Danish Population Based Study of Familial Epilepsy and Childhood Cancer

Objective: Results from studies investigating the association between maternal or child epilepsy, use of anticonvulsants in pregnancy, and childhood cancer are inconsistent and at times contradictory. Methods: Linking Danish national databases, we obtained epilepsy and childhood cancer diagnoses, and anticonvulsant use data. We estimated adjusted odds ratios of all or specific childhood cancers in relation to maternal or child epilepsy and anticonvulsant therapies using conditional logistic regression. Results: Maternal epilepsy was positively associated with all childhood cancers in offspring, specifically, with acute lymphoblastic leukemia (Odds Ratio (OR) = 1.68, 95% Confidence Interval (CI) = 1.16, 2.43) and Wilms tumor (OR = 2.13, 95%CI = 0.97, 4.68). When considering maternal ever (lifetime) ingestion of anticonvulsants, a positive association was found with all cancers (OR = 1.15, 95%CI = 1.01, 1.31), and central nervous system tumors (OR = 1.32, 95%CI = 1.03, 1.69) as well as neuroblastoma (OR = 2.05, 95%CI = 1.29, 3.28) among offspring. Maternal anticonvulsant use before or during the index pregnancy was related to CNS tumors in offspring (OR = 1.78, 95%CI = 0.99, 3.21), however the confidence interval included the null. Significance: Maternal use of certain anticonvulsant medications may be a risk factor for cancer in offspring. Medical providers may need to consider what type of treatments to prescribe to pregnant mothers with epilepsy.


Introduction
Approximately 400,000 children between the ages of 0-14 develop cancer globally each year [1,2]. The most common types of childhood cancers are leukemias, neuroblastomas, brain tumors, and Wilms' tumors. [3] Even though this only accounts for less than 1 percent of the total annual global cancer incidence, [4] cancer remains the second leading cause of childhood morbidity in developed countries in this age group. [5] Suspected causes of childhood cancer include medication use, agricultural or industrial chemicals, and tobacco smoke (and hepatoblastoma) in pregnancy as well as hereditary and acquired mutations. [6][7][8][9] Worldwide approximately 7.2 million women of childbearing age have epilepsy. [10] One study found that approximately 70 percent of women with epilepsy reported adhering to their prescribed epilepsy medications during pregnancy. [11] Yet, some reports found that 40-60 percent of pregnant women with epilepsy reported having been non-compliant with their medications. [12,13] Mothers with epilepsy are slightly more likely to undergo Cesarean section, deliver preterm, and have offspring with low birth weight even when not using antiepileptic drugs. [14,15] Anticonvulsants can be transmitted to the fetus transplacentally, [16] and research has established strong teratogenic effects from pregnancy exposure. [14,[17][18][19] In addition, adults with epilepsy may be at an increased cancer risk. A recent cohort study in Denmark found moderate increases in total cancer incidence and speci cally increases in liver, mouth and throat, central nervous system, and respiratory cancers among adults with epilepsy who took anticonvulsants. [20] Yet, despite the evidence for both gestational teratogenic and adult carcinogenic impacts, there have been relatively few studies evaluating whether offspring of parents with epilepsy are at an increased cancer risk, likely due to the relative rarity of both parental epilepsy and cancer in children. [21,22] Previous studies have suggested a relationship between particular types of anticonvulsants, such as barbiturates and phenytoin, and the development of various cancers in animal models. [23,24] With regards to childhood cancer, elevated risks were reported for acute lymphoblastic lymphoma (ALL) with non-speci c maternal intake of anti-epileptic medications in pregnancy [25] as well as for brain tumors, although some studies had small sample sizes and reported wide con dence intervals that included the null. [26][27][28] A link between childhood lymphoma and prior exposure to phenytoin in utero has also been proposed [29], but several studies were unable to corroborate these associations. [30][31][32][33] A study that found an association with brain tumors in children taking anticonvulsants or having a history of epileptic seizures, may have been affected by reverse causation if the tumor was causing seizures prior to diagnosis. [33] This study attempts to determine whether children are at a higher risk of developing cancers from exposure to anticonvulsant medications in utero. We also evaluated the possibility that the mother's epilepsy plays an independent etiologic role in the development of cancer in their offspring.

