Antiseizure Medications in Poststroke Seizures

Abstract

Background and Objectives

The most effective antiseizure medications (ASMs) for poststroke seizures (PSSs) remain unclear. We aimed to determine outcomes associated with ASMs in people with PSS.

Methods

We systematically searched electronic databases for studies on patients with PSS on ASMs. Our outcomes were seizure recurrence, adverse events, drug discontinuation rate, and mortality. We assessed the risk of bias using Cochrane Risk of Bias tool for randomized controlled trials and Risk Of Bias In Non-randomized Studies of Interventions tools. Using levetiracetam as the reference treatment, we conducted a frequentist network meta-analysis and determined the certainty of evidence using the Grading of Recommendations Assessment, Development, and Evaluation methodology.

Results

Our search yielded 15 studies (3 randomized, 12 nonrandomized, N = 18,676 patients (121 early and 18,547 late seizures), 60% male, mean age 69 years) comparing 13 ASMs. Three studies had moderate and 12 had high risk of bias. Seizure recurrence was 24.8%. Compared with levetiracetam, very low-certainty evidence suggested that phenytoin was associated with higher seizure recurrences (odds ratio [OR] 7.3, 95% CI 3.7–14.5) and more adverse events (OR 5.2, 95% CI 1.2–22.9). Low-certainty evidence suggested that carbamazepine (OR 1.8, 95% CI 1.5–2.2) and phenytoin (OR 1.9, 95% CI 1.4–2.8) were associated with high drug discontinuation rates. Moderate to high-certainty evidence suggested that valproic acid (OR 4.7, 95% CI 3.6–6.3) and phenytoin (OR 8.3, 95% CI 5.7–11.9) were associated with higher mortality rates. Considering all treatments and using the GRADE approach for treatment ranking, very low-certainty evidence suggested that eslicarbazepine, lacosamide, and levetiracetam had the fewest seizure recurrences. Low to very low-certainty evidence suggested that lamotrigine had the fewest adverse events and drug discontinuations, whereas lamotrigine and levetiracetam exhibited low mortality rates with moderate-certainty evidence.

Discussion

We found that levetiracetam and lamotrigine may be safe and tolerable ASMs for PSS. Despite ASM use, the seizure recurrence rate remains high in the PSS population. Owing to bias and confounding risks, these findings should be interpreted cautiously.

Trial Registration Information

PROSPERO: CRD42022363844.

Introduction

Poststroke seizures (PSSs) are associated with increased mortality and poor quality of life.¹ Acute symptomatic seizures are seizures closely related to neurologic or systemic insults.² In case of stroke patients, seizures that occur within 7 days of stroke are termed as early seizures and those that arise after 7 days are termed as late seizures, also referred to as poststroke epilepsy.^2,3 Stroke is one of the most common causes of new-onset epilepsy in old age.⁴ The likelihood of recurrence after a first unprovoked late PSS is approximately 72% during the subsequent 10 years.⁵

Antiseizure medications (ASMs) are often given as a secondary prophylactic measure to prevent subsequent seizures after stroke.^6,7 Studies have investigated a range of ASMs for their impact on seizure recurrence and other clinical outcomes such as cognitive impairment or mortality in patients with epilepsy. However, there is variation in clinical practice regarding the choice of ASMs in managing PSS.^8,9 A recent guideline by the European Stroke Organisation failed to identify high-level evidence favoring any specific ASM or class of ASMs in those with PSS.⁷ Only 2 randomized controlled trials (RCTs) compared the efficacy and tolerability of ASMs in patients with PSS and identified that lamotrigine and levetiracetam were more effective and tolerable than carbamazepine; however, they were limited by small sample sizes (N = 170 patients) and fewer ASMs compared.^10,11 In addition, a retrospective cohort study examined the association between ASM monotherapy and mortality in patients with PSS and identified that lamotrigine was associated with lower mortality compared with carbamazepine¹²; however, it did not assess the adverse event profile and seizure recurrence rates associated with these ASMs. High-level evidence is lacking to support any specific ASM in the PSS population.

We recognized the need to investigate ASM-associated adverse events, seizure recurrence, and drug discontinuation rate in addition to mortality in patients with PSS. We adopted a robust systematic review and network meta-analysis methodology to collate the available evidence and determine the ASM(s) associated with least seizure recurrence, adverse events, drug discontinuation, and mortality in patients with PSS.

Methods

Literature Search

We comprehensively searched databases, including Ovid MEDLINE, EMBASE, Web of Science, Scopus, Cochrane, and PsycInfo until October 17, 2023. We used Covidence to screen the results from the literature search. Examples of key terms used for searching include “Seizures,” “Epilepsy,” “Stroke,” “Ischemic Stroke,” “Intracerebral hemorrhage,” “Hemorrhagic Stroke,” “Infarct,” “Antiepileptic,” “Antiseizure,” “Drug,” “Treatment,” “Intervention,” “Medication,” “Post Stroke Epilepsy,” “Epileptogenesis,” “Tolerability,” “Dose,” “Efficacy,” “Mortality,” “Functional Outcome.” The detailed search strategy is provided in eAppendix 1.

Eligibility Criteria

We included prospective and retrospective studies that consisted of (1) adults age 18 years and older with postischemic stroke or intracerebral hemorrhage epileptic seizures, (2) patients with PSS (early and late seizures) taking any ASM, and (3) relevant outcome data from patients with PSS taking any ASM. We included studies conducted in humans and did not apply restrictions on language, publication date, or sample size. Owing to the lack of uniformity and reporting regarding the classification and definition of early and late-onset seizures, we accepted the definitions reported by individual studies.

