Cognitive outcomes after fetal exposure to carbamazepine, lamotrigine, valproate or levetiracetam monotherapy: Data from the EURAP neurocognitive extension protocol

.001). Conclusions: Consistent with previous reports, our results provide evidence for an adverse effect of prenatal exposure to valproate on verbal development. Our finding of relatively weaker performance of VPA-exposed


Introduction
Antiseizure medications (ASMs) are used commonly in pregnancy, with ASM exposure being recorded in 0.6 % to 1.1 % of pregnancies in the Nordic countries [1].Prenatal exposure to certain ASMs is associated with an increased risk of major congenital malformations [2,3].Additionally, registry-based studies have shown that children born to mothers treated with valproate (VPA) and topiramate (TPM) during pregnancy have a higher incidence of neurodevelopmental disorders, including intellectual disability and autism spectrum disorder (ASD) [4][5][6][7], although data on the association between TPM exposure and neurodevelopmental impairments are conflicting [7,8].
Prospective cohort studies have confirmed that prenatal exposure to VPA [9,10] increases the risk of cognitive impairment.The risk appears to be dose-related [9,11] and to affect especially verbal IQ [12,13].VPAexposed children have been found to have impaired scores in sentence repetition and auditory attention [14].Although adverse effects of prenatal exposure to VPA on memory have been reported [15,16], data on specific cognitive abilities are scarce.
Prenatal exposure to lamotrigine (LTG) [9,11,17,18], carbamazepine (CBZ) [12,18] and oxcarbazepine (OXC) [6,7,19] appear to carry a low risk for cognitive impairment, and preliminary data suggest that the same also applies to levetiracetam (LEV) [6,10].There are, however, preliminary reports of impaired cognitive performance after prenatal CBZ exposure compared with unexposed controls [11,20], as well as signals for a potential association of in utero exposure to LEV with attention deficit hyperactivity disorder (ADHD) and anxiety [19], and for a potential decrease in adaptive functioning in children born to mothers exposed to high plasma concentrations of LTG and LEV [21].As for other newer ASMs, with the exception of the aforementioned potential deleterious effects of TPM, the data are too limited for meaningful interpretation [18,22].
The most important confounding factor for cognitive function in the offspring is maternal intelligence [9].Reliable detection of mild to moderate disorders of cognitive function is not possible before the age of approximately six years, as testing requires sufficient brain maturation and cooperation skills.At that age, however, environmental confounders such as socioeconomic and psychosocial factors can also play an important role.The aim of our study was to provide further information on long-term neurocognitive development in children exposed prenatally to CBZ, LTG, VPA or LEV monotherapy and evaluated at age 6-7 years.

Patients and study design
The neurocognitive extension study (NCEP), an extension of the prospective EURAP study [2], is a registry-based multicenter study performed in the Netherlands, Italy, Germany, Czech Republic, Finland, Sweden and Spain.Eligible mother-child pairs were identified from the core EURAP study.The criteria for eligibility were: 1) prenatal exposure to CBZ, LTG, VPA or LEV monotherapy during the entire period from conception to birth; 2) mother's ability to take care of the child from birth until the time of assessment; 3) availability of information on potential confounders (other than ASMs) that could influence neurodevelopmental outcomes, and 4) no significant pre-, peri-and postnatal neurological co-morbidity of the child, such as known chromosomal/ genetic syndromes or gestational age less than 37 weeks.To minimize selection bias, centers were asked to include, whenever possible, consecutive mother-child pairs as initially enrolled in the EURAP registry.Each participating center confirmed the eligibility of the motherchildren pairs identified by the EURAP central registry in Milan.
Prospectively collected data on maternal epilepsy type and etiology, ASM treatment and seizures during pregnancy were retrieved from the core EURAP registry.ASM doses refer to the doses used at conception.For the neurocognitive protocol, follow-up was semi-prospective because written consent was obtained after the child's birth.Data on family type, maternal and paternal educational level, occupation, employment and current health (including seizures and medications) as well as previous developmental assessments of the child were collected in a structured interview with the mother at the time of the child's neurocognitive evaluation (supplementary data, Appendix, A).All data, including the results of neuropsychological evaluation described below, were entered into a web-database or sent to the NCEP coordinator in Excel format.Details of the study protocol are available online [23].

