Real‐world comparison of first‐line antiseizure monotherapy and the role of age at treatment initiation in newly diagnosed childhood epilepsy: A cohort study from a tertiary center

Abstract

Objective

This study was undertaken to evaluate the real‐world effectiveness and tolerability of first‐line antiseizure medication (ASM) monotherapy in children with newly diagnosed epilepsy, focusing on comparative outcomes across developmental age groups and ASM types, and identifying clinical risk factors of treatment failure.

Methods

This retrospective cohort study analyzed 10 years of electronic medical records from Shenzhen Children's Hospital. Children aged ≤18 years who initiated monotherapy with one of five ASMs (oxcarbazepine, valproate, levetiracetam, topiramate, or lamotrigine) were included. The primary outcome was treatment success, defined as sustained seizure and side effect freedom at the defined analytic endpoint. Outcomes based on the 2010 International League Against Epilepsy (ILAE) definition were also evaluated. Nineteen clinical variables across six domains were analyzed. Multivariable Cox regression and stabilized inverse probability of treatment weighting were used to adjust for confounding.

Results

Among 7060 eligible patients with a median follow‐up of 14.0 months, 47.9% achieved treatment success, and 48.3% met ILAE criteria. The median seizure‐ and side effect‐free survival was 24.8 months (95% confidence interval [CI] = 23.3–26.5 months). Oxcarbazepine (32.3%), valproate (31.7%), and levetiracetam (22.5%) were the most commonly prescribed ASMs. Treatment success did not differ significantly across ASM types or age groups. Epileptic spasms (hazard ratio [HR] = 2.86, 95% CI = 1.98–4.09), generalized seizures (HR = 1.78, 95% CI = 1.49–2.13), seizures before reaching maintenance dose (HR = 1.52, 95% CI = 1.37–1.70), delayed treatment initiation (HR ~ 1.30), and neurodevelopmental delay (HR = 1.42, 95% CI = 1.26–1.60) were significant risk factors of treatment failure.

Significance

First‐line ASMs showed similar effectiveness in routine pediatric care. Childhood epilepsy treatment outcomes were more strongly influenced by seizure characteristics, neurodevelopmental comorbidities, and timing of initiation than by ASM choice or age. These findings support extrapolating ASM efficacy across pediatric age groups and highlight the value of real‐world evidence in treatment decision‐making.

Key points.

• First‐line antiseizure medication (oxcarbazepine, valproate, levetiracetam, topiramate, and lamotrigine) monotherapy showed comparable real‐world effectiveness in newly diagnosed childhood epilepsy.

• Seizure characteristics, neurodevelopmental comorbidities, and treatment delay were stronger predictive risk factors of treatment failure than drug choice or age.

• Findings support extrapolating ASM efficacy across pediatric age groups based on real‐world outcomes.

1. INTRODUCTION

Childhood epilepsy is a common neurological disorder with significant consequences for neurodevelopment and long‐term quality of life. ¹ , ² Despite advancements in antiseizure medication (ASM), achieving sustained seizure control in newly diagnosed children remains challenging. ³ , ⁴ Whereas prior studies report monotherapy success rates of 47%–50.5% in mixed‐age populations of adults and children ≥9 years old, ⁵ , ⁶ data focusing specifically on younger pediatric subgroups remain limited and undercharacterized. ⁷ , ⁸ , ⁹

Current clinical guidelines recommend several ASMs, including oxcarbazepine (OXC), valproate (VPA), and levetiracetam (LEV), as first‐line monotherapies for pediatric epilepsy. ¹⁰ , ¹¹ , ¹² The choice of ASMs is driven largely by seizure type; OXC is favored for focal seizures, whereas LEV may be preferred when an SCN1A mutation is suspected or seizures are provoked by fever. ¹² Ultimately, clinicians make the choice after weighing drug availability, the child's clinical profile, and parents' understanding of safety of each medication. However, comparative effectiveness across developmental stages is not well characterized. ¹³ Most available evidence comes from randomized controlled trials with strict eligibility criteria, limiting their generalizability to real‐world practice.

A major barrier in pediatric epilepsy care is the delayed approval of ASMs for children. ¹⁴ New ASMs are typically tested and approved for adults first, then gradually extended to pediatric populations, usually starting with older children (e.g., ≥12 years old), followed by younger age groups (e.g., 6 years, then 4 years). The stepwise approval process often leads to delays due to the limited availability of pediatric trial data, ¹⁴ , ¹⁵ which reflects both ethical and practical challenges in conducting pediatric clinical trials, including difficulties in recruitment, consent, and endpoint determination. ¹⁶ Consequently, real‐world evidence is essential for bridging the evidence gap and guiding both clinical practice and regulatory decision‐making. ¹⁷ , ¹⁸

This study aims to evaluate the real‐world effectiveness of five first‐line ASMs—OXC, VPA, LEV, topiramate (TPM), and lamotrigine (LTG)—in children with newly diagnosed epilepsy. We examine treatment outcomes by age and ASM type following initial monotherapy, as well as clinical risk factors for treatment failure. Based on these real‐world findings, we further explore their implications for pediatric regulatory extrapolation and clinical treatment guidance.