Materials And Methods
This study was conducted in Denmark via a linkage of large national databases and has been described in detail previously [34]. Epilepsy diagnosis (Supplemental Table 1 We additionally linked our data to the Medical Births Register for information on gestational characteristics, and to the Central Population Register that links parents to their children and provides demographic information. From the Central Population Register we randomly selected controls that we matched to cases by sex and birth date (25:1 matching rate). In order to use information from registries on pregnancy, we restricted the study to children whose mothers were pregnant and gave birth while in Denmark.
Although our primary interest was to examine cancer risks in children in relation to maternal epilepsy or anticonvulsant use during pregnancy, we considered that epilepsy and childhood cancer may have shared genetic risk factors and thus also needed to account for paternal history of epilepsy. [35] We therefore additionally examined childhood cancer in relation to paternal epilepsy. Due to differences in data availability for diseases and treatments, we generated two analytical datasets controls. Pregnancies that were likely not viable (gestational age < = 20 weeks or birth weight < 500g, n = 17) were excluded from these analyses.
Dataset #2 is based on the pharmaceutical registry and births in 1995-2014 and cancer diagnoses in 1995-2016 where we linked study subjects to the National Pharmaceutical Registry (established in 1996 with pilot exposure data collected and available beginning in 1995), which provides information on every prescription lled at a pharmacy in Denmark. For the analysis based on the second dataset, we ascertained epilepsy diagnoses from the National Patient Register and additionally via identi cation of individuals prescribed medications indicated solely for epilepsy (Supplemental Table 1). The National Pharmaceutical Registry lists the dates of all medications lled, and information on medication and dose, using Anatomical Therapeutic Chemical Classi cation (ATC) codes. As this sample was created later, matched controls were also re-sampled from the Danish population. These analyses included 2,521 cases and 63,014 controls. Pregnancies likely not viable (gestational age < = 20 weeks or birth weight < 500g, n = 11) were excluded from analyses.

Statistical Analyses:
We estimated the crude odds ratios of cancer diagnosed in the offspring using conditional logistic regression to account for matching and report results for all cancer types with at least ve exposed cases. Potential confounding variables considered for adjustment were suggested by the literature [9,32,[36][37][38][39]. In our nal model, we adjusted for maternal age (continuous), maternal history of migraines (ever/never), and child's epilepsy before cancer diagnosis (ever/never). We considered additional adjustment for urbanicity of residence, maternal smoking, maternal diabetes, familial socioeconomic status, and maternal asthma but did not include these factors in the nal regression model because they did not change the estimates by 10% or more. [40] Earlier studies suggested an increased risk of lymphoma in mice exposed to phenytoin [24] and brain tumors in children whose mothers used barbiturates during pregnancy [41]. We had su cient statistical power to examine childhood cancer risk for barbiturates but not phenytoin.
Since the mechanisms and presenting symptoms as well as treatment for epileptic seizures vary by the epileptogenic regions of the brain, [42] we separately examined the risk of cancer in offspring of mothers diagnosed with primary generalized epilepsy and focal epilepsy. Epilepsy subtypes were identi ed using ICD codes listed in Christensen [43].
Seizures may be a presenting symptom of various brain tumors, [39] thus our analyses of children's epilepsy in relation to cancer may be at risk of reverse causality. We therefore did not examine childhood epilepsy in relation to cancer diagnosis.
All statistical analyses were conducted using SAS, Version 9.4 (SAS Institute Inc., Cary, NC, USA). Human subject approvals were obtained from the University of California Los Angeles, and the Danish Data Protection Board.

Dataset #1
We present demographic and health data for the study population in Supplemental Table 2. Compared to mothers without epilepsy, mothers with epilepsy tended to be younger and were more likely to live in rural areas. A larger proportion of those who were diagnosed with epilepsy reported having migraines, diabetes, asthma, and smoked at the rst prenatal visit compared to those who did not have epilepsy. Table 1 presents the association between childhood cancer diagnoses in relation to maternal epilepsy diagnosis. Maternal epilepsy was positively associated with all cancers as well as acute lymphoblastic leukemia (ALL) and Wilms tumor.

