Psychiatric comorbidities in epilepsy: population co-occurrence, genetic correlations and causal effects

Swedish population-based registries

To estimate the population-level co-occurrence of epilepsy and psychiatric conditions, we derived a nationwide cohort of 7 628 495 individuals born between 1954 and 2011 from the Swedish population registries (figure 1). Using the Total Population Registry and the National Patient Registry (initiation 1969, end of follow-up 2016), we collected information on vital statistics, immigration status (born in or outside Sweden), birth year and any inpatient diagnosis (and specialised outpatient after 1997) of the investigated psychiatric conditions and epilepsy (see online supplemental table 1 for International Classification of Diseases (ICD) 8/9/10 codes). We also identified disposable income quintiles for each year between 1968 and 2016 and retained the highest quintile obtained over a person’s lifetime from the Income and Taxation Registry, which we conceptualised as both a socioeconomic proxy and a measure of resilience to morbidity manifestation. Individuals born after 1996 with missing disposable income values were designated as a distinct category, as they were probably too young to have accumulated disposable income.

Request permission

The National Patient Registry records the date of diagnosis and not the date of onset. Therefore, we focused solely on estimating lifetime comorbidities of epilepsy and psychiatric disorders.

Genome-wide association data sources and phenotypes

Summary statistics were retrieved from the largest European ancestry genome-wide association study (GWAS) of epilepsy performed by the International League Against Epilepsy (ILAE) Consortium on Complex Epilepsies.7 In brief, all epilepsy cases were classified according to ILAE standard and all phenotypes were assessed by epilepsy specialists at individual sites. Controls (persons without epilepsy) were derived from the available population-based datasets within the Consortium on Complex Epilepsies and partially screened for the presence of neurological conditions. Extended phenotypic information is available in the original publication.7 To reduce bias from population stratification, we limited our investigation to a European ancestry sample, which was identified using principal component analysis in the original GWAS.7 Given the phenotypic and genetic heterogeneity of epilepsies, we retrieved summary statistics separately for epilepsy (all epilepsies) (n cases=27 559), focal epilepsies (n cases=14 939) and genetic generalised epilepsies (n cases=6952) (table 1) per the diagnoses of epilepsy specialists in the original GWAS.7

We then extracted summary statistics from the European ancestry GWAS of commonly studied psychiatric comorbidities of epilepsy (table 1). The recent ILAE GWAS7 performed linkage disequilibrium score regression (LDSC) on some psychiatric phenotypes. However, we extended this analysis by using larger GWAS datasets and investigating a broader spectrum of psychiatric conditions. These included GWAS on anorexia nervosa,17 anxiety,18 attention-deficit/hyperactivity disorder,19 autism,20 bipolar disorder,21 depression,22 intelligence,23 obsessive-compulsive disorder,24 schizophrenia,25 suicide attempts26 and tic-disorders (Tourette syndrome).27 Detailed descriptions of the phenotypes and GWAS analyses can be found in the original publications.17–27

Genetic correlations—cross-trait linkage disequilibrium score regression

To quantify the genetic correlation between epilepsy and psychiatric comorbidities, we employed cross-trait LDSC.14 15 LDSC leverages the pattern of linkage disequilibrium among common genetic variants to estimate the genetic correlation across phenotypes.15 We obtained precomputed linkage disequilibrium scores based on the 1000 Genomes Project European Reference Panel. We estimated the LDSC using an unconstrained intercept to allow for sample overlap and population stratification. Univariate LDSC criteria for consistent estimation appeared to be generally satisfied (heritability Z-score >1.5, mean χ² >1.02, single-nucleotide polymorphism (SNP) heritability intercept 0.9 to 1.1).28 A Bonferroni correction of 11×3 test was applied for the genetic correlation analysis (α=0.001515).

Mendelian randomisation

To assess the potential effect of epilepsy on psychiatric conditions and these conditions on epilepsy, we employed bidirectional two-sample MR. MR is based on the principles of instrumental variable analysis using common genetic variants that are robustly associated with a phenotype (exposure) of interest as instruments, yielding an estimate of the causal effect of genetic liability for a phenotype on the outcome of interest if the genetic variant is a valid instrument.16 A genetic variant is a valid instrument if (i) it is robustly associated with the exposure (relevance), (ii) it is not associated with any measured/unmeasured confounders of the exposure-outcome associations (independence) and (iii) it does not affect the outcome via pathways other than the potential effect on the exposure of interest (exclusion restriction).29 Importantly, as genetic variants are fixed at conception and randomly assorted during meiosis, MR is considered robust against reverse causality and many forms of environmental confounding factors.