2. MATERIALS AND METHODS

2.1. Study design

This retrospective cohort study aimed to evaluate the real‐world effectiveness of initial ASM monotherapy in children newly diagnosed with epilepsy. The deidentified electronic medical records (EMRs) were obtained from the Shenzhen Children's Hospital Epilepsy Specialty Data Platform, which archives both outpatient/emergency and inpatient records from the Department of Pediatric Neurology at Shenzhen Children's Hospital. This study was approved by the ethics committee of Shenzhen Children's Hospital (approval No. 2023061) and registered with the Chinese Clinical Trial Registry (ChiCTR2300075115). Reporting follows the STROBE guidelines. ¹⁹

2.1.1. Inclusion criteria

The study period spanned from February 1, 2014 to March 31, 2023. Patients were eligible if they met all six criteria:

1. Diagnosis of epilepsy according to the International League Against Epilepsy (ILAE) 2014 operational clinical definition. ²⁰

2. First‐time ASM monotherapy initiated with VPA, OXC, LEV, TPM, or LTG.

3. No prior exposure to any ASMs (regardless of duration), including traditional Chinese medicines. Emergency use of injectable ASMs or acute rescue medications (e.g., diazepam, phenobarbital, or midazolam) was not considered prior exposure.

4. Age ≤18 years at treatment initiation.

5. Evidence of long‐term oral ASM use, defined as treatment duration ≥15 consecutive days and at least two dispensations of the same ASM (initial fill plus at least one refill, excluding emergency injections).

6. At least 1 year of follow‐up data after treatment initiation.

2.1.2. Exclusion criteria

We defined six exclusion criteria and assessed each independently; patients meeting at least one criterion were excluded.

1. Poor medication adherence, defined as documented missed doses or treatment discontinuation within the first 3 months of treatment.

2. Discontinuation, switching, or addition of medications before achieving steady‐state maintenance dosing (defined as reaching the maintenance dose plus five drug half‐lives, up to a maximum of 90 days). Specifically, the five drug half‐lives for each ASM were: VPA, 3.125 days; OXC, 5.208 days; LEV, 1.667 days; TPM, 6.25 days; and LTG, 6.25 days. ¹²

3. Fewer than two dispensations of the same ASM within the first year after treatment initiation.

4. Absence of follow‐up data within the first year after treatment initiation.

5. Inappropriate ASM monotherapy according to clinical guidelines, such as OXC in contraindicated settings. ¹²

6. ASM dosages >10 times or <1% of the recommended therapeutic dose.

2.2. Endpoint and outcome definition

The primary outcome was treatment success or failure of initial ASM monotherapy. Treatment success was defined as sustained freedom from both seizures and drug‐related side effects after the drug has reached steady‐state. ²¹ We designated a comprehensive set of events covering all plausible clinical scenarios as the analytic end‐of‐follow‐up for this study (for details, see Analytic Endpoint Definition section in Supplementary Methods). Treatment success was defined operationally as:

1. Medication discontinuation or dose reduction for well‐controlled seizures.

2. No seizures, medication changes, or dose escalations through the final dispensation.

3. Nonadherence, including missed doses or self‐initiated discontinuation of medication. This is treated as a success endpoint regardless of any subsequent seizures or those explicitly reported during the nonadherence period, as such seizures could be presumed to result from nonadherence rather than drug inefficacy.

Treatment failure comprised:

1. Seizure recurrence after steady‐state despite adherence.

2. Discontinuation, switching, or add‐on of an ASM due to poor seizure control or side effects (e.g., rash, hepatotoxicity).

Additionally, treatment outcomes were evaluated using the 2010 ILAE “rule of three” definition (hereafter referred to as the ILAE outcome). Patients who remained seizure‐ and side effect‐free at the time of final dispensation but whose follow‐up ended before the required observation window were classified as “unknown” under the ILAE outcome criteria.

2.3. Definition of variables

Nineteen variables grouped into six categories were included.

1. Demographic characteristics

This category included three variables: sex, age at treatment initiation, and age group. Age groups followed a standardized developmental classification, ²² including:

• Term neonatal: birth to 27 days

• Infancy: 28 days–12 months

• Toddler: 13 months–2 years

• Early childhood: 2–5 years

• Middle childhood: 6–11 years

• Early adolescence: 12–18 years

• 2Seizure characteristics

This category included four variables:

• Seizure type

• Accompanied nonepileptic seizures

• Development of status epilepticus

• Seizures before reaching steady‐state ASM dosage

Seizure types were classified into four major groups based on clinical guidelines: generalized seizures, focal seizures, epileptic spasms, and reflex seizures. ¹² If no definitive classification was available, the seizure type was marked as “uncertain.” As individual patients could present with multiple seizure types, each type was treated as a separate categorical variable with three possible values: yes, no, or uncertain.

