Paternal valproate use and impact of shared genetic susceptibility on child neurodevelopment

Abstract

Paternal use of valproate during spermatogenesis has been associated with increased risk of neurodevelopmental disorders (NDDs) in offspring, yet the role of genetic confounding is unclear. Using data from the Norwegian Mother, Father, and Child Cohort Study (MoBa), we assessed genetic susceptibility to epilepsy, ADHD and autism spectrum disorders (ASD) in fathers with epilepsy treated with valproate (n = 41), lamotrigine or levetiracetam (n = 37), other anti-seizure medications (ASMs; n = 80), and healthy controls (n = 54,752). Fathers using valproate had significantly higher polygenic risk scores (PRSs) for epilepsy compared to those using lamotrigine or levetiracetam (mean difference: 0.66, 95% CI: 0.21–1.11, p ≈ 0.005), other ASMs (0.41, 95% CI: 0.02–0.81, p ≈ 0.04) and controls (0.85, 95% CI: 0.54–1.15, p = 5.8 × 10⁻⁸). No robust associations were found between paternal ASM use or epilepsy PRS and child neurodevelopmental outcomes. Significant genetic overlap was found among the top 1% of weighted SNPs in the PRSs for epilepsy, ADHD and ASD (428 genes, p ≈ 0.0001), enriched for neurodevelopmental pathways. These results emphasize the importance of considering shared genetic susceptibility when assessing risks of paternal valproate exposure.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-23377-1.

Introduction

Epilepsy is a chronic heterogenous neurological condition that often requires long-term treatment with anti-seizure medications (ASMs). Valproate is an effective ASM, particularly for generalized and drug-resistant seizures¹. Valproate should not be prescribed to women of childbearing potential due to well-established teratogenic effects. Recently, concerns have emerged about the potential impact of valproate on child development when used by males during spermatogenesis. A limited number of preclinical studies in rodents have demonstrated abnormal behavioral outcomes in offspring following paternal valproate exposure during spermatogenesis^2,3.

In 2024, the Pharmacovigilance Risk Assessment Committee (PRAC) of the European Medicines Agency (EMA) recommended caution for men using valproate due to a potential increased risk of neurodevelopmental disorders (NDDs) in children⁴. This was based on a Scandinavian study by market authorization holders (MAH), which found a higher incidence of NDDs in children of men treated with valproate compared to those treated with lamotrigine/levetiracetam (adjusted hazard ratio (HR) of 1.50 (1.09, 2.08))⁵. In contrast, a recent Danish study found no increased NDD risk⁶, aligning with a review that highlight the limited and conflicting evidence⁷. EMA advised precautionary measures including specialist initiation of valproate, mandatory patient information on risks, and the need for effective contraception. The UK’s Medicines and Healthcare products Regulatory Agency (MHRA) went further in restricting valproate use. These recommendations sparked debate due to inconclusive evidence and unclear biological mechanisms.

One possible explanation for the reported valproate-associated NDD risk is confounding by shared genetic susceptibility. Epilepsy and NDDs, such as attention-deficit/hyperactivity disorder (ADHD) and autism spectrum disorder (ASD), are associated with common genetic pathways, suggesting that these comorbidities may reflect a partially inherited predisposition and raises the possibility of genetic confounding^8–10. For instance, children of fathers with epilepsy may inherit genetic risk factors that predispose them to NDDs independently of ASM exposure.

To investigate the potential for genetic confounding in the reported neurodevelopmental risks, we compared genetic susceptibility between fathers using valproate and those using other ASMs. Leveraging PRSs derived from genotype data, we assessed whether inherited genetic factors in fathers might confound the association between valproate use and child NDDs.

Methods

Study population

This study is based on data from the Norwegian MoBa study conducted by the Norwegian Institute of Public Health (NIPH)^11,12. MoBa is an ongoing prospective, population-based birth cohort study (n = 114,500 children, n = 95,200 mothers, and n = 75,200 fathers). Participants complete questionnaires throughout pregnancy and in childhood. This study is based on data version 12 released by MoBa in 2020. MoBa was also linked to the Medical Birth Registry of Norway (MBRN).