In two-sample MR, SNP-exposure and SNP-outcome estimates are extracted from two independent GWASs of the same underlying population.29 In bidirectional MR, the procedure is repeated in both postulated directions in separate analyses.

To derive the instruments for our MR analyses, we pooled exposure and outcome GWAS data and excluded non-overlapping SNPs. To retain the highest number of SNPs, we repeated this procedure for each psychiatric GWAS separately. From these datasets, we then extracted all variants with a p value <5×10⁻⁰⁶, a threshold employed to maximise power. This inclusive approach may, under some scenarios, lead to a conservative estimate in the two-sample MR because of weak instruments (so-called weak instrument bias), assuming the absence of sample overlap between GWASs. The same p value threshold has been previously employed in studies of psychiatric phenotypes.30 The identified variants were clumped using PLINK LD-clumping, with an R² <0.01 and a 10 000 kb window, based on the 1000 Genome European phase III reference panel. SNPs from the epilepsy GWAS were then extracted from the GWAS of each outcome and harmonised to ensure that the effect estimates were expressed on the same allele.31

For the primary analysis, we employed the inverse-variance-weighted (IVW) method, which estimates the causal effect under the assumption that all instruments are valid.29 Specifically, the IVW method is the ratio of SNP-exposure and SNP-outcome associations weighted by their relative precision. Because the assumptions of IVW are strong32 (eg, no horizontal pleiotropy), we performed a series of standard sensitivity analyses to assess the robustness of the IVW causal effect estimates (see online supplemental material). A Bonferroni correction of 11×3×2 test was applied for the MR (α=0.00075758).

Population-wide lifetime comorbidity

We observed substantial psychiatric morbidity among patients with epilepsy. Almost half of all individuals diagnosed with epilepsy were also diagnosed with a (neuro-)psychiatric disorder in their lifetime (adjusted lifetime prevalence (%) 44.09, 95% CI 43.78 to 44.39; OR 4.60, 95% CI 4.54 to 4.66, p<0.001) (figure 2A), compared with 15.52% among those with no diagnosis of epilepsy (adjusted lifetime prevalence (%) 15.52, 95% CI 15.50 to 15.55) (for extended results, see online supplemental table 2). Each of the considered psychiatric disorders was more common among those diagnosed with epilepsy (figure 2B), than among those who were not, although the absolute lifetime prevalence of specific psychiatric disorders varied according to the underlying prevalence in the population. There was no apparent distinction between the co-occurrence of neurodevelopmental and psychiatric disorders; both forms of comorbidity were more common among those with epilepsy. The highest relative occurrence was observed for autism (OR 6.09, 95% CI 5.93 to 6.26) and intellectual disability. Intellectual disability was 27 times more common among those with epilepsy (adjusted lifetime prevalence (%) 11.12, 95% CI 10.94 to 11.30; OR 27.51, 95% CI 26.9 to 28.12), compared with those not diagnosed with epilepsy (adjusted lifetime prevalence (%) 0.51, 95% CI 0.51 to 0.52).

Request permission

The estimated lifetime comorbidity did not materially differ before or after controlling for covariates, although all estimates were elevated in crude analysis (unadjusted lifetime prevalence of any psychiatric diagnosis (%) 48.61, 95% CI 48.29 to 48.93; OR 5.16, 95% CI 5.09 to 5.23, p<0.001) (online supplemental figure 2 and online supplemental table 2). Lifetime comorbidity did not vary materially between men and women (online supplemental figure 3) or between focal and generalised epilepsy (online supplemental figure 4) and was not limited to recent birth cohorts (online supplemental figure 5).