• 3Seizure history

This category included two variables: total number of seizure days prior to treatment and interval from first seizure to treatment initiation, which was further categorized into three groups: ≤3 months, 3 months–1 year, and >1 year.

• 4ASM and dosage.

This category included two variables: type of ASM prescribed in monotherapy and average maintenance daily dose during the 90 days prior to outcome, normalized by body weight and dispensation duration (mg/[kg·d]). To enhance cross‐drug comparability, a relative dose was calculated as the ratio of the patient's maintenance dose to the guideline‐recommended values: VPA, OXC, LEV: 20 mg/(kg·d); LTG: 2 mg/(kg·d); TPM: 3 mg/(kg·d). ¹²

• 5Comorbidities

This category included three binary variables related to neurodevelopmental and psychiatric comorbidities: neurodevelopmental delay, cerebral palsy, and autism spectrum disorder.

• 6Seizure‐related personal and family history

This category included history of febrile seizures and family history of seizures or prior ASM treatment.

2.4. Data sources and measurements

Study variables were extracted from structured and unstructured EMRs, including demographic information, visit records, diagnostic information, pharmacy dispensation records, outpatient and emergency records, inpatient records, and vital signs. A custom natural language processing (NLP) and data extraction pipeline was developed using Python‐based regular expression functions to clean data and extract the relevant variables. Details on the EMRs and the NLP‐based data cleaning process are provided in the EMR Data Cleaning by NLP section of Supplementary Methods.

To reflect baseline characteristics at treatment initiation, all variables were defined based on information recorded before or within 90 days after ASM initiation. Febrile seizure history was coded as “yes” only if documented prior to treatment.

To ensure data completeness and standardization, the following assumptions were applied when explicit documentation was absent:

• If seizure type could not be definitively classified, all seizure type‐related variables were coded as “uncertain.”

• Absence of documentation for nonepileptic seizures, status epilepticus, and comorbid neurodevelopmental or psychiatric conditions was interpreted as absence of those conditions.

• If no pretreatment seizure history was recorded, the number of pretreatment seizure days was set to two, and the interval from first seizure to treatment initiation was categorized as ≤3 months.

The rationale for these assumptions is detailed in the Rationale for Assumptions Applied to Incomplete EMR Documentation section in Supplementary Methods.

2.5. Bias

To minimize potential sources of bias, several measures were taken. First, all data extraction and NLP procedures were conducted blinded to treatment outcomes, ensuring that variable definition were not influenced by treatment outcome. Second, no additional selection or filtering was applied, and only patients meeting predefined inclusion and exclusion criteria were analyzed, thereby avoiding post hoc selection. Third, we ensured high data quality in NLP‐based data cleaning to minimize bias. We actively refined the NLP rules and keyword dictionary through repeated testing, updating them whenever extraction errors or inconsistencies were identified, with each adjustment informed by manual review of the source text. Based on our internal validation, the NLP pipeline achieved >90% accuracy across all extracted elements. For details, please refer to the EMR Data Cleaning by NLP section of Supplementary Methods and Table S1.

2.6. Study size

This being a retrospective cohort study, sample size was determined by the number of eligible cases available in the EMRs over the study period. All patients who met the predefined inclusion and exclusion criteria were included without additional sampling or matching. No formal power calculation was performed prior to analysis.

2.7. Quantitative variables

Only two quantitative variables were included: age at treatment initiation and the interval from first seizure to ASM initiation. Age was categorized into standardized developmental stages (see Section 2.3) to enable age group comparisons. The seizure‐to‐treatment interval, originally measured in days, was categorized into three clinically meaningful groups: ≤3 months, 3 months–1 year, and >1 year, improving interpretability and robustness to outliers in statistical modeling.

2.8. Statistical methods

We conducted analyses to examine differences in treatment success across age groups and ASM types, and to identify risk factors associated with treatment failure. For these analyses only, patients in the neonatal group (n = 12) and variable reflex seizures group (only one patient) were excluded due to insufficient sample sizes. However, we still report patient characteristics and outcome summaries for the full cohort, including the neonatal group. In addition, we use age groups instead of continuous age values in these analyses.

2.8.1. Treatment outcome by age group

We applied a multivariable Cox proportional hazards model, ²³ with treatment failure as the event and months to outcome as the time variable. All 16 covariates were included in the model to adjust for confounding. We first used likelihood‐ratio test (LRT) ²⁴ to assess the overall effect of age group on treatment outcome by comparing models with and without the age group variable. Subsequently, age groups were included in the Cox model as categorical variables, with the youngest age group serving as the reference category, and we performed pairwise comparisons using Tukey range test ²⁵ based on constructing contrasts on the developed Cox model.