The establishment of MoBa and initial data collection was based on a license from the Norwegian Data Protection Agency and approval from the Regional Committees for Medical and Health Research Ethics (REC, S-97045, S-95113). The MoBa cohort is currently regulated by the Norwegian Health Registry Act. Informed consent was obtained from all MoBa participants and/or their legal guardian(s) upon recruitment. All data collection, storage, management and analysis were performed in accordance with relevant guidelines and regulations. All data were de-identified, and the linking of MoBa to health registries was handled by NIPH. Our study was approved by the REC South East Norway (reference number: 239237).

Selection of samples and study design

Sample selection was based on paternal responses to a questionnaire distributed around week 15 of pregnancy (n = 76,380 pregnancies). Analyses were restricted to pregnancies with live birth outcomes for which paternal genotype data was available from MoBa Genetics (n = 55,283 pregnancies; Fig. 1)^13,14. Men using valproate together with lamotrigine and/or levetiracetam were excluded (n = 2 pregnancies), but concomitant use of other ASMs was allowed. We restricted the study population to men using ASMs with self-reported epilepsy. In the case of multiple births, only one of the children was randomly selected. Thus, the exposure groups were: paternal valproate use and epilepsy (n = 41); paternal lamotrigine and/or levetiracetam use and epilepsy (n = 37); paternal use of other ASMs and epilepsy (n = 80); population comparison group with no paternal ASM use and no epilepsy (n = 54,752).

Fig. 1.

Flow chart of sample selection. ASM: antiseizure medication; lamot.: lamotrigine; levet.: levetiracetam; Q1 father: first paternal MoBa questionnaire (week 15 of pregnancy).

Exposures

Paternal ASM exposure and epilepsy were defined as self-reported use and condition in the paternal questionnaire. We have previously shown a high agreement between self-reported ASM exposure and dispensed ASM prescriptions (Cohen’s κ ≈ 0.81¹⁵). The fathers were asked if they used any medications in the 6-month period prior to conception, and if yes, for how long the medication was used (0–1 week; 1 week–1 month; 1 month–6 months). Exposure was defined as any use throughout the 6-month period before pregnancy according to the Anatomical Therapeutic Chemical (ATC) codes: valproate (N03AG01), lamotrigine (N03AX09), levetiracetam (N03AX14), and other ASMs (all N03-, excluding N03AG01/N03AX09/N03AX14; N05BA09; S01EC01).

Outcomes

Child neurodevelopmental traits were assessed throughout childhood until eight years of age and measured based on the questionnaires Q4 (0.5 years), Q5 (1.5 years), Q6 (3 years), Q5y (5 years), and Q8y (8 years), using validated psychometric tests. The neurodevelopmental traits included hyperactivity (1.5–8 years), inattention (1.5–8 years), language difficulties (1.5–8 years), motor difficulties (0.5–5 years), repetitive behaviors (1.5–8 years), and social communication (0.5–8 years; Table S1). A z score was calculated (i.e., the mean score standardized to the entire population). Only complete cases were included. An overview of missing data, the psychometric tests and items included is provided in the Supplemental information.

Genotype data

Genotype data for all MoBa participants are available through the MoBa Genetics initiative (https://github.com/folkehelseinstituttet/mobagen). The MoBa biobank contains blood samples from both parents during pregnancy and children at birth^13,14. Quality control (QC) of the data has been performed¹⁶ and details are reported in Corfield et al. (2022)¹⁶.