We excluded studies that did not report relevant outcome data, duplicate publications, narrative or systematic reviews, conference proceedings, dissertations, preprints, ongoing/unpublished studies, and studies without available full texts.

Standard Protocol Approvals, Registrations, and Patient Consents

We registered the protocol of this systematic review on PROSPERO (registration CRD42022363844).¹³ Because this was a systematic review and network meta-analysis of published studies, no ethics committee approval or protocol approvals were required. No informed consent or authorization for disclosure was required to conduct this systematic review.

Outcomes

Our outcome measures include seizure recurrence, adverse events, drug discontinuation, and mortality. We defined drug discontinuation as treatment discontinuation due to adverse events or switching medications because of ineffectiveness. However, the studies need more clarity on the methodology for assessing seizure recurrence, adverse events, and drug discontinuation. The authors merely noted that these outcomes were monitored during the follow-up period.

Data Extraction

We followed the Preferred Reporting Items for Systematic Reviews and Network Meta-Analyses 2015 guidelines standards.¹⁴ Five study authors (S.M., T.H., G.H., S.P., and P.D.) independently screened titles and abstracts of the retrieved articles using Covidence software in duplicate. Subsequently, they screened the full-text articles for inclusion in duplicate. Any disagreements were resolved by consulting with an experienced clinician and reviewer (N.K.M.). We extracted the following data from the eligible studies: first author, publication year, sample size, age, sex, ethnicity, stroke subtypes, risk factors, early seizures, late seizures, ASM type (tested drug and comparator), International Classification of Disease (ICD) code, seizure recurrence, adverse-event profile, patient withdrawals from ASM(s), and mortality. If necessary, the original study authors were emailed twice to acquire the missing data.

Risk of Bias Assessment

We used the Cochrane Risk of Bias (ROB-2) tool for randomized controlled trials¹⁵ and the Risk Of Bias In Non-randomized Studies of Interventions (ROBINS-I) tool for nonrandomized studies of interventions¹⁶ included in our systematic review.

Statistical Analysis

We calculated pooled odds ratio (OR) and 95% CI and performed a pairwise meta-analysis initially to compare the various ASM(s). We assessed heterogeneity for each pairwise comparison using I² and Cochrane Q test. We used a fixed-effect model if I² ≤50% and p > 0.05. Else, we pooled the results using a random-effect model. We categorized heterogeneity into low (I² <25%), moderate (I² = 25%–75%), and high (I² >75%).¹⁷ We made comparisons using 3 types of evidence in our network meta-analysis: direct, indirect, and mixed. Direct evidence refers to head-to-head comparisons of 2 treatments within the same study (e.g., treatment A vs treatment B). Indirect evidence compares 2 treatments not compared directly in any study through a common comparator (e.g., comparing A and B through a shared link with treatment C). Mixed evidence combines both direct and indirect evidence for the same comparison, providing a more precise effect estimate. We used the “netmeta” package in R version 4.3.1 to conduct the frequentist network meta-analysis. We created network graphs using the “netgraph” function. We visualized the proportion of direct and indirect evidence using direct evidence plots (“direct.evidence.plot” function). We analyzed publication bias using the comparison-adjusted funnel plots and quantitatively assessed using Egger test p value. We ranked the treatments using P-scores (“netrank” function). Using the net splitting method (“net.split” function), we visualized the inconsistencies between direct and indirect estimates in our network model. We conducted a network meta-regression analysis which is similar to standard regression, but instead of analyzing individual patient data, it uses entire studies as the units of analysis. This method allows us to investigate whether differences in study characteristics, such as risk of bias, study design, or seizure subtypes, explain variations in the results across the included studies. We used the Bayesian network meta-regression analysis method (“gemtc” package in R version 4.3.1) to explore whether the risk of bias, seizure subtypes, and study design influenced the magnitude of effect sizes in our network.

Certainty of Evidence

We used the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) approach to rate the certainty/quality of evidence for each network estimate.^18,19 For direct estimates, we rated down for risk of bias, inconsistency, indirectness, and publication bias. We rated up for large-effect, dose-response, and antagonistic bias in the case of nonrandomized studies. The indirect estimates were based on the lowest ratings of the 2 direct comparisons forming the first-order loop and were rated down for intransitivity. The network estimates were based on the rating of direct or indirect estimates or the estimate that contributed the most and were rated down for incoherence and imprecision. Two study authors (S.M. and A.E.) rated the certainty of evidence for each outcome measure. We used κ statistics to evaluate the inter-rater agreement.

Categorization of Interventions

We used the wordings from GRADE's Minimally Contextualized Framework approach to categorize interventions from least to most harmful based on effect estimates, certainty of evidence, and P-score rankings.²⁰ We a priori selected levetiracetam as our reference treatment for consistency and to serve as a common comparator. We selected a decision threshold where treatment in comparison was considered superior only if 95% CI excluded the null, that is, p < 0.05. Because all our outcomes were “harm outcomes,” a standard wording recommended by the GRADE framework,²⁰ we categorized treatments as more/less harmful than the reference (i.e., levetiracetam) for a given outcome measure. We first classified the treatments based on comparison with reference and categorized them as category 0: statistically less harmful than the reference and some alternatives (95% CI <1); category 1: no statistical difference compared with the reference, that is, no more harmful than the reference (95% CI includes the null) but statistically less harmful than some alternatives; category 2: no more harmful than the reference (95% CI includes the null); and category 3: statistically more harmful than the reference (95% CI >1) but no other alternatives. Furthermore, we classified the treatments based on pairwise comparisons and categorized them as category 4: statistically more harmful than the reference and some alternatives (95% CI >1). We then classified interventions as those with high-certainty or moderate-certainty evidence and those with low-certainty or very low-certainty evidence relative to the reference treatment, that is, levetiracetam.

