providing external validation for our methodological approach.
The identification of open-label bias as a significant predictor of inflated efficacy estimates carries important implications for evidence synthesis in emergency neurology. Studies have demonstrated that lack of blinding can introduce performance and detection bias, especially for outcomes requiring subjective assessment such as seizure cessation timing [32,33]. Our findings indicated that open-label RCTs reported cessation rates of 79.0% compared to 65.0% in double-blind designs, a difference of 14% points that cannot be attributed to patient or intervention characteristics alone. This observation is consistent with the target trial emulation framework, which focuses on alignment of study design features with an ideal randomized trial specification to minimize bias.
The superiority of intramuscular midazolam over intravenous lorazepam observed in our network meta-analysis is consistent with findings from the RAMPART trial, which demonstrated that the practical advantages of intramuscular administration translate into improved clinical outcomes in prehospital settings. The RAMPART investigators attributed this advantage partly to the faster time to treatment initiation (median 1.2 minutes for intramuscular midazolam vs 4.8 minutes for intravenous lorazepam), which may offset the slightly longer time from drug administration to seizure cessation [30,25]. Our component-based analysis further supported this interpretation, showing that intramuscular route was associated with OR=1.63 (95% CI 1.25–2.13, P-value<0.001) compared to intravenous administration.
The heterogeneity decomposition analysis revealed that design features explained 69.6% of between-study variance, leaving residual I² of only 22.3% after adjustment. This finding suggested that much of the observed heterogeneity in benzodiazepine efficacy literature reflects methodological rather than true clinical variation. Previous studies investigating second-line agents for benzodiazepine-resistant status epilepticus have reported similar challenges with heterogeneity [34-36]; Yasiry and Shorvon (2014) documented that 77.7% of studies in their analysis were observational and retrospective, contributing to significant methodological heterogeneity [37]. By modeling design features as bias covariates, our approach provides a framework for separating methodological noise from clinically meaningful variation.
The safety profile observed in our analysis was consistent with established evidence from earlier discussed studies. Respiratory depression rates ranged from 2.9% for clonazepam to 9.7% for lorazepam, with the pooled benzodiazepine-versus-placebo comparison showing a protective effect (OR=0.40, 95% CI 0.18–0.88, P-value= 0.023). This finding aligns with the PHTSE study, which documented lower cardiorespiratory complication rates in benzodiazepine-treated patients (10.4%) compared to placebo recipients (22.5%), focusing on that the risks of untreated status epilepticus exceed those of benzodiazepine administration.
Several limitations should be acknowledged. First, the target trial emulation framework relies on the assumption that design features can be modeled as additive bias terms, which may not capture complex interactions between methodological characteristics. Second, the small number of double-blind RCTs (k=4) limited our ability to precisely estimate the target trial effect, as reflected in the wider 95% CIs for θ*. Third, our analysis included studies spanning four decades (1983–2025), during which definitions of status epilepticus, treatment protocols, and outcome assessment methods have evolved. Fourth, publication bias assessment showed borderline significance (Egger's P-value= 0.060), suggesting possible selective reporting of positive results. Fifth, the network meta-analysis detected inconsistency in the midazolam-diazepam comparison, indicating that indirect and direct evidence may not be fully compatible for this drug pair.
Despite these limitations, our study offers several contributions to the field. The T3-Meta framework provides a principled approach for integrating heterogeneous evidence while explicitly modeling design-induced bias, which may be applicable to other therapeutic areas where RCT evidence is limited. The quantification of open-label bias magnitude (OR=2.13) can inform sample size calculations for future trials and sensitivity analyses in systematic reviews and meta-analyses. In addition to that, our route-stratified analysis suggested that the practical advantages of intramuscular administration may outweigh theoretical pharmacokinetic advantages of intravenous delivery in prehospital settings.
Conclusions
This target trial emulation meta-analysis has demonstrated that traditional pooled estimates of benzodiazepine efficacy for status epilepticus are inflated by around 3.9% points due to open-label study design bias. The bias-adjusted seizure cessation rate of 64.6% represents a more accurate estimate of true treatment effectiveness under ideal double-blind RCT conditions. Open-label design was identified as the only significant predictor of bias inflation (OR=2.13, P-value= 0.048), explaining 69.6% of observed between-study heterogeneity. Network meta-analysis confirmed intramuscular midazolam as the most effective first-line agent (P-Score 96.2%), with significant superiority over lorazepam (OR=1.60, P-value= 0.001) and diazepam (OR=2.21, P-value= 0.002). These findings support current guideline recommendations for intramuscular midazolam in prehospital settings. Future studies should prioritize double-blind trial designs and standardized outcome definitions to minimize methodological heterogeneity in this therapeutic area.