Neuropsychological assessment
Neurocognitive evaluations of the children and their mothers were carried out at the age of 6-7 years at each site by a neuropsychologist or psychologist blinded to the exposure data.The WAIS was used for IQ testing of the mothers.The majority of the children were assessed with the WISC-III and NEPSY-II scales.Some children were tested with different versions of WISC and NEPSY (see Results and supplementary Appendix, B) and due to changes in content and scoring systems between versions, only the children assessed with WISC-III and NEPSY-II were included in the analysis.The scoring system of some of the NEPSY-II tasks also varied between countries and we included only those NEPSY-II tasks for which standard scores were available for most of the participants (subtests and numbers shown in supplementary Appendix, C).Based on normal distribution approximately 68 % of the WISC-III indices (mean = 100, standard deviation = 15) in population fall between 85-115 and NEPSY-II standard scores (mean = 10, standard deviation = 3) between 7-13.In clinical context, certain interpretative limits can also be applied.For example, WISC-III indices between 90-109 and NEPSY-II test scores between 8-12 are considered to be average or expected.
The primary endpoint was the verbal IQ (VIQ) of the child at age 6 to 7 years.The secondary endpoints were performance IQ (PIQ) and fullscale IQ (FSIQ) as well as specific cognitive functions of the child at age 6 to 7 years.

Statistical analysis
The sample size targeted in the study protocol was 95 mother-child pairs in each exposure group.This was based on setting the clinically significant verbal IQ difference at 7.5 points with an estimated SD of 15 points, and 80 % power to detect a difference at a family-wise type I error rate equal to 0.05 (two-tailed) for each of four individual pairwise comparison of one ASM exposure group against another, assuming an attrition rate equal to 10 %.
Data was analysed and visualized using SPSS and Python.Differences in background variables between the exposure groups were tested with chi square or ANOVA.Differences in proportions of children with IQ over and below 85 were tested with the Freeman and Halton extension of the Fisher's test.Differences in VIQ, PIQ, FSIQ and neurocognitive tests were tested with ANOVA using observed estimates and adjustment for confounding variables.Maternal IQ (VIQ, PIQ or FSIQ), sex assigned at birth (male or female), type of maternal epilepsy (generalized or other) and paternal education (university or other) were included as confounding variables.Bonferroni's correction was applied for testing statistical significances of multiple pairwise comparisons between individual ASM groups within each of the separate ANOVA models to avoid family-wise type-I error.Additionally, each ASM group was also contrasted against the overall average of all children (all ASM groups combined).No imputation methods were applied in case of missing data.

Ethics statement
The study was approved by the ethical committees of each participating center.All mothers gave their consent in writing.

Patient characteristics
The first participant was enrolled in the study on September 27, 2008.By February 18, 2020, 177 mother-child pairs had been entered in the database, and the study had to be terminated due to enrolment difficulties.One reason for some mothers declining participation relates to the fact that they were contacted several years postpartum, and were probably less motivated to contribute than they would have been should the invitation had come prior to delivery.An additional hurdle to enrolment was the unavailability of a neuropsychologist, particularly in centers with small number of participants and prolonged intervals between assessments.
Seven children were found not to meet eligibility criteria and no data was available for one child, leaving 169 children for analysis, including eight sibling pairs (LTG=5, CBZ=1, VPA=1, LEV=1) and one set of unidentical twins in the LTG-exposed group.The proportion of included children versus all eligible children recruited by EURAP during the study period in the participating centers was 169/641 or 26 %.
Of them, 99 were investigated in The Netherlands and included in a previous report [24].Malformations diagnosed after the first year of life were reported in two cases, including a persistent ductus Botalli in a child exposed to LTG (maternal dose, 150 mg/day) and funnel chest in a child exposed to VPA (1200 mg/day).Febrile seizures or epilepsy were reported in two children, one exposed to LTG and the other to VPA.
Maternal, paternal and child demographic data and other clinically relevant details are shown in Table 1.LTG was the most common exposure (54 %).Maternal epilepsy types differed significantly between the exposure groups, with generalized epilepsy being more prevalent in the VPA group and least prevalent in the CBZ group.There were more children with no siblings in the VPA group compared to other groups.There were no other significant differences in demographic variables across groups.
A total of 162 children who had results for WISC-III and/or NEPSY-II (LTG n = 80, CBZ n = 37, VPA n = 27, LEV n = 18) were included in the analysis.The number of children included in different comparisons varies depending on whether adequate data was available (Figs. 1-2, Tables 2-3, Supplementary Tables S1-S3 and Appendix B).

