Abstract
Dreier JW, Bjørk M-H, Alvestad S, Gissler M, Igland J, Leinonen MK, Sun Y, Zoega H, Cohen JM, Furu K, Tomson T, Christensen J. JAMA Neurol. 2023;80(6):568-577. doi:10.1001/jamaneurol.2023.0674
Importance: Prenatal antiseizure medication (ASM) exposure has been associated with adverse early neurodevelopment, but associations with a wider range of psychiatric end points have not been studied. Objective: To examine the association between prenatal exposure to ASM with a spectrum of psychiatric disorders in childhood and adolescence in children of mothers with epilepsy. Design, setting, and participants: This prospective, population-based register study assessed 4 546 605 singleton children born alive in Denmark, Finland, Iceland, Norway, and Sweden from January 1, 1996, to December 31, 2017. Of the 4 546 605 children, 54 953 with chromosomal disorders or uncertain birth characteristics were excluded, and 38 661 children of mothers with epilepsy were identified. Data analysis was performed from August 2021 to January 2023. Exposures: Prenatal exposure to ASM was defined as maternal prescription fills from 30 days before the first day of the last menstrual period until birth. Main outcomes and measures: The main outcome measure was diagnosis of psychiatric disorders (a combined end point and 13 individual disorders). Estimated adjusted hazard ratios (aHRs) using Cox proportional hazards regression and cumulative incidences with 95% CIs are reported. Results: Among the 38 661 children of mothers with epilepsy (16 458 [42.6%] exposed to ASM; 19 582 [51.3%] male; mean [SD] age at the end of study, 7.5 [4.6] years), prenatal valproate exposure was associated with an increased risk of the combined psychiatric end point (aHR, 1.80 [95% CI, 1.60-2.03]; cumulative risk at 18 years in ASM-exposed children, 42.1% [95% CI, 38.2%-45.8%]; cumulative risk at 18 years in unexposed children, 31.3% [95% CI, 28.9%-33.6%]), which was driven mainly by disorders within the neurodevelopmental spectrum. Prenatal exposure to lamotrigine, carbamazepine, and oxcarbazepine was not associated with an increased risk of psychiatric disorders, whereas associations were found for prenatal exposure to topiramate with attention-deficit/hyperactivity disorder (aHR, 2.38; 95% CI, 1.40-4.06) and exposure to levetiracetam with anxiety (aHR, 2.17; 95% CI, 1.26-3.72) and attention-deficit/hyperactivity disorder (aHR, 1.78; 95% CI, 1.03-3.07). Conclusions and relevance: Findings from this explorative study strengthen the evidence for the warning against the use of valproate in pregnancy and raise concern of risks of specific psychiatric disorders associated with topiramate and levetiracetam. This study provides reassuring evidence that lamotrigine, carbamazepine, and oxcarbazepine are not associated with long-term behavioral or developmental disorders but cannot rule out risks with higher doses.
Commentary
We strive for no seizures and no side effects. Unfortunately, no perfect anti-seizure medication (ASM) currently exists. This situation becomes even more complicated when you have not 1, but 2 patients in front of you.
Clinicians recognize that certain ASMs increase adverse fetal structural and neurodevelopmental outcomes. 1,2 One prominent example is valproate, which offers broad-spectrum seizure suppression and treats several comorbidities, yet has been linked to neural tube defects, attention deficit disorder (ADHD), autistic spectrum disorder (ASD), and cognitive impairment. However, many other neurobehavioral outcomes have remained underexplored.
Dreier et al recently studied associations between prenatal ASM exposure and essentially every possible psychiatric diagnosis. 3 They pooled up to a decade of data from 5 Scandinavian countries. This feat was possible because each country maintains a national registry with common elements, containing data on all births, maternal, and child diagnoses, and prescriptions including their indications (which are typically unavailable in most claims databases). They followed live-born singleton births from mothers with epilepsy (defined as any epilepsy-related inpatient International Classification of Diseases-10 code or use of ASM with epilepsy as the indication, before birth) through adolescence. They compared the time until the first psychiatric ICD-10 code for 16 000 children versus 22 000 without prenatal exposure to ASMs (between 30 days before the last menstrual period and birth).
Prenatal ASM exposure was associated with children developing any psychiatric disorder. After adjusting for maternal factors (eg, age, parity, education, smoking, psychiatric comorbidity), the hazard ratio (HR) was 1.17 (95% confidence interval [CI]: 1.09-1.25). The authors also nicely put everything into absolute terms, with cumulative incidences of 17% versus 14% by age 10 comparing any ASM versus no ASM. However, this overall difference washed out by age 18: 31% versus 31%. This is new information, given previous studies had examined psychiatric studies only in isolation rather than combined.