Data Availability

Data are presented in this study, its supplementary materials, or within the texts or appendices of the included studies. Data and analysis codes are available on reasonable request from the corresponding author.

Results

Our literature search yielded 2,412 articles, of which 101 full texts were assessed for eligibility. See eTable 1 for the list of studies excluded after full-text review. We identified 15 eligible studies^10-12,21-32 in our systematic review and network meta-analysis (Figure 1). The studies reported investigations on the following ASMs: levetiracetam, carbamazepine, phenytoin, valproic acid, lamotrigine, lacosamide, phenobarbital, oxcarbazepine, zonisamide, eslicarbazepine, brivaracetam, perampanel, and topiramate. Fourteen studies reported ASM monotherapy,^{10-12,21-30,32} and 1 study reported a combination of lamotrigine and valproic acid.³¹ No study included the use of 2 or more ASMs within a single patient group. We included 3 RCTs^10,11,27 and 12 nonrandomized interventional studies (2 prospective^28,31 and 10 retrospective cohorts^{12,21-26,29,30,32}). The follow-up duration ranged from 6 months to 5 years. We included studies from 9 countries: 3 studies from Italy^11,22,27, 2 each from Japan,^28,29 Sweden,^12,25 Taiwan,^23,24 and Germany^30,32; and 1 study each from China,³¹ Israel,¹⁰ the Netherlands,²¹ and Tunisia.²⁶ Studies were published from 2007 to 2023 in English.

Figure 1. PRISMA Flow Diagram.

ASM = antiseizure medication; PRISMA = Preferred Reporting Items for Systematic Reviews and Meta-Analyses.

The studies included data on 18,676 patients with PSS (60% male patients), of whom 121 had early seizures, 18,547 had late seizures, and 8 had both early and late seizures. Seven studies defined early/late-onset seizures using the 7-day cutoff, 5 used the 14-day cutoff, and 3 did not provide any definition. See Table 1 for baseline characteristics of the included studies. There were 13,067 patients with ischemic stroke and 5,567 patients with hemorrhagic stroke. Five of 15 studies diagnosed PSS using ICD codes (including 17,462 patients, 94.1%).^12,22-25

Table 1.

Baseline Characteristics of Studies Included in the Systematic Review and Network Meta-Analysis

S. No.	Study	Country	Study design	PSS (N)	Early/late seizures	IS/HS (PSS)	Age, mean (SD)	Male, n (%)	Follow-up duration	ASM	Outcome measures	ICD codes	Seizure definition
1	Ouerdiene et al., 202326	Tunisia	Retrospective cohort	52	21/31	52/NA	55.1 (median)	39 (75)	6 mo	VPA, CBZ, PBR	Seizure recurrence	No	7 d
2	Winter et al., 202332	Germany	Retrospective cohort	138	NA/138	138/NA	70.8 (8.1)	71 (51.4)	NA	BRV, LCM, PER, ESL, TPM, ZNS	Seizure recurrence	No	7 d
3	Larsson et al., 202212	Sweden	Retrospective cohort	2,577	NA/2,577	2,164/413	78 (69–85)	1,400 (54.3)	5 y	LTG, OXC, LEV, VPA, CBZ, PHT	Mortality	Yes	7 d
4	Winter et al., 202230	Germany	Retrospective cohort	181	NA/181	181/NA	68.1 (8.39)	95 (52.3)	1 y	LEV, LCM, LTG, ESL	Seizure recurrence	No	Updated ILAE
5	Costa et al., 202122	Italy	Retrospective cohort	275	NA/275	149/126	64 (16)	157 (57.1)	1 y	LEV, VPA, OXC	Drug discontinuation	Yes	7 d
6	Hsu et al., 202123	Taiwan	Retrospective cohort	5,997	NA/5,997	3,952/2,045	67.8 (14.05)	3,771 (62.9)	5 y	CBZ, PHT, VPA	Mortality	Yes	2 wk
7	Tanaka et al., 202128	Japan	Prospective cohort	322	NA/322	199/114	74 (65–82)	195 (60.6)	371 (IQR 347–420) d	CBZ, VPA, LEV, LTG, LCM, ZNS, PHT	Seizure recurrence, AE, drug discontinuation, mortality	No	7 d
8	Bekelaar et al., 202021	Netherlands	Retrospective cohort	53	4/41	45/8	59 (13)	36 (67.9)	62 (IQR 69) mo	CBZ, LEV, PHT, VPA	AE, drug discontinuation	No	7 d
9	Tao et al., 202031	China	Prospective cohort	145	91/54	80/34	59.95 (8.82)	81 (55.9)	1 y	VPA + LTG, VPA	AE, seizure recurrence	No	2 wk
10	Larsson et al., 201925	Sweden	Retrospective cohort	4,991	NA/4,991	4,152/839	74 (12)	2,732 (54.7)	5 y	LTG, OXC, LEV, VPA, CBZ, PHT	Drug discontinuation	Yes	Not defined
11	Tanaka et al., 201529	Japan	Retrospective cohort	104	NA/104	69/43	74 (63.3–81)	71 (68.3)	362 (IQR 172–552) d	VPA, CBZ, PHT	Seizure recurrence	No	2 wk
12	Huang et al., 201524	Taiwan	Retrospective cohort	3,622	NA/3,622	1,729/1,893	60.3 (13.1)	2,450 (67.6)	1 y	PHT, VPA, CBZ	Seizure recurrence	Yes	2 wk
13	Siniscalchi et al., 201427	Italy	RCT	49	NA/49	24/25	73.09 (18.03)	26 (53.1)	30 mo	LEV, PBR	Seizure recurrence	No	Not defined
14	Consoli et al., 201211	Italy	RCT	106	NA/106	69/27	71.86 (12.27)	58 (54.7)	1 y	LEV, CBZ	AE, seizure recurrence, drug discontinuation	No	Not defined
15	Gilad et al., 200710	Israel	RCT	64	5/59	64/NA	67.45 (2.5)	46 (71.9)	1 y	LTG, CBZ	AE, seizure recurrence, drug discontinuation	No	2 wk