Calculation of polygenic risk scores (PRSs)

The LDpred2-auto method was used to compute the PRSs¹⁷. We used GWAS summary statistics, the linkage disequilibrium (LD) reference HapMap3+ of European individuals¹⁸ and genotype data from MoBa. PRSs were generated for three traits, using data from large GWASs for epilepsy¹⁹, ADHD²⁰ and ASD²¹. The GWAS summary statistics were quality-controlled in accordance with recommended steps^17,22,23. Single nucleotide polymorphisms (SNPs) with an INFO score ≤ 0.8 and a MAF ≤ 0.01 were removed (except for epilepsy, where the INFO score was not provided). Further, any SNPs where the standard deviations of the summary statistics and the allele frequencies of the LD reference deviated were removed²³. The PRSs were computed using the ‘snp_ldpred2_auto()’ function of the bigsnpr R package²⁴. The default settings were used, except the initial values for p (polygenicity) was set to 50 different points varying evenly on the logarithmic scale between 0.0001 and 0.9. Further, the ‘allow_jump_sign’ option was disabled and the ‘shrink_corr’ was set to 0.95²³. We only included chains that converged. The PRSs were standardized to have a mean of 0 and a standard deviation of 1.

Gene annotation, functional enrichment, and genetic overlap analyses

The top 1% weighted SNPs in the three PRSs were annotated in accordance with hg19/GRCh37 using the gprofiler2 R package²⁵ and protein-coding genes available in the Ensembl Variation database (updated February, 2024)²⁶. Gene ontology analysis was performed using the R package clusterProfiler^27,28 and the function ‘enrichGO()’. The gene set tested was the overlapping genes between the PRSs, and the background genes were defined as all genes annotated to the top 1% weighted SNPs. The p values were adjusted for multiple testing with a false-discovery rate (FDR) < 0.05, using the Benjamini-Hochberg method²⁹. Semantic similarities between ontology terms were calculated using Jaccard’s similarity index between the 30 most significant ontologies to create five common categories, as implemented in the enrichplot package³⁰. Clustering was performed using Ward’s method and k-means clustering. To test the significance of the genetic overlap between the PRSs, we performed a permutation test by randomly selecting three sets of genes (one for each condition) from the genes annotated to the top 1% of weighted SNPs in each PRS. The number of overlapping genes was calculated for each random selection, and this process was repeated 10,000 times to generate a distribution of overlaps expected by chance. Finally, the observed overlap between the PRSs was compared to this distribution to assess whether it was greater than would be expected by random chance.

Statistical analyses

The difference between the treatment groups was tested using the Kruskal-Wallis test (continuous variables), the chi-squared test (categorical variables; all expected cell counts ≥ 5), or the Fisher’s exact test (categorical variables; expected cell counts < 5). The differences in standardized PRSs between groups were tested using two-tailed t tests. Differences between the treatment groups in the central tendency and distribution of z scores were tested using the Mann-Whitney U and Kolmogorov-Smirnov tests, respectively. All regression models were adjusted for infant sex and paternal education, and only run if the exposure groups for the given trait and time point were ≥ 10 individuals. The p values were adjusted for multiple testing (25 tests) with an FDR < 0.05, using the Benjamini-Hochberg method²⁹. All analyses were performed in the programming language R (v. 4.2.3).

Results

Sample characteristics and group comparisons

Table 1 summarizes the demographic and clinical characteristics of the fathers and mothers included. Significant differences were observed between the treatment groups and population controls for paternal age and educational level, but not in the prevalence of psychiatric comorbidities. Fathers using valproate were generally younger compared to those on other ASMs and population controls. Educational attainment also varied significantly, with a lower proportion of fathers using valproate having completed higher education compared to the other groups. In contrast, there were no significant differences in maternal demographic or clinical characteristics, except for a small difference in sporadic smoking.

Table 1.

Overview of sample characteristics.