IQ assessments (WISC-III scores)
Observed (unadjusted) VIQ, PIQ and FSIQ scores in each exposure group are shown in Table 2 and Fig. 1.Plots of the same scores in relation to maternal ASM doses are illustrated in supplementary Figures S1-S4.Mean values for VIQ, PIQ and FSIQ scores were within age-based norms (IQ>85) in all exposure groups (Table 2, Fig. 1A).There were no statistically significant differences in PIQ and FSIQ across groups, but a statistically significant difference was identified for VIQ scores (P=0.026)(Table S1).In pairwise comparisons, children exposed to VPA had lower VIQ scores than children exposed to LEV (P=0.036) and children from all groups combined (P=0.007) (Fig. 1, Table S1).Conversely, children exposed to LEV had higher VIQ scores (P=0.036)When adjustment for confounding variables was introduced in the model, there were no longer any significant differences in VIQ, PIQ or FSIQ scores across groups, although the difference between the VPA group and all groups combined approached borderline significance (P=0.051) (Fig. 1A-B, Table 3).

Neuropsychological tasks (NEPSY-II scores)
Observed NEPSY-II scores in each exposure groups with results of ANOVA testing are shown in Fig. 2 and supplementary Table S2, whereas plots of NEPSY-II scores in relation to maternal ASM doses are illustrated in supplementary Figures S1-S4.Group averages of neuropsychological task scores were within the age-based norm (standard scores ≥ 7) in all ASM groups for every task, with one exception.Children exposed to LEV had below age norm scores in delayed memory for names.Significant differences across exposure groups and pairwise comparisons were observed for comprehension of instructions, immediate and delayed memory for faces, and delayed memory for names (Fig. 2 A and supplementary Table S2).
After adjusting for confounders, differences across ASM groups remained significant for comprehension of instructions, immediate and delayed memory for faces, and delayed memory for names (Fig. 2B and supplementary Table S3).Children exposed to VPA had lower comprehension of instructions scores than children exposed to LTG, LEV or children from all groups combined.On the other hand, LTG-exposed children performed better in comprehension of instructions than children from all groups combined.Children exposed to CBZ had higher scores than the children from all groups combined in immediate and delayed memory for faces and in delayed memory for names.In the latter task, the LTG group also performed better than children from all groups combined.The VPA group had lower scores in both immediate and delayed memory for faces compared to children from all groups combined.In pairwise comparisons, VPA group had lower scores than LTG and CBZ groups in immediate and in delayed memory for faces.LEV-exposed children had lower scores in delayed memory for names than children from all groups combined and any of the other ASM groups separately.