Abbreviations: AE = adverse event; ASM = antiseizure medication; BRV = brivaracetam; CBZ = carbamazepine; ESL = eslicarbazepine; HS = hemorrhagic stroke; ICD = International Statistical Classification of Diseases and Related Health Problems, Tenth Revision; ILAE = International League Against Epilepsy; IS = ischemic stroke; LCM = lacosamide; LEV = levetiracetam; LTG = lamotrigine; NA = not applicable; OXC = oxcarbazepine; PBR = phenobarbital; PER = perampanel; PHT = phenytoin; PSS = poststroke seizure; RCT = randomized controlled trial; TPM = topiramate; VPA = valproic acid; ZNS = zonisamide.

Risk of Bias Assessment

Overall, 3 studies (20%)^11,12,23 had a moderate risk of bias and 12 (80%)^{10,21,22,24-32} had a high risk of bias. No study had a low risk of bias. Using the ROB-2 tool for RCTs, 1 study had a moderate¹¹ risk of bias and 2 studies^10,27 had a high risk of bias (eFigure 1 and eTable 2). Using the ROBINS-I tool for noninterventional studies, 2 studies had a moderate^12,23 risk of bias and 10 had a high risk of bias (eFigure 2 and eTable 3).^{21,22,24-26,28-32}

Seizure Recurrence

We observed a 24.8% seizure recurrence rate within a follow-up duration ranging from 6 months to 2.5 years. We constructed a network of 13 treatments from 10 studies (3 RCTs^10,11,27 and 7 non-RCTs^24,26,28-32) (Figure 2A). From the pairwise meta-analysis, we obtained 49 comparisons, including 4,470 patients with PSS with moderate heterogeneity (I² = 39.5%). In the network meta-analysis, we obtained 76 paired estimates, of which 29 had direct and indirect evidence, 6 had only direct evidence, and 41 had only indirect evidence (eFigure 3).

Figure 2. Network Diagram of Treatment Comparisons for (A) Seizure Recurrence, (B) Adverse Events, (C) Drug Discontinuation, and (D) Mortality.

The node size in red represents the total number of patients with PSS taking ASMs, and the edge thickness in black represent the total number of studies comparing ASMs in patients with PSS. ASM = antiseizure medication; PSS = poststroke seizure.

When considering all the interventions, moderate-certainty evidence suggested that carbamazepine (OR 2.97, 95% CI 1.65–5.34) was more harmful than levetiracetam but less harmful than phenytoin.

Very low-certainty evidence suggested that eslicarbazepine (OR 0.52, 95% CI 0.25–1.10), lacosamide (OR 0.64, 95% CI 0.34–1.22), lamotrigine (OR 1.13, 95% CI 0.57–2.23), perampanel (OR 1.86, 95% CI 0.56–6.18), and brivaracetam (OR 1.91, 95% CI 0.60–6.09) may be no more harmful than levetiracetam (statistically not different from the reference) but may be less harmful than some other alternatives and associated with fewer seizure recurrences. Very low-certainty evidence suggested that zonisamide (OR 2.10, 95% CI 0.58–7.59), topiramate (OR 2.12, 95% CI 0.62–7.26), and phenobarbital (OR 2.58, 95% CI 0.83–8.04) may be no more harmful than levetiracetam (statistically not different from the reference). Valproic acid (OR 4.19, 95% CI 2.11–8.32) (very low certainty) and lamotrigine + valproic acid (OR 5.76, 95% CI 2.07–16.05) (low certainty) may be more “harmful” (terminology used per GRADE recommendation) than levetiracetam but less harmful than phenytoin. When considering all interventions, very low-certainty evidence suggested that phenytoin (OR 7.33, 95% CI 3.71–14.50) may be the most harmful treatment associated with poor seizure control (Figure 3A, Table 2, and eTable 4).

Figure 3. Forest Plots for the Association of Various Treatments With (A) Seizure Recurrence, (B) Adverse Events, (C) Drug Discontinuation, and (D) Mortality Compared With Levetiracetam (Reference Treatment).

ASM = antiseizure medication; LEV = levetiracetam.

Table 2.