Dataset #2
Compared to mothers without epilepsy, the majority of mothers diagnosed with epilepsy in this subsample were younger at the time of the offspring's birth while the largest proportion of mothers without epilepsy were between 30-34 years of age at that time (Supplemental Table 5). A higher proportion of this subgroup of mothers with epilepsy also resided in rural areas, had migraines, diabetes, and asthma prior to pregnancy compared to mothers without epilepsy. Table 2 presents the association between maternal anti-epileptic medication use and childhood cancer. When considering maternal ever (lifetime) use of anticonvulsants, a positive association was found with all cancers (OR = 1.15, 95%CI = 1.01, 1.31), and central nervous system tumors (OR = 1.32, 95%CI = 1.03, 1.69) as well as neuroblastoma (OR = 2.05, 95%CI = 1.29, 3.28). When we examined maternal intake before or during the index pregnancy, a positive association was suggested for central nervous system tumors, but con dence intervals contained the null (OR = 1.78, 95%CI = 0.99, 3.21). Limiting to barbiturates, there were only 5 exposed childhood cancer cases, and the estimate was uninformative for all childhood cancers. (OR = 1.54, 95% CI 0.62, 3.80).

Discussion
In this population-based study of childhood cancer, we found that maternal epilepsy -irrespective of anticonvulsant medication use -was positively associated with acute lymphoblastic leukemia and Wilms tumors in offspring. Additionally, maternal anticonvulsant use was linked to CNS tumors and neuroblastoma in offspring. In contrast, we did not observe associations with paternal epilepsy. The lack of association with paternal epilepsy would suggest that the increased risk of cancer is likely maternally in uenced, either directly from maternal use of anticonvulsant medications, or indirectly via negative impacts of maternal epilepsy on pregnancy.
Such negative impacts on pregnancy have been the focus of some past studies targeting clinician guidelines. A recent systematic review and meta-analysis found increased risks of spontaneous miscarriage, ante and post-partum hemorrhage, caesarean section, hypertensive disorder, and induction of labor among mothers with epilepsy, primarily among those who took anticonvulsant medications [44]. Another study found elevated risks of anemia, uterine broids, ovarian cysts, and thyroid dysfunction among pregnant epileptic mothers [10]. Hypothyroidism has recently been linked to childhood cancer [45].
However, the degree to which the mother's epileptic condition itself or the medications used may be responsible for developing childhood cancer can be di cult to elucidate. In our study, we cannot rule out that the association noted between maternal epilepsy and childhood cancer is due to confounding by indication. Previous literature has suggested that epileptic mothers, independent of anticonvulsant medications, are at a slightly higher risk of having low birthweight or preterm offspring (OR's: 1.1-1.3), [14,15] which has been linked to the development of childhood cancers such as gliomas and retinoblastomas [46,47]. This seems to support that the mother's condition related to adverse birth outcomes possibly contribute to the carcinogenic process regardless of medication use, but our results are inconclusive. When we examined women with epilepsy strati ed by medication use, we observed lower point estimates among medication users compared to non-users and con dence intervals included the null (Supplemental Table 6). This may be due to the fact that only 7 case women with epilepsy were reported with a registered ATC code. This is contrary to previous reports indicating high adherence to medications prescribed during pregnancy.
[48] Yet, our observations were comparable to another Danish population-based study which found similar rates of anticonvulsant consumption during pregnancy. [49] We were not able to conclude whether the observed increased risk of cancer in offspring is a result of maternal medication use or epilepsy diagnosis, as both estimates have wide and overlapping con dence intervals.
The teratogenic impacts of anticonvulsant therapies are well recognized. It has been established that some anticonvulsant medications used to treat epilepsy, particularly barbiturates and phenytoin, are potent carcinogens in animal models, [23,24] and these drugs have been observed in human studies passing through the transplacental barrier to the fetus.