Treiman DM. The role of benzodiazepines in the management of status epilepticus. Neurology. 1990
Treiman DM. Pharmacokinetics and clinical use of benzodiazepines in the management of status epilepticus. Epilepsia. 1989. doi:10.1111/j.1528-1157.1989.tb05824.x
Glauser T, Shinnar S, Gloss D, et al, Treiman DM. Evidence-Based Guideline: Treatment of Convulsive Status Epilepticus in Children and Adults: Report of the Guideline Committee of the American Epilepsy Society. Epilepsy Currents. 2016. doi:10.5698/1535-7597-16.1.48
Jain P, Aneja S, Cunningham J, et al, Sharma S. Treatment of benzodiazepine-resistant status epilepticus: Systematic review and network meta-analyses. Seizure. 2022. doi:10.1016/j.seizure.2022.09.017
Sathe AG, Tillman H, Coles LD, et al, Cloyd JC. Underdosing of Benzodiazepines in Patients With Status Epilepticus Enrolled in Established Status Epilepticus Treatment Trial. Academic Emergency Medicine. 2019. doi:10.1111/acem.13811
Pannucci CJ, Wilkins EG. Identifying and avoiding bias in research. Plastic and Reconstructive Surgery. 2010. doi:10.1097/PRS.0b013e3181de24bc
Berkman ND, Santaguida PL, Viswanathan M, et al, Morton SC. The Empirical Evidence of Bias in Trials Measuring Treatment Differences. AHRQ Methods for Effective Health Care. Agency for Healthcare Research and Quality (US). 2014
Dekkers OM. Meta-analysis: Key features, potentials and misunderstandings. Research and Practice in Thrombosis and Haemostasis. 2018. doi:10.1002/rth2.12153
Fu EL. Target Trial Emulation to Improve Causal Inference from Observational Data: What, Why, and How? Journal of the American Society of Nephrology: JASN. 2023. doi:10.1681/ASN.0000000000000152
Hernán MA, Wang W, Leaf DE. Target Trial Emulation: A Framework for Causal Inference From Observational Data. JAMA. 2022. doi:10.1001/jama.2022.21383
Wong KHF, Hinchliffe RJ. Target trial emulation: harnessing real-world data to evaluate surgery and perioperative care interventions. British Journal of Surgery. 2025. doi:10.1093/bjs/znaf182
Hernán MA, Dahabreh IJ, Dickerman BA, et al, Swanson SA. The Target Trial Framework for Causal Inference From Observational Data: Why and When Is It Helpful? Annals of Internal Medicine. 2025. doi:10.7326/ANNALS-24-01871
Cashin AG, Hansford HJ, Hernán MA, et al, McAuley JH. Transparent Reporting of Observational Studies Emulating a Target Trial-The TARGET Statement. JAMA. 2025. doi:10.1001/jama.2025.13350
Page MJ, McKenzie JE, Bossuyt PM, et al, Moher D. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. Revista Espanola de Cardiologia (English Edition). 2021. doi:10.1016/j.rec.2021.07.010
Azzam AY. T3-Meta: Target Trial-Centric Meta-Analysis Framework, version 1.0.0. GitHub. 2025
Zitek T, Scheppke KA, Antevy P, et al, Farcy DA. Midazolam and Ketamine for Convulsive Status Epilepticus in the Out-of-Hospital Setting. Annals of Emergency Medicine. 2024. doi:10.1016/j.annemergmed.2024.11.002
Guterman EL, Sporer KA, Newman TB, et al, Burke JF. Real-World Midazolam Use and Outcomes With Out-of-Hospital Treatment of Status Epilepticus in the United States. Annals of Emergency Medicine. 2022. doi:10.1016/j.annemergmed.2022.05.024
Halliday AJ, Santamaria J, D'Souza WJ. Pre-hospital benzodiazepines associated with improved outcomes in out-of-hospital status epilepticus: A 10-year retrospective cohort study. Epilepsy Research. 2021. doi:10.1016/j.eplepsyres.2021.106846
Guterman EL, Burke JF, Sporer KA. Prehospital Treatment of Status Epilepticus in the United States. JAMA. 2021. doi:10.1001/jama.2021.15964
Maier S, Godau J, Bösel J, et al, Rösche J. Recognition and treatment of status epilepticus in the prehospital setting. Seizure. 2021. doi:10.1016/j.seizure.2020.12.014
Guterman EL, Sanford JK, Betjemann JP, et al, Sporer KA. Prehospital midazolam use and outcomes among patients with out-of-hospital status epilepticus. Neurology. 2020. doi:10.1212/WNL.0000000000010913