Final Classification of 13 Interventions, Based on Network Meta-Analysis of ASM Treatments of Patients With PSS

Outcomes	Certainty of evidence	Treatment (CoE)	OR (95% CI) vs LEV	P-score ranking
Seizure recurrence	High certainty (high to moderate evidence)
	Category 0: Among the least harmful	—	—	—
	Category 1: Among the less harmful	—	—	—
	Category 2: Among no more harmful
	Category 3: Inferior to no more/less harmful or superior to the most harmful	CBZ (M)	2.97 (1.65–5.34)	0.36
	Category 4: Among the most harmful	—	—	—
	Low certainty (low to very low evidence)
	Category 0: Might be among the least harmful	—	—	—
	Category 1: Might be among the less harmful	ESL (VL)	0.52 (0.25–1.10)	0.97
		LCM (VL)	0.64 (0.34–1.22)	0.92
		LEV	—	0.76
		LTG (VL)	1.13 (0.57–2.23)	0.72
		PER (VL)	1.86 (0.56–6.18)	0.52
		BRV (VL)	1.91 (0.60–6.09)	0.51
	Category 2: Might be among no more harmful	ZNS (VL)	2.10 (0.58–7.59)	0.47
		TPM (VL)	2.12 (0.62–7.26)	0.47
		PBR (VL)	2.58 (0.83–8.04)	0.41
	Category 3: Might be inferior to no more/less harmful or superior to the most harmful	VPA (VL)	4.19 (2.11–8.32)	0.22
		LTG + VPA (L)	5.76 (2.07–16.05)	0.13
	Category 3: Might be among the most harmful	PHT (VL)	7.33 (3.71–14.50)	0.04
Adverse events	Low certainty (low to very low evidence)
	Category 0: Might be among the least harmful
	Category 1: Might be among the less harmful	LTG (L)	0.62 (0.16–2.49)	0.88
		LEV	—	0.76
	Category 2: Might be among no more harmful	LCM (VL)	1.57 (0.19–13.14)	0.54
		VPA (VL)	2.05 (0.41–10.30)	0.49
		LTG + VPA (VL)	3.32 (0.56–19.66)	0.25
	Category 3: Might be inferior to no more/less harmful or superior to the most harmful	CBZ (VL)	1.89 (1.02–3.53)	0.45
	Category 4: Might be among the most harmful	PHT (VL)	5.16 (1.16–22.90)	0.14
Drug discontinuation	Low certainty (low to very low evidence)
	Category 0: Might be among the least harmful	—	—	—
	Category 1: Might be among the less harmful	LTG (VL)	0.95 (0.72–1.25)	0.83
		LEV	—	0.78
	Category 2: Might be among no more harmful	LCM (VL)	0.60 (0.08–4.75)	0.79
		OXC (VL)	1.36 (0.87–2.12)	0.5
		ZNS (VL)	7.17 (0.44–117.39)	0.12
	Category 3: Might be inferior to no more/less harmful or superior to the most harmful	VPA (VL)	1.30 (1.04–1.64)	0.53
	Category 4: Might be among the most harmful	CBZ (L)	1.78 (1.46–2.18)	0.25
		PHT (L)	1.93 (1.35–2.76)	0.20
Mortality	High certainty (high to moderate evidence)
	Category 0: Among the least harmful	—	—	—
	Category 1: Among the less harmful	LEV	—	0.99
		LTG (M)	1.33 (0.98–1.81)	0.81
	Category 2: Among no more harmful	—	—	—
	Category 3: Inferior to no more/less harmful or superior to most harmful	CBZ (M)	2.13 (1.68–2.70)	0.58
		OXC (M)	3.33 (1.72–6.45)	0.39
	Category 4: Among the most harmful	VPA (H)	4.74 (3.55–6.34)	0.23
		PHT (M)	8.27 (5.74–11.92)	0.001

Abbreviations: ASM = antiseizure medication; BRV = brivaracetam; CBZ = carbamazepine; CoE = certainty of evidence; H = high; L = low; LCM = lacosamide; LEV = levetiracetam; LTG = lamotrigine; M = moderate; OXC = oxcarbazepine; PBR = phenobarbital; PER = perampanel; PHT = phenytoin; TPM = topiramate; VL = very low; VPA = valproic acid; ZNS = zonisamide.

Category 0: less harmful than the reference and some alternatives; category 1: no more harmful than the reference but less harmful than some alternatives; category 2: no more harmful than the reference; category 3: more harmful than the reference but no worse than any alternatives; category 4: more harmful than the reference and other options.

Adverse Events

The most common adverse events included allergy (19.8%), excessive sleepiness (14.3%), mood symptoms (14.3%), skin eruption (9.4%), and nausea and vomiting (8.9%). The adverse-event profile with different ASMs is given in eTable 5. We constructed a network of 7 treatments from 5 studies (2 RCTs^10,11 and 3 non-RCTs^21,28,31) (Figure 2B). From the pairwise meta-analysis, we obtained 19 comparisons, including 666 patients with PSS with moderate heterogeneity (I² = 50%). In the network meta-analysis, we obtained 21 paired estimates, of which 13 had direct and indirect evidence, 1 had only direct evidence, and 7 had only indirect evidence (eFigure 4).

Very low-certainty evidence suggested that lamotrigine may be no more harmful than levetiracetam (OR 0.62, 95% CI 0.16–2.49) (statistically not different from the reference) but less harmful than phenytoin and associated with fewer adverse events; lacosamide (OR 1.57, 95% CI 0.19–13.14), valproic acid (OR 2.05, 95% CI 0.41–10.30), and lamotrigine + valproic acid (OR 3.32, 95% CI 0.56–19.66) may be no more harmful than levetiracetam (statistically not different from the reference). Very low-certainty evidence suggested that carbamazepine (OR 1.89, 95% CI 1.02–3.53) may be more harmful than levetiracetam but not phenytoin. When considering all interventions, very low-certainty evidence suggested that phenytoin (OR 5.16, 95% CI 1.16–22.90) may be the most harmful treatment associated with the highest adverse events (Figure 3B, Table 2, eTable 6).