	Valproate and epilepsy (n = 41)	Lamot./levet. and epilepsy (n = 37)	Other ASMs and epilepsy (n = 80)	Controls (n = 54,752)	p value
Paternal characteristics
Paternal age (mean years ± SD)	30.5 ± 5.7	30.9 ± 4.3	33.5 ± 5.1	32.6 ± 5.2 79 missing	b, c
Paternal psychiatric disorder* (yes; n (%))	1 (2.4)	1 (2.7)	0 (0)	715 (1.3)	N.S.
Paternal ASM use** (yes; n (%))					a***,b, c
Valproate	41 (100)	0 (0)	0 (0)	0 (0)
Lamotrigine/levetiracetam	0 (0)	37 (100)	0 (0)	0 (0)
Other ASMs	2 (4.9)	7 (18.9)	80 (100)	0 (0)
Paternal education					a
University/college (n (%))	15 (36.6)	22 (59.5)	39 (48.8)	28,329 (51.7)
High school or lower (n (%))	25 (61.0)	14 (37.8)	40 (50.0)	25,354 (46.3)
	1 missing	1 missing	1 missing	1,069 missing
Maternal characteristics
Maternal age (mean years ± SD)	28.3 ± 4.7	29.5 ± 4.1	30.9 ± 4.0	30.2 ± 4.4 21 missing	b, c
Maternal epilepsy (yes; n (%))	1 (2.4)	0 (0)	0 (0)	337 (0.6)	N.S.
Maternal psychiatric disorder (yes; n (%))	6 (14.6)	8 (21.6)	11 (13.8)	8,128 (14.8)	N.S.
Maternal ASM use (yes; n (%))					N.S.
Valproate	0 (0)	0 (0)	0 (0)	47 (0.1)
Lamotrigine/levetiracetam	0 (0)	0 (0)	0 (0)	106 (0.2)
Other ASMs	1 (2.4)	0 (0)	0 (0)	187 (0.3)
Maternal education					N.S.
University/college (n (%))	25 (61.0)	23 (62.2)	43.0 (53.8)	34,875 (63.7)
High school or lower (n (%))	13 (31.7)	10 (27.0)	26.0 (32.5)	16,714 (30.5)
	3 missing	4 missing	11 missing	3,163 missing
Smoking during pregnancy					c
Daily (n (%))	4 (9.8)	1 (2.7)	4.0 (5.0)	2,839 (5.2)
Sometimes (n (%))	3 (7.3)	1 (2.7)	1.0 (1.3)	1,302 (2.4)
Never (n (%))	33 (80.5)	35 (94.6)	75.0 (93.8)	50,494 (92.2)
	1 missing			117 missing
Alcohol during pregnancy					N.S.
Weekly (n (%))	0 (0)	0 (0)	0 (0)	172 (0.3)
Monthly (n (%))	2 (4.9)	1 (2.7)	1 (1.3)	3,174 (5.8)
Never (n (%))	35 (85.4)	34 (91.9)	77 (96.3)	49,505 (90.4)
	4 missing	2 missing	2 missing	1,901 missing
Infant characteristics
Infant sex (female; n (%))	18 (43.9)	18 (48.6)	39 (48.8)	26,730 (48.8)	N.S.
Birth weight (mean grams ± SD)	3,530.8 ± 607.0	3,548.4 ± 482.1	3,556.0 ± 617.7	3,577.1 ± 590.6 61 missing	N.S.
Gestational age (mean weeks ± SD)	39.8 ± 2.0	40.4 ± 1.4	39.6 ± 2.4 2 missing	39.8 ± 2.0 1,290 missing	N.S.
Year of birth					N.S.
1999–2004	6 (14.6)	7 (18.9)	23.0 (28.8)	11,773 (21.5)
2005–2009	27 (65.9)	23 (62.2)	44.0 (55.0)	35,090 (64.1)
	8 missing	7 missing	13 missing	7,889 missing

^a Significant difference between the valproate and the lamotrigine/levetiracetam exposed groups; ^b Significant difference between the valproate and the other ASMs exposed groups; ^c Significant difference between the valproate exposed and control groups.

* Paternal psychiatric disorder includes self-report(s) of ADHD, eating disorder, bipolar disorder, schizophrenia, and/or other mental illness(es).