Sairanen JJ, Kantanen AM, Hyppölä HT, et al, Kälviäinen RK. Treatment delay in status epilepticus - more effective prehospital symptom recognition warranted. Scandinavian Journal of Trauma, Resuscitation and Emergency Medicine. 2019. doi:10.1186/s13049-019-0605-7
Semmlack S, Yeginsoy D, Spiegel R, et al, Sutter R. Emergency response to out-of-hospital status epilepticus: A 10-year observational cohort study. Neurology. 2017. doi:10.1212/WNL.0000000000004147
Navarro V, Dagron C, Elie C, et al, Carli P. Prehospital treatment with levetiracetam plus clonazepam or placebo plus clonazepam in status epilepticus (SAMUKeppra): a randomised, double-blind, phase 3 trial. Lancet Neurology. 2015. doi:10.1016/S1474-4422(15)00296-3
Silbergleit R, Durkalski V, Lowenstein D, et al, Barsan W. Intramuscular versus intravenous therapy for prehospital status epilepticus. New England Journal of Medicine. 2012. doi:10.1056/NEJMoa1107494
Nakken KO, Lossius MI. Buccal midazolam or rectal diazepam for treatment of residential adult patients with serial seizures or status epilepticus. Acta Neurologica Scandinavica. 2011. doi:10.1111/j.1600-0404.2010.01474.x
de Haan GJ, van der Geest P, Doelman G, et al, Edelbroek P. A comparison of midazolam nasal spray and diazepam rectal solution for the residential treatment of seizure exacerbations. Epilepsia. 2009. doi:10.1111/j.1528-1167.2009.02333.x
Fitzgerald BJ, Okos AJ, Miller JW. Treatment of out-of-hospital status epilepticus with diazepam rectal gel. Seizure. 2003. doi:10.1016/S1059-1311(02)00170-X
Alldredge BK, Gelb AM, Isaacs SM, et al, Lowenstein DH. A comparison of lorazepam, diazepam, and placebo for the treatment of out-of-hospital status epilepticus. New England Journal of Medicine. 2001. doi:10.1056/NEJMoa002141
Silbergleit R, Lowenstein D, Durkalski V, et al, Conwit R. RAMPART (Rapid Anticonvulsant Medication Prior to Arrival Trial): a double-blind randomized clinical trial of the efficacy of intramuscular midazolam versus intravenous lorazepam in the prehospital treatment of status epilepticus by paramedics. Epilepsia. 2011. doi:10.1111/j.1528-1167.2011.03235.x
Lowenstein DH, Alldredge BK, Allen F, et al, Gelb A. The prehospital treatment of status epilepticus (PHTSE) study: design and methodology. Controlled Clinical Trials. 2001. doi:10.1016/S0197-2456(01)00120-9
Kahan BC, Cro S, Doré CJ, et al, Jairath V. Reducing bias in open-label trials where blinded outcome assessment is not feasible: strategies from two randomised trials. Trials. 2014. doi:10.1186/1745-6215-15-456
Salazar J, Moustgaard H, Bracchiglione J, et al, Hróbjartsson A. Empirical evidence of observer bias in randomized clinical trials: updated and expanded analysis of trials with both blinded and non-blinded outcome assessors. Journal of Clinical Epidemiology. 2025. doi:10.1016/j.jclinepi.2025.111787
Al-Sofyani KA. An Insight into the Current Understanding of Status Epilepticus: From Concept to Management. Neurology Research International. 2021. doi:10.1155/2021/9976754
Brigo F, Del Giovane C, Nardone R, et al, Lattanzi S. Second-line treatments in benzodiazepine-resistant convulsive status epilepticus: An updated network meta-analysis including the ESET Trial - What did change? Epilepsy & Behavior. 2020. doi:10.1016/j.yebeh.2020.107035
Nevitt SJ, Sudell M, Cividini S, et al, Tudur Smith C. Antiepileptic drug monotherapy for epilepsy: a network meta-analysis of individual participant data. Cochrane Database of Systematic Reviews. 2022. doi:10.1002/14651858.CD011412.pub4
Yasiry Z, Shorvon SD. The relative effectiveness of five antiepileptic drugs in treatment of benzodiazepine-resistant convulsive status epilepticus: a meta-analysis of published studies. Seizure. 2013. doi:10.1016/j.seizure.2013.12.007
No conflicts of interest to disclose.
Research reported in this publication was supported by the National Institute of General Medical Sciences of the National Institutes of Health under Award Number U54 GM104942. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
None.