Drug Discontinuation

We constructed a network of 8 treatments from 6 studies (2 RCTs^10,11 and 4 non-RCTs^21,22,25,28) (Figure 2C). The pairwise meta-analysis obtained 41 comparisons, including 5,413 patients with PSS with moderate heterogeneity (I² = 41.4%). In the network meta-analysis, we obtained 28 paired estimates, of which 21 had direct and indirect evidence, 3 had only direct evidence, and 4 had only indirect evidence (eFigure 5).

Very low-certainty evidence suggested that lamotrigine may be no more harmful than levetiracetam (OR 0.95, 95% CI 0.72–1.25) but less harmful than carbamazepine, phenytoin, and valproic acid and may be associated with high retention rates. Very low-certainty evidence suggested that lacosamide (OR 0.60, 95% CI 0.08–4.75), oxcarbazepine (OR 1.36, 95% CI 0.87–2.12), and zonisamide (OR 7.17, 95% CI 0.44–117.39) may be no more harmful than levetiracetam (statistically not different from the reference); valproic acid (OR 1.30, 95% CI 1.04–1.64) may be more harmful than levetiracetam but less harmful than carbamazepine and phenytoin. When considering all interventions, low-certainty evidence suggested that carbamazepine (OR 1.78, 95% CI 1.46–2.18) and phenytoin (OR 1.93, 95% CI 1.35–2.76) may be among the most harmful treatments associated with least retention rates (Figure 3C, Table 2, eTable 7).

Mortality

We constructed a network of 6 treatments from 3 studies (all non-RCTs)^12,23,28 (Figure 2D). From the pairwise meta-analysis, we obtained 19 comparisons, including 9,739 patients with PSS with no substantial heterogeneity (I² = 0%). In the network meta-analysis, we obtained 15 paired estimates, all of which had direct and indirect evidence (eFigure 6).

Moderate-certainty evidence suggested that lamotrigine (OR 1.33, 95% CI 0.98–1.81) was no more harmful than levetiracetam but less harmful than carbamazepine, oxcarbazepine, phenytoin, and valproic acid and had reduced mortality rates. Carbamazepine (OR 2.13, 95% CI 1.68–2.70) and oxcarbazepine (OR 3.33, 95% CI 1.72–6.45) were more harmful than levetiracetam and lamotrigine but less harmful than valproic acid and phenytoin. When considering all interventions, valproic acid (OR 4.74, 95% CI 3.55–6.34) (high certainty) and phenytoin (OR 8.27, 95% CI 5.74–11.92) (moderate certainty) were among the most harmful treatments associated with highest mortality rates (Figure 3A, Table 2, eTable 8).

Publication Bias

We did not detect publication bias for any outcome measure. See the funnel plots displayed in eFigure 7.

Network Meta-Regression Analysis

Neither study design nor risk of bias or seizure subtypes were associated with network effect estimates (eTable 9).

Discussion

This systematic review and network meta-analysis synthesized evidence from 15 studies and tested the strength of the association of 13 ASMs with seizure recurrence, adverse events, drug discontinuation, and mortality in patients with PSS. Levetiracetam exhibited fewer seizure recurrences and adverse events and the lowest mortality risk. Lamotrigine emerged as the ASM with the least adverse events, lowest drug discontinuation rate, and lower mortality risk. Eslicarbazepine and lacosamide had the least seizure recurrences among the treatments considered. Still, because of their novelty, they could not be evaluated for mortality or discontinuation because of a lack of available data.

Using the GRADE methodology, we categorized our data to determine the certainty of evidence. We report them in Figure 4, which will guide physicians when choosing ASMs for patients with PSS. We observed substantial agreement between the 2 study authors in rating the GRADE certainty of evidence for all outcome measures (eTable 10). Most included studies had a high risk of bias (80%), with none attaining a low-risk status.

Figure 4. NMA Results Sorted Based on GRADE Certainty of Evidence and Effect Estimate for the Comparisons of ASM Treatments for Seizure Recurrence, Adverse Events, Drug Discontinuation, and Mortality in Patients With PSS.

ASM = antiseizure medication; NMA = network meta-analysis; PSS = poststroke seizure.

Achieving and maintaining seizure freedom is a primary goal in epilepsy management. In our study, patients treated with eslicarbazepine, lacosamide, or levetiracetam demonstrated the most favorable seizure control outcomes. Although we graded their level of evidence as low to very low, particularly because of the risk of bias, inconsistency, incoherence, and imprecision (eTable 4), the findings are not surprising. These agents are safer than old-generation ASMs. For example, we found that phenytoin was associated with the highest seizure recurrences among all the treatments, very likely because of the high discontinuation rate and narrow therapeutic index, which, at times, is difficult to maintain in the stroke population. Similarly, carbamazepine, valproic acid, and lamotrigine + valproic acid also exhibited significant association with more seizure recurrences compared with levetiracetam.