** Numbers may add up to more than 100%, as some individuals have been exposed to more than one ASM.

*** No significant difference in the use of other ASMs between the valproate and lamotrigine/levetiracetam groups.

ASM: antiseizure medication; lamot.: lamotrigine; levet.: levetiracetam; N.S.: not significant; SD: standard deviation.

Comparison of PRS for epilepsy, ADHD, and ASD among fathers using valproate and other ASMs

The PRSs for ADHD and ASD were similar across treatment groups and controls (Figure S1). In contrast, the PRSs for epilepsy were significantly higher in fathers using valproate compared to those treated with lamotrigine or levetiracetam (mean difference: 0.66, CI = 0.21–1.11, p ≈ 0.005), other ASMs (mean difference: 0.41, CI = 0.02–0.81, p ≈ 0.04), and population controls (mean difference: 0.85, CI = 0.54–1.15, p = 5.8*10⁻⁸), reflecting distinct genetic profiles (Fig. 2A). Restricting the analyses to fathers using only one ASM did not alter the results (data not shown).

Fig. 2.

Comparison of epilepsy PRSs across treatment groups and associations with child neurodevelopmental traits.(A) Density plots of standardized epilepsy PRSs for the valproate exposed group compared to the lamotrigine/levetiracetam group, the other ASMs group, and the control group. Mean epilepsy PRS per group is indicated with a dotted line. A two-tailed t test was used to compare differences in means, the p value of which is indicated in the plots. (B) Plots of standardized beta coefficients from multiple linear regression models with neurodevelopmental trait as the dependent variable, and epilepsy PRS, infant sex, and paternal education as the independent variables. Only the beta coefficients of the epilepsy PRS are shown. One model was run per trait and per time point. ASM: antiseizure medication; PRS: polygenic risk score; y: years.

Associations between ASM exposure, epilepsy PRS, and child neurodevelopmental outcomes

We investigated the differences in central tendency and distribution between the valproate exposed group and comparison groups and found a borderline significant difference between the valproate and lamotrigine/levetiracetam groups for motor difficulties at three years (difference in location of −0.74 [CI: −1.48– (−4.3*10⁻⁶)]; p ≈ 0.03; Figure S2). We did not identify any robust associations between paternal valproate exposure and any of the six child neurodevelopmental outcomes when adjusting for infant sex and paternal education. The only exception was a borderline significant association of paternal valproate use with motor difficulties in children at three years compared to the lamotrigine/levetiracetam group, and with language difficulties at five years of age compared to the control group (standardized beta coefficients of −0.74 [CI: −1.23–(−0.24)], and − 0.64 [CI: −1.10–(−0.19)]; p ≈ 0.03, and p ≈ 0.04, respectively) in a small sample (Figure S3A). Including paternal epilepsy PRS in the models did not meaningfully alter the results (Figure S3B), and no significant associations were found between epilepsy PRS alone and child neurodevelopmental outcomes when adjusting for infant sex and paternal education (Fig. 2B).

Overlap between highest weighted SNPs in the PRSs for epilepsy, ADHD, and ASD

To explore shared genetic susceptibility between epilepsy, ADHD and ASD, we conducted an overlap analysis of the top 1% of SNPs contributing to the PRSs. These results are visualized in UpSet plots³¹, highlighting both overlap of specific genetic variants and annotated genes (Fig. 3A and B, respectively). The analyses of the specific overlapping SNPs revealed a notable overlap between the PRSs, with five SNPs overlapping between all three PRSs (Fig. 3A). Furthermore, the extended analyses examining the overlap of genes annotated to the top 1% of SNPs in the PRSs revealed a larger, significant overlap of annotated genes (n = 428 genes, p ≈ 0.0001, Fig. 3B). The overlapping genes (n = 428) were enriched for categories clustering to five key biological processes (Fig. 3C), including “postsynaptic density specialization membrane”, “anterograde chemical trans-synaptic signaling”, “cell-cell adherens binding adhesion”, “cell junction assembly organization”, and “generation differentiation neuron development”. These categories are highly relevant to the shared pathophysiology underlying epilepsy, ADHD, and ASD^32,33. Notably, the “postsynaptic density specialization membrane”, “anterograde chemical trans-synaptic signaling”, and “cell junction assembly organization” pathways are crucial for synaptic plasticity and neural communication, which are frequently disrupted in both epilepsy and neurodevelopmental disorders^34,35. Similarly, categories such as “generation differentiation neuron development” are critical for neurodevelopment and neuronal guidance, processes implicated in the development of neural circuits involved in both epilepsy, ADHD and ASD^33,36.