Genetic correlation

We observed a positive genetic correlation between epilepsy and attention-deficit/hyperactivity disorder (r_g=0.18, 95% CI 0.09 to 0.27, p<0.001) (figure 3) (for extended results see online supplemental tables 3–5). The genetic correlation between epilepsy and attention-deficit/hyperactivity disorder was more pronounced for focal epilepsy (r_g=0.23, 95% CI 0.09 to 0.36, p<0.001). Furthermore, we identified a negative genetic correlation between epilepsy and intelligence (r_g=−0.20, 95% CI −0.28 to −0.13, p<0.001), which was also more pronounced for focal epilepsy (r_g=−0.24, 95% CI −0.36 to −0.12, p<0.001). The negative genetic correlation between epilepsy and intelligence may be consistent with the observational association between epilepsy and intellectual disability, as the intelligence scale is inversely related to intellectual disability. We observed modest-to-weak genetic correlations between the remaining psychiatric disorders and epilepsy, regardless of the type of epilepsy (figure 3).

Request permission

Mendelian randomisation

In the MR analyses, limited evidence existed of an association between genetic liability to epilepsy and any of the considered psychiatric conditions after Bonferroni correction (figure 4) (for extended results, see online supplemental tables 3–5). The largest observed association was for genetic liability to epilepsy and anxiety disorder, which did not pass Bonferroni correction (p>0.000758). Specifically, genetic liability to epilepsy was associated with an increased likelihood of anxiety disorder (OR 1.55, 95% CI 1.09 to 2.22, p=0.016); a finding that was not replicated in sensitivity analysis (online supplemental methods and online supplemental table 7).

Request permission

Limited evidence existed of an association between genetic liability to most psychiatric conditions and epilepsy (figure 4). One exception was intelligence. Genetic predisposition to higher intelligence reduced the likelihood of epilepsy (OR 0.89, 95% CI 0.85 to 0.93, p<0.001), and the focal (OR 0.88, 95% CI 0.84 to 0.92, p<0.001) and generalised subtypes of epilepsy (OR 0.87, 95% CI 0.87 to 0.94, p<0.001).

Although there were strong assumptions in our primary MR analysis (IVW), our findings were mostly consistent across sensitivity analyses (online supplemental table 7). Generally, our sensitivity analysis was inconsistent with widespread horizontal pleiotropy in our MR analysis (online supplemental figure 6 and online supplemental table 6). That is, the MR-Egger intercepts did not suggest that the selected genes were invalid instruments through an independent effect on the outcome (violation of the third MR assumption: exclusion restriction).

Main findings

Using data from nationwide health registries in Sweden, we found that individuals with epilepsy had a substantially higher prevalence of neurodevelopmental and psychiatric comorbidities. Using cross-trait linkage disequilibrium analysis, we found genetic correlations between epilepsy and certain psychiatric conditions, which, in some cases, such as attention-deficit/hyperactivity disorder, were more pronounced in focal epilepsy. The majority of genetic correlations observed between epilepsy and psychiatric conditions were modest, supporting the partial role of a genetic basis for these comorbidities. Finally, by leveraging common genetic variations, we did not find strong evidence to suggest that these comorbidities are explained by the effects of epilepsy on psychiatric conditions or psychiatric conditions on epilepsy. Taken together, these findings suggest that there are important genetic correlations between epilepsy and some psychiatric conditions. However, these are unlikely to explain the full extent of psychiatric comorbidities in epilepsy.

Our finding that the genetic correlation between epilepsy and attention-deficit/hyperactivity disorder is more pronounced in focal epilepsy is novel, although the general correlation between epilepsy and attention-deficit/hyperactivity disorder is consistent with recent work.9 33 For example, a recent study on the co-aggregation of epilepsy and attention-deficit/hyperactivity disorder in Swedish families found a strikingly similar genetic correlation (0.21, 95% CI 0.02 to 0.40),33 although they were unable to explore subtypes of epilepsy. Our findings suggest that focal epilepsy is the main cause of this association. This contrasts with the aim of a previous study that examined generalised epilepsy because of its higher heritability.9 However, using updated GWAS data, we also confirmed the identification of a shared genetic overlap between epilepsy and attention-deficit/hyperactivity disorder.9 Although focal epilepsy has a lower heritability than generalised epilepsy,7 our findings suggest that pleiotropic genes are distinctly associated with focal epilepsy and attention-deficit/hyperactivity disorder. However, we noted that there was no substantial difference in the population-wide prevalence of psychiatric comorbidities between focal and generalised epilepsy (online supplemental figure 4), although subtype characterisation of epilepsy in Swedish registries may not be accurately recorded, and many individuals may have multiple seizure types.