Discussion
Our study provides data on IQ scores and performance in neuropsychological tasks at age 6-7 years in children exposed prenatally to LTG, CBZ, VPA or LEV.Although enrolment difficulties did not permit the achievement of the target sample size for all cohorts, a number of differences in the assessed outcomes among exposure groups could be identified.As expected, based on literature data [18], most adverse findings concerned VPA-exposed children.Children exposed to VPA had lower observed VIQ scores than children from all groups combined, a difference which had borderline statistical significance after adjusting for confounders.Of note, while every ASM group had mean WISC-III indices within age-based norms (85-115), the group exposed to VPA showed the highest proportion (15 %) of children with an observed VIQ below 85.On the other hand, we did not find any evidence for an adverse effect of VPA exposure on PIQ or visuospatial performance.After adjustment for confounders, children born to mothers treated with VPA were found to have lower scores in comprehension of instructions compared to children exposed to LTG, LEV or children from all groups combined, as well as lower scores in immediate and delayed memory for faces compared to children exposed to CBZ and LTG, and children from all groups combined.Deficits in verbal development after prenatal exposure to VPA have been reported in earlier studies [9,10,12].Our results concerning VIQ and comprehension of instructions are consistent with an adverse effect of VPA exposure in utero on language development but does not explain how the verbal development is compromised.Several neuropsychological features are associated with developmental language deficits and language-related learning difficulties, including phonological processing, verbal short-term memory and working memory [25,26].The neuropsychology of developmental language deficits is likely to involve several aspects of neurocognition [27], which could not be fully assessed based on the data collected in our study.Previous investigations have not reported specific deficits in phonological processing in VPA-exposed children, but impairments in attention regulation and short-term verbal memory have been reported [14], as well as possible adverse effects on memory functions in general [15].Our observation that VPA-exposed children also had lower scores in immediate and delayed memory for faces compared to children exposed to CBZ and LTG, and children from all groups combined, could reflect the presence of general deficits in memory functions.This is in line with reports suggestive of prenatal VPA exposure being a risk factor for memory dysfunction [14][15][16].
Our findings on the effect of prenatal VPA exposure on verbal tasks, especially comprehension of verbal instructions, and face memory, could also indicate altered processing of socially relevant information such as faces and semantic content of spoken language.Register-based studies have shown that children born to mothers treated with VPA are at increased risk for developing ASD [5,7].A previous prospective investigation which included 13 of the VPA-exposed children assessed in this study reported a high incidence of borderline or clinically significant behavioral problems (32 %) and social difficulties (16 %) after prenatal VPA exposure, based on parental reports [28].Persisting difficulties in social communication and interaction are one of the core features in ASD, and impairments in face recognition have been associated with ASD [29].Difficulties in verbal communication, such as deficits in semantic or pragmatic aspects of language are not required for ASD diagnosis, but they are also common among individuals with ASD [30].The core features of ASD were not covered in our study, but the findings concerning comprehension of instructions and face memory may be compatible with some features commonly associated with ASD.
Many neurodevelopmental disorders are not associated with distinct neurocognitive profiles or known specific comorbid neurocognitive deficits.Hence WISC may not identify all neurodevelopmental disorders or delays.This may be reflected in our study by valproate exposed infants having the highest incidence (25 %) of previously diagnosed developmental delay while the differences measured by WISC between the ASM groups were less evident.
Although confounding by indication might affect some of the findings in VPA-exposed children, it has probably a minor effect compared to the adverse effects of valproate, which has been established in several well designed studies [9,10,12].Further, adverse outcomes associated with VPA exposure in our study remained significant after adjustment for confounders, including type of maternal epilepsy.Growing awareness of the adverse reproductive effects of VPA has led over the years to a major decline in the use of this drug in women of childbearing potential [1,2,31,32].This decline has resulted in a major reduction in the prevalence of birth defects in the offspring of ASM-treated mothers with epilepsy [2].It is reasonable to assume that changes in ASM prescribing patterns are also resulting in a decrease in the prevalence of neurodevelopmental disorders in these children.
In our study, prenatal exposure to LTG or CBZ was not associated with adverse outcomes in VIQ, PIQ, FSIQ or neuropsychological tasks.This was also true for LEV-exposed children, with the exception that children born to LEV-treated mothers showed delayed memory for names scores below age norms, and lower than children exposed to CBZ, LTG and VPA.LEV-exposed children, however, showed no impairment in other memory tasks, and therefore a possible signal for an adverse effect of prenatal LEV exposure on memory remains inconclusive.Most of the previous studies on neurodevelopment in children exposed prenatally to LEV have reported favourable outcomes [10], 2024; [6], but there has been a population-based register study that described a potentially increased risk of ADHD and anxiety in these children [19].
Overall, our results are consistent with those summarized in a recent systematic review that assessed studies published between 1990 and 2023 [18].Based on data from the 43 studies that met eligibility criteria for inclusion, the authors concluded that use of VPA and TPM during pregnancy is associated with a significantly increased risk of neurodevelopmental effects in the offspring, that LTG seems to be free of neurodevelopmental effects, and that the data concerning potential neurodevelopmental risks with other ASMs are 'mixed or inadequate to draw definite conclusions'.The review also highlighted additional limitations of existing evidence, including heterogeneity in the reported outcomes measures and a small sample size in most studies.The fact that the risk of adverse neurodevelopmental effects after prenatal exposure remains unknown for the majority of currently used ASMs highlights the need for further research in this area.

