A population-based cohort study published in The BMJ reports that prenatal exposure to valproate and zonisamide carries the most substantial associations with neurodevelopmental disorders in children, while levetiracetam and phenytoin demonstrated no statistically significant increased risk across the outcomes evaluated. The findings, drawn from healthcare utilization data spanning two decades in the United States, offer the most comprehensive comparative assessment to date of neurodevelopmental risk across ten antiseizure medications (ASMs) used during pregnancy.

Study Design and Population

The cohort study drew on healthcare claims data from both publicly and commercially insured beneficiaries in the United States between 2000 and 2021. Investigators linked records of pregnant patients with epilepsy to their offspring, establishing a study population that enabled longitudinal follow-up of children from birth through the period during which neurodevelopmental diagnoses typically emerge.

The exposure window was defined as the second half of pregnancy, a period corresponding to active synaptogenesis in the developing fetal brain. This design decision reflects current understanding of when the central nervous system is most vulnerable to pharmacological disruption of neural circuit formation. The ten ASMs examined were carbamazepine, lacosamide, lamotrigine, levetiracetam, oxcarbazepine, phenobarbital, phenytoin, topiramate, valproate, and zonisamide.

The reference group consisted of pregnant patients with a diagnosed epilepsy condition who received no ASM dispensation from three months prior to conception through delivery. This comparison group, rather than the general obstetric population, was selected to reduce confounding by indication, a methodological concern of considerable weight in pharmacoepidemiology research. Epilepsy itself, independent of medication exposure, may confer neurological risk to offspring, and investigators sought to isolate the contribution of specific drugs by holding underlying diagnosis constant across exposed and unexposed groups.

The cohort comprised 8,887 prenatally unexposed children. Among exposed pregnancies, sample sizes varied substantially by drug: lacosamide represented the smallest exposure group at 219 pregnancies, while levetiracetam represented the largest at 5,261.

Statistical Methods and Confounding Adjustment

Investigators estimated hazard ratios (HRs) using Cox proportional hazard models. To address potential confounding from baseline maternal and pregnancy characteristics, the analysis applied propensity score overlap weighting. This approach redistributes analytical weight toward participants with covariate profiles common to both exposure groups, improving comparability without excluding observations. Sensitivity analyses included the use of lamotrigine as an active comparator, a strategy that further isolates drug-specific effects by comparing ASM-exposed children against children exposed to a reference medication with a comparatively well-characterized safety profile.

Neurodevelopmental outcomes were identified through validated diagnostic algorithms applied to claims data. The outcomes evaluated included any neurodevelopmental disorder, attention deficit hyperactivity disorder (ADHD), autism spectrum disorder (ASD), behavioral disorder, developmental coordination disorder, intellectual disability, learning difficulty, and speech or language disorder.

Primary Findings by Drug

Valproate demonstrated the most substantial and consistent associations across neurodevelopmental outcomes. Adjusted HRs for valproate-exposed children ranged from 1.26 to 4.50 depending on the specific outcome, representing a spectrum from modest to marked elevated risk. These findings are consistent with a substantial body of prior literature implicating prenatal valproate exposure in adverse neurodevelopmental trajectories, including reduced cognitive performance and elevated rates of ASD and ADHD. The present study strengthens that evidence base through its large sample, extended follow-up period, and methodologically rigorous confounding control.

Zonisamide, by contrast, has received considerably less research attention than valproate, and the study’s identification of associations with several neurodevelopmental outcomes represents a notable signal warranting further investigation. The investigators described the zonisamide findings as suggesting a need for further evaluation, a measured characterization appropriate to the current state of the evidence.

Levetiracetam and phenytoin were not associated with an increased risk for any of the evaluated neurodevelopmental outcomes. Levetiracetam in particular has become one of the most widely prescribed ASMs in pregnancy given its perceived tolerability profile, and the absence of a detectable neurodevelopmental signal in this large cohort provides clinically relevant reassurance, subject to the limitations discussed below.

Carbamazepine and oxcarbazepine, two structurally related sodium channel blocking agents, demonstrated a modest risk increase for ADHD and behavioral disorders, with HRs ranging from 1.23 to 1.40. While these estimates are statistically modest, they merit attention given the prevalence of both drugs in clinical practice and the cumulative burden that ADHD and behavioral disorders impose on affected children and families.