Notably, 2 RCTs^10,11 in patients with PSS did not identify any statistically significant differences in seizure freedom when comparing carbamazepine with levetiracetam and lamotrigine; we believe this was possibly due to small sample sizes. Furthermore, indirect evidence from a previous network meta-analysis found no difference in seizure freedom for carbamazepine compared with levetiracetam and lamotrigine.³³ By contrast, our network meta-analysis with moderate-certainty evidence showed a significant association between carbamazepine and high seizure recurrences compared with levetiracetam. Given that the individual patient data are lacking, we cannot determine the cause of the greater recurrence rate in this population. A study examined the effectiveness and tolerability of levetiracetam in patients with PSS age 60 years and older. Their findings showed that 82.4% of patients achieved seizure freedom.³⁴ Consistent with our findings, previous studies have identified fewer seizure recurrences for eslicarbazepine and lacosamide in patients with and without PSS.^30,35,36

Our analyses show that lamotrigine is associated with fewer adverse events, whereas phenytoin had the most. Whereas we graded the former with low-certainty evidence and the latter with very low-certainty evidence, the findings are not surprising.³⁷ Carbamazepine also exhibited substantially higher adverse events than levetiracetam. These findings align with previous observations regarding the tolerability and side-effect profiles of older generation ASMs.²⁸ Our results align with a prior network meta-analysis, which identified fewer adverse events with lamotrigine and levetiracetam than with carbamazepine.³³ Levetiracetam demonstrated better tolerability profiles, particularly in older individuals with epilepsy.³⁸ Lamotrigine was well tolerated in the SANAD trial, which compared newer generation ASMs with carbamazepine in patients with focal epilepsy.³⁹ In addition, a survey of 42 epileptologists in the United States further supported the preference for lamotrigine and levetiracetam over older ASMs in the elderly population.⁴⁰ Preference for levetiracetam mainly was attributed to its favorable efficacy, ease of administration, minimal potential for drug interactions, suitability for rapid oral loading, and the presence of intravenous formulations. These differences in adverse-event profiles highlight the importance of selecting ASMs that minimize the burden on patients while effectively controlling seizures.

Whether drug resistance is a significant problem in patients with PSS is currently not established, although it is a very relevant outcome measure for patients with PSS. High rates of drug discontinuation, whether because of adverse events or treatment ineffectiveness, can lead to suboptimal seizure control and adverse clinical outcomes and contribute to drug resistance.^41,42 Whereas we could not analyze the data for drug resistance, we could for the drug discontinuation rate. In our analysis, lamotrigine with very low-certainty evidence may be associated with the fewest drug discontinuations. Our findings align with a nationwide cohort study, and 2 RCTs identifying levetiracetam and lamotrigine had better retention rates than carbamazepine.^10,11,25 Data regarding treatment switches over time in PSS are scarce. The tolerance profile and potential drug interactions are critical for switching between ASMs. Previous RCTs reported a consistent risk of treatment discontinuation, with 3%–31% of individuals withdrawing from their initial ASM because of adverse events.^10,11 This dropout is also commonly observed in prospective studies and can affect seizure management and hospitalization rates.^43,44 Therefore, promptly identifying adverse events using standardized scales and considering treatment adjustments are crucial.⁴⁵

There is significant variation in practice regarding the duration of ASM used in patients with early seizures. Whereas some might consider stopping it at 7 days, others might use it for a prolonged duration. The decision to stop ASM after an early seizure is guided by various factors such as patient preference, side effects, and, at times, the perceived low seizure recurrence risk in certain stroke subtypes.

We found moderate to high-certainty evidence that levetiracetam and lamotrigine were associated with lower mortality risk. By contrast, valproic acid and phenytoin were significantly associated with the highest mortality risk than other treatments in the network. Carbamazepine and oxcarbazepine also exhibited significantly higher mortality rates than levetiracetam. Our findings are consistent with a recent study by Larsson et al.¹² and a Danish registry-based cohort that revealed decreased mortality associated with lamotrigine.⁴⁶

There is no straightforward explanation for the differences in mortality rates between the ASMs assessed. Given the differences in patient characteristics, concomitant drug use, and comorbidities, clinical practice significantly varies for the choice of ASM.³ Older ASMs, such as phenytoin, carbamazepine, valproic acid, or phenobarbital, are associated with increased mortality in patients with epilepsy, consistent with our findings.^46,47 For example, phenytoin use in patients with PSS has been linked to a higher death risk,²³ very likely because of its effect on a patient's cardiac rhythm. However, other ASMs, such as valproic acid, were also associated with higher mortality rates compared with lamotrigine and levetiracetam, despite not being known to cause cardiac arrhythmias. We provide a table with ASM's mechanism of action and their interaction to guide its use in patients with PSS (eTable 11).

Our study has several limitations. First, the included studies exhibited variations in design and follow-up durations. This and the practice patterns can introduce statistical and clinical heterogeneity in our results. Variability in seizure definition is listed in Table 1. These may have an impact, particularly on the outcome of seizure recurrence. Second, the high risk of bias in many studies could affect the validity of our findings. Third, the sample size per study ranged from 49 to 5,997. We adjusted for individual study sizes when rating the certainty of evidence and in the risk-of-bias assessment, using the weighted overall risk-of-bias scores based on sample size. Meta-regression analysis showed no influence of risk-of-bias scores on the overall network effect estimates (eTable 9). Fourth, we combined evidence from RCTs and non-RCTs for seizure recurrence, adverse events, and drug discontinuation. However, the transitivity assumption of network meta-analysis was not violated. In addition, we conducted a network meta-regression analysis and found that the study design did not influence the network effect estimates (eTable 9). Fifth, unmeasured confounders may also influence the results. Sixth, we could not stratify the treatment comparisons based on seizure and stroke subtypes because of a lack of available data. We, however, conducted network meta-regression analysis for seizure recurrence, adverse events, and drug discontinuation outcome measures and identified no influence of seizure subtypes (early + late and late seizures) on the overall network effect estimates (eTable 9). Seventh, although we demonstrated poor outcomes in patients with PSS in our previous work,¹ we could not systematically evaluate the influence of ASMs on patient outcomes. Eighth, data on ASM doses were unavailable, which could affect the efficacy, adverse events, discontinuation rates, and mortality associated with ASMs. Finally, many newer ASMs have been introduced in the market, and because outcome data on these agents have not been reported in patients with PSS, we cannot provide evidence on them.