2.8.2. Treatment outcome by ASM

Due to nonrandomized ASM assignment and imbalanced group sizes, we applied a propensity score‐based adjustment using stabilized inverse probability of treatment weighting (IPTW) ²⁶ to account for confounding and treatment selection bias. A multinomial logistic regression model was fitted using all 16 covariates to estimate each patient's probability of receiving their observed treatment, from which IPTW weights were derived. To mitigate the influence of extreme weights and improve the stability of the estimates, weights were truncated at a threshold of 15. Then, pairwise comparisons of treatment outcomes across ASM groups were performed using weighted log‐rank tests. ²⁷ To account for multiple testing, we controlled the false discovery rate at 5% (α = .05) using the Benjamini–Hochberg (BH) procedure. ²⁸ Pairwise comparisons with BH‐adjusted p‐values (i.e., q‐values ²⁹ ) < .05 were considered statistically significant.

2.8.3. Risk factor screening

We screened predictive risk factors for treatment failure with an adaptive least absolute shrinkage and selection operator (LASSO) penalized Cox model. ³⁰ , ³¹ A Tikhonov penalized ³² Cox model was first fitted to all 13 candidate factors plus the four prespecified covariates (sex, age group, ASM, and relative dose), and the resulting ridge estimates were used as adaptive weights for the subsequent adaptive LASSO penalized Cox model fit. For both the ridge pilot and the adaptive LASSO, the penalty parameter λ was chosen by 10‐fold cross‐validation on the partial likelihood deviance, and the value that minimized the mean deviance was selected. To correct for selection optimism, ³³ we applied the exact postselection inference procedure proposed by Lee et al., ³⁴ which was recently shown to perform well on adaptive LASSO in the benchmark study by Kammer et al. ³⁵ We report 95% selective confidence intervals (CIs) and one‐sided p‐values conditional on the chosen active variable set and coefficient signs; candidate factors with a selective p < .05 are regarded as statistically significant risk factors.

2.8.4. Additional analytical considerations

This study did not involve formal subgroup or interaction analyses, as the primary aim was to evaluate overall treatment success rates and identify risk factors across age groups and ASM types. Missing data were handled based on rational assumptions during the data preprocessing stage, resulting in a complete analytical dataset. Loss to follow‐up was addressed by using the most recent visit or dispensation date to define outcome status. No sensitivity analyses were conducted.

For descriptive statistics and baseline comparisons, continuous variables were tested using Wilcoxon rank‐sum test, ³⁶ whereas categorical variables were tested using chi‐squared test. ³⁷ Unless otherwise specified, all statistical tests were two‐sided, and statistical significance was defined as p < .05. When accounting for multiple testing, hypotheses with Tukey range test or BH procedure adjusted p‐values < .05 were considered statistically significant.

2.8.5. Statistical software and packages

All statistical analyses were conducted using R software (version 4.3.0). ³⁸ Cox models and Kaplan–Meier curves were constructed using the survival package (version 3.8.3). ³⁹ Multinomial logistic regression was implemented using the nnet package (version 7.3.18). ⁴⁰ Tikhonov and adaptive LASSO penalized Cox models were fitted using the glmnet package (version 4.1–8), ⁴¹ , ⁴² and exact postselection inference was conducted using the selectiveInference package (version 1.2.5). ⁴³ Pairwise comparisons with Tukey range test were performed using the glht function from the multcomp package (version 1.4.28). ⁴⁴ Heatmaps were produced with the ggplot2 package (version 3.5.0) ⁴⁵ and forest plots with the forestplot package (version 3.1.6). ⁴⁶

3. RESULTS

3.1. Characteristic of all patients

We identified 10 188 patients with epilepsy‐ or seizure‐related encounters on the data platform. Of these, 7974 patients met the inclusion criteria, and after exclusions, 7060 patients were included in analysis (Figure 1). Baseline characteristics are summarized in Table 1. Among the cohort, 41.4% were females, and mean age at treatment initiation was 5.77 years. There were marked differences in the number of patients across age groups, with the three largest groups being middle childhood (6–11 years, 36.8%), early childhood (2–5 years, 29.9%), and infancy (28 days–12 months, 14.3%). Substantial variation in ASM usage was also observed, and OXC (32.3%), VPA (31.7%), and LEV (22.5%) were most commonly prescribed. The mean relative dose before outcome was 1.04, indicating most patients received a dosage close to the recommended level.

FIGURE 1.

Flowchart of patient selection based on inclusion and exclusion criteria. ASM, antiseizure medication.

TABLE 1.