Consistent with the population rate of intellectual disability among those with epilepsy (OR 27.5, p<0.001), we found that epilepsy and intelligence had a negative genetic correlation, and that genetic liability to higher intelligence was associated with a lower risk of epilepsy. Our findings are consistent with a shared genetic aetiology of intellectual disability and epilepsy. However, it is important to acknowledge that intelligence and intellectual disability are two distinct phenotypes. We were only able to perform MR and genetic correlation analyses of intelligence because of the absence of GWAS data for intellectual disabilities.

Intriguingly, using MR, we were unable to confirm the common assumption that epilepsy is a ‘cause’ of an increased risk of psychiatric disorders, or that these disorders are a ‘cause’ for an increased risk of epilepsy.12 This finding should be interpreted in light of the MR assumptions being met and the possibility of the role of rare genetic variations not captured within the GWAS data which comprise common variants.

However, this finding may also indicate the importance of environmental factors in the origin of psychiatric comorbidities in epilepsy. Future work could help identify the causal role of such factors (eg, prenatal or perinatal events unique to the individual, or discrimination, stigma and life events during the individual’s lifetime, which are common risk factors for several neurological and psychiatric phenotypes2 33 34), ideally using causal triangulation across study designs with orthogonal sources of bias.

Limitations

An important limitation to note is that existing epilepsy and psychiatric conditions GWAS are not sufficiently large, potentially leading to reduced power in identifying genetic correlations, and necessitating an inclusive threshold (p<5×10⁻⁶) to identify instruments for MR analyses. Although this threshold has been widely employed in previous studies, the possibility of reduced specificity in instrument selection, leading to weak instrument bias, cannot be ruled out. Analyses like ours, repeated using a larger GWAS, when available, may provide stronger evidence. Furthermore, rare genetic variants are known to play an important role in the aetiology of epilepsy,7 and their contribution to psychiatric comorbidities in epilepsy cannot be excluded. Although rare genetic variants may not be sufficiently common in the population to explain the high rates of psychiatric conditions reported in observational studies including our nationwide study, they may still be common among individuals with epilepsy. Nonetheless, future studies will help clarify the role of rare genetic variants in the overlap between psychiatric conditions and epilepsy. We were also unable to perform any age-stratified or sex-stratified cross-trait genetic analysis; however, this could be a target for future analysis if sex-stratified GWAS data were available. Another limitation of the Swedish data analysis is that our study relied on the accuracy of the epilepsy and psychiatric diagnoses recorded in the registers. Validation studies of epilepsy35 and a range of psychiatric and neurodevelopmental conditions suggest a high validity of these recorded diagnoses. However, the possibility of measurement bias in the data cannot be excluded. This limitation is common across large-scale epidemiological studies. Further large-scale efforts to estimate the burden of psychiatric conditions among people with epilepsy are warranted, with particular emphasis on including diverse and heterogeneous populations. Future work based on high-resolution phenotyping of smaller representative cohorts will shed further light on the complexities of the relationships studied.

Implications

Our study confirms the notable prevalence of psychiatric comorbidities among individuals with epilepsy and provides mechanistic insights into this co-occurrence. These findings underscore the importance of comprehensive patient care that acknowledges and addresses associated conditions during epilepsy diagnosis and treatment. This is particularly important, as psychiatric comorbidities notably impact the prognosis of epilepsy, influencing mortality and life expectancy.36 37 While our study revealed some genetic correlations suggesting a potential genetic contribution to comorbidities, it did not strongly support bidirectional relationships. Consequently, our study emphasises the importance of exploring alternative factors, including environmental influences and rare genetic variations, to unravel the origins of psychiatric comorbidities in epilepsy. Understanding the underlying causes of comorbidities is crucial, as it can guide treatment decisions and offer more effective strategies for managing both epilepsy and associated psychiatric conditions. This may enhance the overall quality of care for individuals affected by epilepsy.

In conclusion, psychiatric comorbidities are common in patients with epilepsy. Genetic correlations may partly explain some comorbidities; however, little evidence exists of a bidirectional relationship between the genetic liability of epilepsy and psychiatric conditions. These findings highlight the need to understand the role of environmental factors or rare genetic variations in the origins of psychiatric comorbidities in epilepsy.