They also evaluated specific ASMs and diagnoses. Examples: Valproate was the worst, with a 42% cumulative incidence of any psychiatric disorder by age 18, and associations with intellectual disability (HR: 2.7), ASD (HR: 2.5), and ADHD (HR: 1.4). Several new associations were observed as well, such as attachment disorder and “other developmental disorders” referring to speech, language, “scholastic,” and “motor function.” They also found that valproate was not associated with anxiety or mood problems, substance use, or schizophrenia. Lamotrigine and oxcarbazepine were the best, both associated with nothing. Carbamazepine was also mostly bland, except for an unexpected association with tic disorder (HR: 1.8). Topiramate was associated with ADHD (HR: 2.4). There were several nearly significant associations such as intellectual disability and ASD. Levetiracetam had an adjusted HR of 1.30 (95% 0.99-1.71) for any psychiatric disorder, just barely missing significance, and new possible associations including anxiety (HR: 2.2) and ADHD (HR: 1.8).
This generally reinforces existing knowledge. For example, avoid valproate if possible, whereas lamotrigine and oxcarbazepine should remain among preferred ASMs during pregnancy. The study also supports carbamazepine. Though, the association with tic disorders is difficult to interpret, given an unclear biological explanation, but also because false positives almost certainly exist. The study presented over 200 total HR’s without correction for multiple comparisons, in addition to a similarly large number of cumulative incidences and incident rate ratios across 5 countries, 13 ICD categories (including codes F10-F95, a huge number of specific diagnoses), of about 9 ASM mono- or polytherapy combinations, with 3 sensitivity analyses. This amounts to an exhaustive analysis but at the expense of being unable to distinguish noise from signal (type I error).
Other findings are somewhat difficult to interpret considering type II error. Even when pooling a decade of experience from 5 countries, there were only about 300 children exposed to each of topiramate and clonazepam, leading to wide confidence intervals for topiramate (eg, ASD: HR: 1.93, 95% CI 0.95-3.94) and clonazepam being not significantly associated with any outcome. Really only 5 ASMs had meaningful sample sizes, and the reader is left to determine for themselves what a minimal clinically meaningful difference might be.
One interesting point is whether these findings should dissuade clinicians from prenatal levetiracetam. Increased anxiety and ADHD would be cause for concern. Presumably, the association might be even stronger than reported given mean follow-up for children exposed to levetiracetam was only about 4 years, and mental health diagnoses may take time to develop. Likewise, mothers taking ASMs could have been more likely to adopt other protective health behaviors, which could have made ASMs less bad.
However, there are several potentially reassuring points suggesting that we should probably not totally abandon levetiracetam yet. First, even if there were some ill effects, this must be weighed against levetiracetam remaining an often-well-tolerated broad-spectrum ASM with minimal interactions and availability of levels, and data have not borne out other major organ anomalies. Also, having any psychiatric diagnoses was unfortunately quite common in all exposure groups, so it is not as if the risk is zero if untreated or treated with other ASMs.
Second, confounding could alternatively make ASMs appear worse than they truly are if mothers taking ASMs have worse prepregnancy epilepsy and genetic loading. There are also complex observational time dynamics regarding how ASMs affect seizures and vice versa. Disentangling the effects of seizures versus ASMs may require detailed data about changes in both during pregnancy and potentially also advanced statistical models (eg, mediation, or g methods for treatment-confounder feedback 4 ). Furthermore, mothers taking ASMs could possess better health care leading to greater early detection of psychiatric diagnoses (detection bias), though admittedly double-blinding to overcome this issue would be obviously ethically problematic. Finally, these data do not capture postpregnancy factors that could also differ between mothers who did versus did not take prenatal ASMs. So, causal effects are challenging to infer from these data, both due to having 200+ HR’s producing multiple comparisons, but also because of complex issues regarding time-varying confounding and real-world biases. Third, claims generally do not measure the severity of psychiatric diagnosis, which would be important to understand before interpreting the meaning of associations.
This work opens the door for several more future directions. Another research need includes assembling sufficient sample sizes regarding newer ASMs (eg, lacosamide). It is a bit of a catch-22. Newer ASMs have little data thus clinicians may avoid them, but then we are still left with little guiding clinical data, thus high-quality preclinical work is crucial to evaluate safety. Moreover, further data are needed regarding dose–response relationships such as how maternal adherence, daily dose, or timing/duration of exposure might influence the tipping point.