Characteristics of patients.

Characteristic	Overall, N = 7060
Female	2921 (41.4%)
Age at treatment initiation, years	5.77 ± 4.10
Age group at treatment initiation
Neonatal	12 (.2%)
Infancy	1012 (14.3%)
Toddler	717 (10.2%)
Early childhood	2110 (29.9%)
Middle childhood	2597 (36.8%)
Early adolescence	612 (8.7%)
Antiseizure medication
Valproate	2237 (31.7%)
Oxcarbazepine	2280 (32.3%)
Levetiracetam	1592 (22.5%)
Topiramate	555 (7.9%)
Lamotrigine	396 (5.6%)
Relative dose	1.04 ± .48
Generalized seizure
Yes	208 (2.9%)
No	1325 (18.8%)
Uncertain	5527 (78.3%)
Focal seizure
Yes	1155 (16.4%)
No	378 (5.4%)
Uncertain	5527 (78.3%)
Epileptic spasms
Yes	38 (.5%)
No	1495 (21.2%)
Uncertain	5527 (78.3%)
Reflex seizure
Yes	1 (0%)
No	1532 (21.7%)
Uncertain	5527 (78.3%)
Developed status epilepticus	176 (2.5%)
Accompanied by nonepileptic seizures	163 (2.3%)
Seizure days prior to treatment	2.53 ± 1.62
Interval from seizure onset to treatment
≤3 months	5118 (72.5%)
3 months–1 year	800 (11.3%)
>1 year	1142 (16.2%)
Seizure occurrence before reaching maintenance dose stability	576 (8.2%)
Neurodevelopmental delay	489 (6.9%)
Cerebral palsy	56 (.8%)
Autism spectrum disorder	32 (.5%)
History of febrile seizures	292 (4.1%)
Family history	531 (7.5%)

Note: For continuous variables, statistics are reported as mean ± SD. For categorical variables, statistics are reported as the frequency of each category, with the percentage in parentheses indicating the proportion of that category among all 7060 patients.

Of the 1533 patients (21.7%) with clearly defined seizure types, focal seizures were most common (75.3%), followed by generalized seizures (13.6%), epileptic spasms (2.5%), and reflex seizures (<.1%). Sixty‐one patients exhibited two seizure types, and one patient exhibited three. Status epilepticus occurred in 2.5% of patients, and nonepileptic seizures were documented in 2.3% of patients. The average number of seizure days prior to treatment was 2.53, and 72.5% of patients initiated ASM treatment within 3 months of their first seizure onset; 8.2% of patients experienced seizures during the titration period after treatment initiation.

Regarding comorbidities, 6.9% had neurodevelopmental delay, .8% had cerebral palsy, and .5% had autism spectrum disorder. Febrile seizure history was present in 4.1% of patients, and 7.5% reported a family history of seizures or prior ASM treatment.

The median follow‐up duration was 14.00 months, with a maximum of 9.23 years.

3.2. Outcomes of all patients

At the analytic endpoint, 3677 patients experienced treatment failure due to uncontrolled seizures or drug side effects, yielding an overall failure rate of 52.1%. Among the 3383 patients (47.9%) who achieved treatment success, the majority (3212 patients, 94.9%) remained seizure‐free through their last dispensation. In addition, 38 patients achieved success due to dose reduction following sustained seizure control, and 56 and 77 patients were classified as successes at early analytic end of follow‐up due to missed doses or self‐initiated discontinuation, respectively. The Kaplan–Meier curve for all patients is shown in Figure S1. The median seizure‐ and side effect‐free survival time was 24.8 months (95% CI = 23.3–26.5 months). According to the ILAE outcome definition, 3412 patients (48.3%) met the criteria for treatment success and 2778 (39.3%) for failure, and 870 (12.3%) were classified as unknown due to insufficient follow‐up duration.

3.3. No significant differences in treatment success across age groups

Treatment outcomes by age groups based on endpoint and ILAE criteria, along with median seizure‐ and side effect‐free survival times, are summarized in Table 2. Kaplan–Meier curves for each age group are shown in Figure 2A. An LRT comparing models with and without the age group variable indicated a statistically significant association between age group and treatment outcome (LRT p = .032). However, the trend in treatment outcomes across age groups did not follow a strictly monotonic pattern. Unadjusted results indicated better outcomes in toddlers (13 months–2 years) compared to infancy (28 days–12 months), with declines in the early childhood (2–5 years) and middle childhood groups (6–11 years). Notably, early adolescents (12–18 years) had the highest overall success, with 54.1% endpoint success, 55.6% ILAE success, and a median event‐free survival of 33.1 months (95% CI = 28.0–42.0 months). These findings suggest a nonlinear age‐related response to treatment.

TABLE 2.

Treatment outcomes by age group.