Fig. 3.

Genetic overlap and pathway enrichment across epilepsy, ADHD and ASD PRSs. A and B UpSet plots showing the overlap of the top 1% highest weighted SNPs for each of the three PRSs, where (A) shows the number of overlapping SNPs, and (B) shows the overlapping annotated genes among the highest weighted SNPs. Overlapping SNPs/genes are indicated with filled dots for the respective PRSs. The vertical bars indicate the number of SNPs/genes part of the respective intersection. (C) Tree plot showing the hierarchical clustering of semantically similar gene ontology terms into biologically meaningful categories. The 30 most significant ontology terms are visualized with its own branch and dot, color-coded by FDR-adjusted p value, with the size of the dot indicating the number of the overlapping genes being part of the ontology. ADHD: attention-deficit/hyperactivity disorder; adj.: adjusted; ASD: autism spectrum disorder; FDR: false-discovery rate; GABA: gamma-aminobutyric acid; no.: number; PRS: polygenic risk score; SNP: single nucleotide polymorphism.

Discussion

We investigated the genetic predispositions to epilepsy, ADHD and ASD in fathers with epilepsy using ASMs during the spermatogenic period, and the association with neurodevelopmental traits in their children. We found significant differences in PRSs for epilepsy between fathers using valproate and those using other ASMs. We could not find robust associations of ASM use and neurodevelopmental traits. Finally, we found a genetic overlap between epilepsy, ADHD, and ASD, which together with the higher epilepsy PRSs highlight the possibility of genetic confounding of paternal ASM use and child neurodevelopmental outcomes associations.

The genetic susceptibility of fathers with epilepsy using ASMs is an important factor to consider when assessing potential neurodevelopmental risks in offspring. Genetic predisposition to epilepsy, which may differ depending on the specific ASMs used, could confound associations between ASM exposure and child neurodevelopmental outcomes. We identified significantly higher PRSs for epilepsy in fathers using valproate compared to those using other ASMs or controls, likely reflecting the genetic underpinnings tied to specific epilepsy subtypes typically treated with valproate. This is consistent with studies showing distinct genetic architectures associated with different epilepsy subtypes^19,37. While PRSs capture only a portion of overall genetic susceptibility, we still observed significant differences in epilepsy PRS between treatment groups. This is in line with research indicating that PRSs explain a modest, but significant proportion of epilepsy risk^19,37. This finding raises concerns about genetic confounding in pharmaco-epidemiological studies, where paternal genetic differences might contribute to the observed neurodevelopmental risks independently of ASM exposure. Despite the identified differences in epilepsy PRS, we did not identify associations with child neurodevelopmental outcomes.

In contrast to the findings of the PASS study⁵, we did not identify consistent associations, except for a borderline association of paternal valproate use with motor difficulties in children at three years, and with language difficulties at five years of age. However, this signal should be interpreted with caution given the small sample size and wide confidence intervals. These results are consistent with previous studies by Christensen et al.⁶ and Tomson et al.³⁸, which similarly found no evidence linking paternal valproate use to NDDs.