Topiramate and lamotrigine showed no meaningful associations across most outcomes evaluated. However, both drugs produced a potential signal for intellectual disability, and topiramate additionally showed an elevated HR of 1.23 for learning difficulty. The investigators explicitly noted that these estimates were based on small numbers, rendering them statistically imprecise. The wide confidence intervals associated with rare outcomes such as intellectual disability preclude firm conclusions, and the authors appropriately framed these findings as signals requiring confirmation as data accumulate.

Phenobarbital, an older ASM with known teratogenic concerns, was included in the analysis. The published summary does not specify the direction or magnitude of its associations in the same detail provided for other agents, though its inclusion in a study of this scale enables future subgroup and pooled analyses.

Lacosamide, with only 219 exposed pregnancies, represented a statistically constrained group. Any estimates derived from this subgroup carry substantial uncertainty, and conclusions regarding lacosamide’s neurodevelopmental safety profile cannot be drawn from this dataset alone.

Limitations

Several methodological constraints bear on interpretation of these results. The reliance on healthcare claims data introduces limitations inherent to administrative records, including potential misclassification of diagnoses and incomplete capture of relevant covariates such as maternal seizure frequency, severity, and polypharmacy patterns. Neurodevelopmental diagnoses may also be subject to surveillance bias, whereby children with more medical contact, possibly those born to mothers on more closely monitored regimens, receive diagnoses at higher rates regardless of true prevalence differences.

The variation in sample sizes across drug groups creates asymmetric statistical power. Levetiracetam’s large cohort enables relatively precise estimates, while lacosamide and zonisamide subgroups generate estimates with wider confidence intervals and greater susceptibility to chance findings. The reference group of ASM-unexposed patients with epilepsy, while methodologically preferable to a general population comparator, may still differ from exposed patients in ways not fully captured by propensity score methods. Patients who received no ASM dispensation during pregnancy may represent individuals with milder epilepsy or those who elected to discontinue medication, and their offspring’s neurodevelopmental outcomes may reflect those underlying differences.

The study’s observational design precludes causal inference. Associations identified in this cohort require replication in independent datasets and, where feasible, mechanistic investigation.

Clinical Implications

For clinicians counseling patients of reproductive age who require ASM therapy, these findings reinforce existing guidance recommending avoidance of valproate during pregnancy when alternative agents provide adequate seizure control. The valproate signal observed here, spanning multiple outcomes and ranging to an adjusted HR of 4.50 for specific disorders, is consistent with current teratogenicity warnings incorporated into prescribing guidance across multiple regulatory jurisdictions.

The zonisamide findings introduce a new area of clinical concern. While the evidence base for zonisamide is considerably less mature than that for valproate, the signal observed in this large cohort is sufficient to prompt heightened monitoring of children with prenatal zonisamide exposure and to encourage investigators to prioritize this agent in future pharmacoepidemiological analyses.

The modest associations identified for carbamazepine and oxcarbazepine with ADHD and behavioral disorders suggest that clinicians should not assume structural analogs within the sodium channel blocker class carry identical developmental risk profiles. These findings may influence prescribing preferences when multiple agents achieve comparable seizure control.

The absence of a detectable signal for levetiracetam across outcomes evaluated in a cohort of more than 5,000 exposed pregnancies is a clinically consequential finding. Although this result does not constitute proof of safety, it provides the most robust observational evidence currently available for this widely used agent.

Broader Context

The management of epilepsy in pregnancy presents an inherent clinical tension. Uncontrolled seizures during pregnancy carry maternal and fetal risks that in many circumstances exceed the teratogenic risks associated with ASM use. Abrupt medication discontinuation during pregnancy is associated with status epilepticus, a potentially life-threatening condition. These findings do not argue for medication avoidance in all pregnant patients with epilepsy. Rather, they provide a comparative framework within which clinicians and patients can evaluate the relative neurodevelopmental profiles of available agents when clinical circumstances permit a choice among therapeutically equivalent options.

The population-based design and extended