Outcome	Infancy, n = 1012	Toddler, n = 717	Early childhood, n = 2110	Middle childhood, n = 2597	Early adolescence, n = 612
Endpoint success	506 (50.0%)	376 (52.4%)	896 (42.5%)	1268 (48.8%)	331 (54.1%)
ILAE outcome
Success	491 (48.5%)	360 (50.2%)	927 (43.9%)	1291 (49.7%)	340 (55.6%)
Unknown	103 (10.2%)	76 (10.6%)	260 (12.3%)	353 (13.6%)	74 (12.1%)
Median event‐free survival, months	25.3 (19.7–31.0)	29.7 (18.5–50.1)	18.3 (15.6–21.4)	27.0 (24.5–30.4)	33.1 (28.0–42.0)

Note: Median seizure‐ and side effect‐free survival times are reported along with their corresponding 95% confidence intervals.

Abbreviation: ILAE, International League Against Epilepsy.

FIGURE 2.

Treatment outcomes by age group. (A) Kaplan–Meier curves showing seizure‐ and side effect‐free survival stratified by age group. (B) Heatmap of pairwise hazard ratios (HRs) with 95% confidence intervals for treatment failure across age groups. Each cell represents the HR comparing the age group on the x‐axis to the reference group on the y‐axis. The p‐values were adjusted (adj.) using Tukey range test and are presented in the third row in each cell. adol., adolescence; child., childhood.

Furthermore, the adjusted results revealed a similar nonlinear pattern in treatment response across age groups. Based on the multivariable Cox model that included all covariates to account for potential confounding, outcomes improved from infancy to toddlers but declined thereafter in a near‐monotonic fashion through early and middle childhood, reaching the lowest point in early adolescence. Notably, toddlers showed the most favorable treatment outcomes, whereas both infancy and early adolescence were associated with higher risks of treatment failure compared to the intermediate age groups, highlighting the complexity of age‐related treatment responses even after adjustment. Despite these trends, pairwise comparisons using Tukey range test revealed no statistically significant differences between any two age groups after adjustment for multiple comparisons (Figure 2B).

3.4. No significant differences in treatment success across the five ASMs

Treatment outcomes by the five ASMs based on endpoint and ILAE criteria, along with median seizure‐ and side effect‐free survival times, are summarized in Table 3, including both unadjusted and propensity score‐adjusted results. Unadjusted and adjusted Kaplan–Meier curves for each ASM are shown in Figure S2. Based on unadjusted results, TPM had the best treatment success by endpoint (57.8%) and the longest median event‐free survival (38.5 months, 95% CI = 30.9–57.8 months). In contrast, LEV had the lowest outcomes across all three unadjusted metrics (44.7% endpoint success, 40.6% ILAE success, 16.2 months survival with 95% CI = 13.8–20.0 months). However, after propensity score adjustment, the ranking shifted substantially. LTG demonstrated the best outcomes, with 65.1% endpoint success, 53.0% ILAE success, and the longest median event‐free survival (58.4 months, lower 95% confidence bound = 39.4 months). Conversely, TPM, despite its high unadjusted performance, had the shortest adjusted survival time (17.5 months, 95% CI = 12.6–37.3 months). Despite these differences in ranking, pairwise comparisons using weighted log‐rank test revealed no statistically significant differences between any two ASMs after adjustment for multiple comparisons (all q‐values > .05).

TABLE 3.

Treatment outcomes by antiseizure medication.

Outcome	OXC	VPA	LEV	TPM	LTG
Unadjusted
Sample size	2279	2236	1582	555	396
Endpoint success	1041 (45.7%)	1113 (49.8%)	707 (44.7%)	321 (57.8%)	195 (49.2%)
ILAE outcome
Success	1126 (49.4%)	1158 (51.8%)	643 (40.6%)	277 (49.9%)	205 (51.8%)
Unknown	251 (11.0%)	234 (10.5%)	232 (14.7%)	101 (18.2%)	48 (12.1%)
Median event‐free survival, months	24.1 (22.1–26.4)	29.7 (26.2–31.8)	16.2 (13.8–20.0)	38.5 (30.9–57.8)	30.9 (18.8–39.7)
Propensity score‐adjusted
Sample size	2235	2104	1664	717	433
Endpoint success	1089 (48.7%)	1042 (49.5%)	915 (55.0%)	373 (52.1%)	282 (65.1%)
ILAE outcome
Success	1131 (50.6%)	1080 (51.3%)	780 (46.9%)	305 (42.5%)	229 (53.0%)
Unknown	260 (11.6%)	234 (11.1%)	288 (17.3%)	145 (20.3%)	88 (20.2%)
Median event‐free survival, months	25.7 (23.1–29.0)	27.4 (23.6–30.5)	37.6 (25.4–52.5)	17.5 (12.6–37.3)	58.4 (39.4–NA)

Note: The neonatal age group (n = 12) was excluded from calculations in the table. Median seizure‐ and side effect‐free survival times are reported along with their corresponding 95% CIs. “NA” indicates that either the upper or lower bound of the CI could not be estimated due to insufficient sample.