Currently, recommendations for PSS therapy are primarily based on studies conducted on older adults with epilepsy of diverse etiology. Although there is no evidence to support the use of certain ASMs, experts have recommended lamotrigine and levetiracetam as effective therapy alternatives.⁴⁸ The results of our network meta-analysis lend support to these recommendations. Levetiracetam seems to be a favorable choice in terms of lower mortality and fewer seizure recurrences when compared with other ASMs. At the same time, lamotrigine demonstrated the least adverse events, fewest drug discontinuations, and low mortality rates. These results will guide neurologists in selecting an appropriate ASM for patients with PSS, considering both efficacy and safety profiles.

Glossary

ASMs

antiseizure medications

GRADE

Grading of Recommendations Assessment, Development, and Evaluation

ICD

International Classification of Disease

OR

odds ratio

PSS

post-stroke seizure

RCT

randomized controlled trial

ROB-2

Cochrane Risk of Bias tool for randomized controlled trials

ROBINS-I

Risk Of Bias In Non-randomized Studies of Interventions

Author Contributions

S. Misra: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data. S. Wang: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. T.J. Quinn: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. J. Dawson: drafting/revision of the manuscript for content, including medical writing for content. J. Zelano: drafting/revision of the manuscript for content, including medical writing for content. T. Tanaka: drafting/revision of the manuscript for content, including medical writing for content. J.C. Grotta: drafting/revision of the manuscript for content, including medical writing for content. E. Khan: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. N. Beriwal: drafting/revision of the manuscript for content, including medical writing for content. M.C. Funaro: major role in the acquisition of data; study concept or design. S. Perla: major role in the acquisition of data. P. Dev: major role in the acquisition of data. D. Larsson: drafting/revision of the manuscript for content, including medical writing for content. T. Hussain: major role in the acquisition of data. D.S. Liebeskind: drafting/revision of the manuscript for content, including medical writing for content. C.L. Yasuda: drafting/revision of the manuscript for content, including medical writing for content. H.H. Altalib: drafting/revision of the manuscript for content, including medical writing for content. H.P. Zaveri: drafting/revision of the manuscript for content, including medical writing for content. A Elshahat: major role in the acquisition of data. G. Hitawala: major role in the acquisition of data. E.Y. Wang: major role in the acquisition of data. R. Kitagawa: major role in the acquisition of data. A. Pathak: drafting/revision of the manuscript for content, including medical writing for content. F. Scalzo: drafting/revision of the manuscript for content, including medical writing for content. M. Ihara: drafting/revision of the manuscript for content, including medical writing for content. K.S. Sunnerhagen: drafting/revision of the manuscript for content, including medical writing for content. M.R. Walters: drafting/revision of the manuscript for content, including medical writing for content. Y. Zhao: drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data. N. Jette: drafting/revision of the manuscript for content, including medical writing for content. S.E. Kasner: drafting/revision of the manuscript for content, including medical writing for content. P. Kwan: drafting/revision of the manuscript for content, including medical writing for content. N.K. Mishra: drafting/revision of the manuscript for content, including medical writing for content; study concept or design; analysis or interpretation of data.

Study Funding

No targeted funding reported.

Disclosure

S. Misra, S Wang, T.J. Quinn, and J. Dawson report no disclosures relevant to the manuscript. J. Zelano has received speaker honoraria from UCB and Eisai and has, as an employee of Sahlgrenska University Hospital (no personal compensation), been an investigator in clinical trials sponsored by Bial, UCB, GW Pharma, SK Life Science, and Angilini Pharma. T. Tanaka reports lecture fees from Daiichi Sankyo and Eisai. J.C. Grotta, E. Khan, N. Beriwal, M.C. Funaro, S. Perla, P. Dev, D. Larsson, and T. Hussain report no disclosures relevant to the manuscript. D.S. Liebeskind is a consultant as Imaging Core Lab to Cerenovus, Genentech, Medtronic, Stryker, and Rapid Medical. C.L. Yasuda reports lecture fees from UCB, Adium, Torrent, and Libbs. H.H. Altalib, H.P. Zaveri, A. Elshahat, G. Hitawala, E.Y. Wang, R. Kitagawa, A. Pathak, F. Scalzo, M. Ihara, K.S. Sunnerhagen, M.R. Walters, Y. Zhao, and N. Jette report no disclosures relevant to the manuscript. S.E. Kasner has received grant funding from Bayer, Bristol-Myers Squibb, Daiichi Sankyo, DiaMedica, WL Gore, and Stryker, and royalties from UpToDate. P. Kwan and N.K. Mishra report no disclosures relevant to the manuscript. N.K. Mishra and H.H. Altalib are employees of the US Department of Veterans Affairs; any opinions, findings, and conclusions or recommendations expressed in this material are their own and do not necessarily reflect the views of the US Department of Veterans Affairs. Go to Neurology.org/N for full disclosures.