Unlike other studies that have used composite outcomes or diagnostic categories for NDDs, we explored six distinct traits associated with ADHD and ASD, such as hyperactivity, inattention, and motor function, as recommended by expert consensus³⁹. This approach provided a finer resolution of the potential effects of paternal ASM use on child neurodevelopment. Composite outcomes aggregate multiple heterogeneous traits, which can introduce variability and increase the likelihood of false-positive associations⁴⁰ and may mask the variability in how genetic and environmental factors affect specific traits due to variability in diagnostic criteria^41,42. Specific traits may be influenced by different genetic pathways or environmental factors, making trait-based analysis particularly valuable in understanding the heterogeneous nature of ADHD and ASD⁴⁰. The observed genetic overlap between PRSs for epilepsy, ADHD and ASD, suggests that shared genetic susceptibility could contribute to the outcomes.

The observed genetic overlap between the PRSs for epilepsy, ADHD, and ASD, provides evidence for shared genetic susceptibility, which may influence both paternal epilepsy and comorbidity, as well as child neurodevelopmental outcomes. Genes annotated to these top SNPs were significantly enriched for pathways related to for example synaptic plasticity and neurodevelopmental processes. These findings align with previous reports of genetic overlap^8,9. Our results suggest that the increased risk of NDDs in children could partly be explained by genetic susceptibility, questioning the attribution of NDD risks to direct effects of valproate. The fathers in our study taking valproate have a higher genetic predisposition for epilepsy compared to those on other ASMs. Concomitantly, the shared genetic pathways between epilepsy, ADHD and ASD, may result in children inheriting genetic risk factors independently of ASM exposure and creates a potential for genetic confounding. As a result, associations between paternal valproate use and child NDDs may be confounded by these shared genetic susceptibilities. To mitigate the potential for genetic confounding in future studies investigating paternal ASM use and neurodevelopmental outcomes, several approaches could be considered. Incorporating PRSs as covariates, conducting sibling-control or family-based analyses, and using Mendelian randomization can help disentangle genetic risks from ASM exposures.

Strengths and limitations

This study’s strengths include the use of the large, population-based MoBa cohort study. Unlike previous studies, our approach focused on specific trait measures associated with ADHD and ASD. The use of PRSs allowed for analyses of genetic susceptibility and its potential impact on neurodevelopmental outcomes. However, the study was underpowered to investigate group differences in neuropsychiatric traits, especially for the 5-year and 8-year age groups. Also, we attempted trio analyses to disentangle direct genetic effects from indirect paternal genetic influences on child neurodevelopment, but the sample size was too small to support these analyses. Further, the MoBa questionnaires do not include information on epilepsy types, and consequently, we could not differentiate on generalized and focal epilepsies in this study.

Conclusions

The significant genetic differences in PRSs for epilepsy among fathers using valproate compared to other ASMs, along with the genetic overlap between epilepsy, ADHD, and ASD, suggest that genetic factors may confound studies of paternal ASM use and neurodevelopmental outcomes in children.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

EWO performed the analyses and drafted the manuscript. KG conceived the study idea and co-drafted the manuscript. HMEN, MHB and KS contributed to study design, interpretation of findings, and critical revision of the manuscript. All authors read and approved the final version. KG and KS contributed equally to this work and share senior authorship.

Data availability

The datasets presented in this article are available from the MoBa study, but restrictions apply to the availability of these data and so are not publicly available. However, data are available from the authors (EWO; Emilie Willoch Olstad) upon reasonable request and with permission from the MoBa study by applying to datatilgang@fhi.no.

Declarations

Competing interests

MHB reports advisory board honoraria or/and speaking honoraria from Pfizer, Jazz Pharmaceuticus, Angelini Pharma, AbbVie, Lundbeck and Eisai. Received biostatistical research support from Novartis unrelated to the present study; her department has received funding from valproate marked authorization holders to conduct postmarketing drug safety research outside the submitted work, and she is PI on an industry financed phase IV study of a migraine preventive drug (eptinezumab) unrelated to the drugs studied in the present work. KS has received consultant and speaker’s honoraria from Roche and OrionPharma. She also received non‐personal sponsorships from Desitin and Eisai AB in relation to organizing conferences. The other authors declare no conflicts of interest.