Abbreviations: CI, confidence interval; ILAE, International League Against Epilepsy; LEV, levetiracetam; LTG, lamotrigine; OXC, oxcarbazepine; TPM, topiramate; VPA, valproate.

3.5. Risk factor screening

Cross‐validation identified the optimal penalty parameter as λ = .0015 for the ridge pilot (Figure S3A) and λ = .0023 for the subsequent adaptive LASSO fit (Figure S3B). The adaptive LASSO penalized Cox model retained all four prespecified covariates (sex, age group, ASM, and relative dose) together with 11 of the 13 screened factors, and seven of these factors met the significance criterion (selective p < .05, Figure 3), including the following:

• Four seizure characteristics variables: generalized seizures, epileptic spasms, status epilepticus, and seizures before reaching maintenance dose stability.

• Two seizure history variables: number of seizure days prior to treatment and interval from seizure onset to treatment.

• One comorbidity: neurodevelopmental delay.

FIGURE 3.

Selective hazard ratios (HRs) with 95% selective confidence intervals (CIs) and one‐sided p‐values for the 11 candidate factors retained by the adaptive least absolute shrinkage and selection operator (LASSO) penalized Cox proportional hazards model in risk factor screening. The four prespecified covariates (sex, age group, antiseizure medication, and relative dose) were also retained but are omitted from the figure for clarity. All estimates and intervals were obtained with the exact postselection inference procedure applied to the adaptive LASSO fit. M, months; Y, year.

Among these, the strongest risk factor was epileptic spasm (hazard ratio [HR] = 2.86, 95% CI = 1.98–4.09), followed by generalized seizure (HR = 1.78, 95% CI = 1.49–2.13). Seizures before reaching maintenance dosing were also predictive of treatment failure (HR = 1.52, 95% CI = 1.37–1.70), indicating poor early seizure control. A longer interval between seizure onset and treatment initiation was associated with worse outcomes, with an HR of approximately 1.30 for intervals of >3 months compared to those of ≤3 months. The number of seizure days before treatment showed a weak effect (HR = 1.05, 95% CI = 1.02–1.07). Neurodevelopmental delay (HR = 1.42, 95% CI = 1.26–1.60) was also a significant comorbid risk factor.

4. DISCUSSION

This retrospective cohort study evaluated the real‐world effectiveness of five commonly used ASMs in children (≤18 years old) with newly diagnosed epilepsy, focusing on age‐ and ASM‐specific treatment outcomes and risk factors for treatment failure. Among 7060 patients included in analysis, the overall treatment success rate—defined as freedom from both seizures and drug‐related side effects—was approximately 47.9% based on the study endpoint definition and 48.3% according to ILAE criteria, with a median seizure‐ and side effect‐free survival time of 24.8 months (95% CI = 23.3–26.5 months). These findings highlight the long‐term effectiveness and clinical benefit of initial ASM monotherapy in nearly half of the pediatric patients. OXC, VPA, and LEV were the most frequently prescribed ASMs, consistent with current clinical guidelines. ¹¹ , ¹²

Unadjusted comparisons by age group and ASM type differed markedly from the fully adjusted results because risk factor profiles were uneven across groups. Early adolescents, for instance, had no epileptic spasms, almost no status epilepticus, and fewer pretitration seizures than younger children (Table S2). These advantages inflated their crude success rate, but after adjusting for these factors the underlying age effect emerged; early adolescents actually showed a slightly higher risk of failure. A similar pattern explains the shift seen across ASM types (Table S3). However, the absence of significant differences in treatment success across both age groups and ASM types after adjustment suggests that treatment outcomes were more strongly influenced by seizure characteristics, neurodevelopmental comorbidities, and timing of initiation than by ASM choice or age. ⁴⁷ , ⁴⁸ , ⁴⁹

These findings align with recent large‐scale cohort studies, which show that nonsyndromic epilepsy tends to follow a predictable therapeutic trajectory when appropriately managed with first‐line ASMs. ⁵ , ⁶ Our findings also support the hypothesis that seizure control mechanisms are relatively conserved across developmental stages, consistent with the meta‐analysis by Janmohamed et al. ⁵⁰ and other studies that report comparable first‐treatment success rates across pediatric and adult populations. ⁵ , ⁶ , ⁵¹ The less frequent use of TPM and LTG in our cohort may reflect clinical concerns about cognitive side effects or titration difficulties, particularly in younger children, highlighting the importance of balancing efficacy with tolerability and dosing practicality.