Strengths and limitations
Our study has strengths but also limitations.Strengths include the prospective collection of exposure data, follow-up until 6-7 years of age, inclusion of a relatively large cohort of LTG-exposed children, use of standardized structured tools for assessment of neurocognitive function, and adjustments for some of the most important confounding factors.The most important limitation is the inclusion of fewer numbers of childmother pairs than target for the VPA, CBZ and LEV exposure groups.This limited the statistical power for between-group comparisons, and prevented application of more complex statistical modelling as well as evaluation of risks associated with other variables such as ASM dose, maternal seizures during pregnancy, household socio-economic status, and gender of the offspring.The influence of dose could be particularly important in view of the reported dose-dependency in risk of neurodevelopment disorders after prenatal exposure to VPA and TPM [18].Moreover, a recent report signaled a potential decrease in adaptive functioning in children born to mothers exposed to high plasma concentrations of LTG and LEV [21], suggesting that ASMs generally considered as safe could have deleterious effects when applicable dose thresholds are exceeded.
Although we aimed at enrolling consecutive pregnancies, recruitment proved more difficult than envisaged and the fact that many mothers declined participation may have generated selection bias.Additionally, data was gathered in seven different countries with the Netherlands providing the bulk of the data and the possible crosscultural differences were not controlled in the analyses.During the study period, WISC and NEPSY were updated to new versions and some of the tasks underwent some changes concerning content and scoring.
Controlling statistically for the effect of different test versions was deemed to be unfeasible due to the small sample size.On the other hand, disregarding the effect of test versions and pooling all data could have led to biased estimates due to the unbalanced application of different test versions across ASM groups.Therefore, we restricted the analysis to children assessed with the same test versions (WISC-III, NEPSY-II) to enhance the reliability of the results.
To this end we also chose to include only the tasks which were scored with comparable methods in different sites/countries, but this limited the number of tasks included in the analyses.

Conclusions
Our results are in line with previous studies reporting an increased risk of impaired verbal (but not non-verbal) reasoning skills in preschool age children exposed to VPA in utero.Language difficulties combined with deficits in comprehension of instructions and face recognition in VPA-exposed children may be associated with ASD-like features or a more general adverse effect on memory.Exposures to LTG or CBZ were not associated with cognitive dysfunction.The number of children exposed to LEV was too small to allow conclusions.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Table 1
Details of study participants by exposure group.childrenfromall groups combined.The proportion of children with VIQ below 85 (Table2) was highest in the VPA group (15 %), followed by the LTG group (5 %); none of the children in the CBZ and LEV groups had a VIQ below 85 (P=0.051).Proportions of children with PIQ or FSIQ scores below 85 were not statistically significantly different across groups.
Abbreviations: ASM, antiseizure medication; NA=not applicable.PIQ, performance IQ; VIQ, verbal IQ.Statistical tests: Chi square for categorical and one-way ANOVA for continuous variables.than

Table 2
Means, standard deviations and range of the observed WISC-III scores, number of children included in the analysis with proportion of scores below 85. Results of ANOVA comparisons are shown in supplementary Table1.