[16] A study of anticonvulsant medications used in pregnancy found increased risks for neural tube defects, oral clefts, and heart defects among offspring, but conclusions were limited to valproic acid and carbamazepine consumed during the rst trimester. [50] Another study found similar malformations, including spina bi da and trisomy 18, [51] in offspring exposed to anticonvulsants in utero, and yet another found minor craniofacial and digital malformations and concluded that the mother's condition may have played a role. [52] Several of these birth defects are also related to childhood cancers. [53,54] We also considered examining the relationship between childhood epilepsy and cancers, and we observed elevated estimates for central nervous system tumors, including gliomas. This was expected as it is well known that seizures may be a presenting symptom of such brain tumors, [39] yet a 2-year lagged analysis still produced elevated estimates. This suggests multiple possibilities. Slow growing brain tumors may remain undiagnosed beyond our lagging cut-off, and therefore these seizures may in fact be a product of the cancer itself. [27] Conversely, the medications used to treat them could also be involved in the carcinogenic process. There are past studies that have found elevated cancer risks among children exposed to anticonvulsants, notably barbiturates, either in utero or in early childhood. [25-29, 33, 41] However, limited sample size made it impossible to conduct a lagged analysis beyond 2 years, and delays in tumor diagnosis would require far longer lags. [55] Seizures are also a common symptom of individuals with gliomas.
[56] As such, we were unable to address the possibility of reverse causation.
We did not nd any association for maternal barbiturate use. Our point estimate was lower than what has been seen in other studies, however our analysis was limited by the small number of exposed children (5 exposed cases and 80 exposed controls), likely due to decreasing use of barbiturates in high income countries. [57] Our study design did not allow us to con rm that all mothers with a registered ATC code for an anticonvulsant purchase actually ingested the medication. In previous studies, noncompliance in pregnancy ranged from one-fth to two-third for antiepileptic drug users; still others continued treatment but switched to safer alternatives. [14,48,58,59] However, four-fth of mothers with epilepsy in our study did not purchase anticonvulsant medications during pregnancy. Yet, noncompliance would have been non-differential as none of the mothers could have predicted the child's cancer and it would have biased our results towards the null, thus the increased point estimates are notable.
Another factor to consider from our study is that some of these older drugs studied in the past, such as phenobarbital, may be facing increased prescription hesitancy from medical providers in high income countries. First generation anti-epileptic drugs were not subject to the same preclinical carcinogenicity testing required for drugs developed in the last thirty-eight years.
[60] This shift towards more current antiepileptic medications may decrease the relevance of our study for those regions, however phenobarbital is still widely prescribed as a treatment for epilepsy in most of low-and middle-income countries, where 80% of people with epilepsy live. [61] In summary, this study provides some indication that certain anticonvulsant medications may be carcinogenic to children exposed in utero. While our data does not seem to show an association between maternal epilepsy diagnosis when anticonvulsant medications are not used, maternal epilepsy has been associated with conditions linked to childhood cancer. Untreated epilepsy can have perinatal risks, thus discontinuation of anticonvulsant use may not be the safest option for patients. However, as certain anticonvulsant medications can irreparably harm the fetus, these conditions and treatments require consideration and further study.

Competing Interests
There are no con icts of interest to declare. We con rm that we have read the Journal's position on issues involved in ethical publication and a rm that this report is consistent with those guidelines.

Author Contributions
Corbin Platamone served as principal investigator during this study and drafted the manuscript, Xiwen Huang performed data analyses, Rajarshi Mazumder served as coinvestigator and epilepsy neurology specialist, Beate Ritz served as co-investigator and lab director, Jorn Olsen and Johnni Hansen provided the Danish government data, Chai Saechao provided an anti-convulsant medication list, and Julia Heck served as co-investigator and preceptor.

Ethics Approval
Human subject approvals were obtained from the University of California Los Angeles, and the Danish Data Protection Board. ISPRC approval granted on 06/26/2022 as it is an observational study. IRB #:  Consent to Participate Human subjects consent was approved by the Danish Data Protection Board as part of this observational study.

Consent to Publish
The authors con rm that no individual level data is present in this study. AOR values generated by simultaneous entry of covariates into a conditional logistic regression model a*: Adjusted for maternal age and maternal history of migraine b*: Adjusted for maternal age, maternal history of migraine, child's epilepsy before cancer diagnosis, and maternal history of cancer