Key risk factors of treatment failure included epileptic spasms, generalized seizures, status epilepticus, and early seizures before achieving maintenance dosing. These findings emphasize the need for individualized, more aggressive strategies in patients with severe or atypical seizure types. Neurodevelopmental delay was also associated with poorer outcomes, illustrating the complex relationship between epilepsy and neurodevelopment. Additionally, a longer interval between seizure onset and ASM initiation was linked to higher failure risk, reinforcing the importance of early diagnosis and timely intervention.

The challenges in pediatric ASM development underscore the critical role of real‐world data in bridging evidence gaps. Our findings support current US and EU regulatory frameworks that permit efficacy extrapolation from older children to younger populations (≥1 month) when supported by robust pharmacokinetic and safety data. ¹⁵ , ⁵² This approach balances ethical considerations with the need for timely access to effective treatments.

This study has several limitations. First, its retrospective design introduces potential selection bias, and reliance on data from a single tertiary center may limit the generalizability of findings to broader clinical settings. Second, heterogeneity in EMR documentation and the resulting missing data may have affected the precision of analyses. Despite our accurate NLP pipeline (>90% overall accuracy), multiple quality checks, and the use of clinically grounded assumptions to handle undocumented items, residual misclassification and hence potential bias cannot be fully excluded. Third, our dataset lacked multimodal clinical biomarkers, including genetic testing, electroencephalographic features, and neuroimaging data, so we did not have systematic data on epilepsy etiology or well‐defined epilepsy syndromes. This gap restricts interpretation of the biological mechanisms driving treatment response. Incorporating these data in future work would permit etiological and syndrome‐specific analyses of ASM effectiveness. Fourth, our analysis evaluated only the five first‐line ASMs most commonly used at our center, and does not cover newer ASMs (e.g., cenobamate, lacosamide, perampanel) due to insufficient sample size; larger datasets will be needed to assess these drugs. Finally, although the median follow‐up duration was 14 months, the maximum follow‐up extended up to 10 years, introducing potential survivorship bias in long‐term outcome assessments.

Despite these limitations, this study provides valuable real‐world insights from a large, longitudinal pediatric cohort, spanning a broad developmental age range and seizure characteristics. The use of ILAE‐standardized outcome and adjusted survival models strengthens the methodological rigor and interpretability of results. Future prospective multicenter studies incorporating genetic, neurophysiological, and imaging data are needed to validate the risk factors and to advance precision medicine approaches for childhood epilepsy.

5. CONCLUSIONS

In this large, retrospective real‐world study of children with newly diagnosed epilepsy, treatment success following initial ASM monotherapy was influenced more by seizure characteristics, comorbidities, and treatment timing than by ASM choice or age at treatment initiation alone. These findings highlight the importance of early intervention and individualized risk assessment in pediatric epilepsy management and provide valuable real‐world evidence to inform clinical decision‐making and support regulatory policies for ASM use in children.

AUTHOR CONTRIBUTIONS

Ningshan Li and Jianxiang Liao contributed to study design. Huiting Zhang, Xiaoying Qiao, Dezhi Cao, and Lina Men contributed to clinical data management and data quality control. Ningshan Li and Chaoran Yang contributed to data extraction and cleaning. Ningshan Li contributed to data analysis. Ningshan Li and Jianxiang Liao contributed to drafting of the manuscript. Guangjun Yu and Jianxiang Liao supervised the study. All authors reviewed and approved the final manuscript.

FUNDING INFORMATION

This research was funded by the Shenzhen Science and Technology Program, the Sanming Project of Medicine in Shenzhen (No. SZSM202311028), the Shenzhen Key Medical Discipline Construction Fund (No. SZXK033), the Shenzhen Fund for Guangdong Provincial High‐Level Clinical Key Specialties (No. SZGSP012), the Shenzhen Fund (No. JCYJ20200109150818777), and the Shenzhen and Guangdong High‐Level Hospital Construction Fund (No. KCXFZ20211020163549011).

CONFLICT OF INTEREST STATEMENT

None of the authors has any conflict of interest to disclose. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.

PATIENT CONSENT STATEMENT

The requirement for written informed consent from children or their guardians was waived due to the retrospective nature of the study and the use of deidentified data.

CLINICAL TRIAL REGISTRATION

This study protocol was registered with the Chinese Clinical Trial Registry (ChiCTR2300075115) in accordance with the requirements for observational study transparency.

Supporting information

Appendix S1.

Li N, Zhang H, Yang C, Qiao X, Cao D, Men L, et al. Real‐world comparison of first‐line antiseizure monotherapy and the role of age at treatment initiation in newly diagnosed childhood epilepsy: A cohort study from a tertiary center. Epilepsia. 2025;66:4738–4751. 10.1111/epi.18607

DATA AVAILABILITY STATEMENT

The datasets generated and analyzed during the current study are available from the corresponding author on reasonable request.