Article Text


Assessing epidemiological evidence for the teratogenic effects of anticonvulsant medications
  1. H Dolk1,
  2. P McElhatton2
  1. 1Department of Epidemiology and Health Services Research, Faculty of Life and Health Sciences, University of Ulster, Shore Road, Newtownabbey BT37 0QB, UK
  2. 2National Teratology Information Service (NTIS), Regional Drug and Therapeutics Centre, Wolfson Unit, Claremont Place, Newcastle upon Tyne NE2 4HH, UK
  1. Correspondence to:
 Professor H Dolk, Department of Epidemiology and Health Services Research, Faculty of Life and Health Sciences, University of Ulster, Shore Road, Newtownabbey BT37 0QB, UK;

Statistics from

Antiepileptic medication in pregnancy

Epidemiological studies of pregnancy outcomes of women with epilepsy are important, primarily to inform choice of treatment options in clinical care of pregnant women and to inform counselling, the need for prenatal and postnatal screening, and the need for early medical, behavioural, or educational intervention.

In this issue (251), Dean et al report on a long term follow up of the health and neurodevelopment of children exposed to antiepileptic drugs before birth. They did this by consulting obstetric records to identify mothers with epilepsy who delivered in a single maternity unit between 1976 and 2000, inviting mothers to participate in the study by letter via their primary care physicians, conducting an interview with participant mothers, and consulting a variety of medical and child health records. The oldest children in the study were 15 years of age.

There is a fairly solid consensus that epilepsy in pregnancy treated with anticonvulsants is associated with an overall two- to three-fold increased risk of congenital malformation compared to the general population. The first challenge of the epidemiological approach is to mount sufficiently large studies. Since epilepsy is fortunately uncommon (6 per 1000 pregnancies or less), and since major congenital malformations are uncommon (2-3% of all births), every 10 000 pregnancies will yield only three or four cases of major congenital malformation born to epileptic women. Kallen,1 reviewing the results of 24 studies, indicated that nearly half of these studies included fewer than 150 infants born to epileptic mothers and did not have the statistical power to reveal a two- to three-fold raised risk of major malformation. It is generally necessary therefore to make a combined assessment of existing studies, particularly when assessing risks of specific anomalies. Schardein2 reviewed 17 cohort studies including a total of 2168 infants of women receiving anticonvulsant treatment to calculate absolute risks of 1.8% for congenital heart disease and 1.7% for facial clefts, the two types of congenital anomaly most commonly associated with maternal epilepsy. Similarly, very large studies or pooling of studies are needed to investigate differences in risk between different types of drug. The study of Dean et al makes a further contribution to the pool of available data on the risk of specific anomalies and their association with specific drugs.

A smaller body of research has tackled the longer term neurodevelopmental outcome of exposed children, these studies having the added problem of achieving follow up of the cohort of children identified and obtaining more complex neurodevelopmental data. In the study of Dean et al, of 411 women identified from obstetric records as having epilepsy, one quarter had moved away and were therefore not available for follow up, and two fifths of those still in the area did not participate in the study, leaving 149 participants. This rather low follow up rate reduced the statistical power of the study and also raised the question of bias: were those who moved away or who did not participate more or less likely to have experienced adverse pregnancy outcomes than the participants? Developmental delay was defined as having any one or more of a need for speech therapy in child health records, not sitting by 10 months, not walking by 18 months, or special educational needs at school. This record based assessment contrasts with the methodology available to prospective cohort studies, where standardised neurodevelopmental assessments can be made. For example, Gaily et al3 assessed their exposed and control children at the age of 5 years with standardised psychological and neurodevelopmental evaluations, and Wide et al4 applied a standard developmental test to 9 month old infants. Such studies also have the opportunity to carry out testing blind to the disease or treatment status of the mother.

Why do some epidemiological studies find evidence of an increased risk of developmental delay or malformations and others not? The most superficial reason is alluded to above, that some studies simply lack statistical power to find differences in risk between the exposed cohort and the comparison group, particularly for rarer/newer drugs or rarer health outcomes. In addition, owing to the very wide range of drugs and outcomes studied, multiple statistical testing will throw up a certain number of “false positive” results where the difference between exposed and unexposed groups is simply the result of chance. At the usual significance level of 5%, for example, 5% of comparisons will find a statistically significant difference where there is no true underlying difference in risk. Differences in outcome between different drugs can similarly be chance differences. Thus, for example, Dean et al's finding that developmental delay is associated with all the major drug types but not phenobarbitone should be interpreted with caution until assessed for its consistency with other studies.

There are also more informative reasons why results of studies may conflict, relating to study design. Since studies are observational rather than experimental, they are open to various sources of bias. Bias can result from selection of a comparison group which differs from the cohort of epileptic women in other respects than disease or treatment status. For example, one study found epileptic mothers to be younger at delivery of their children and more likely to be of lower social class5 than the population of the same region, emphasising the need to control or match for these characteristics. Bias may also result from differing levels of and reasons for non-participation in the exposed and comparison group. Apparently conflicting study results can arise through differences in the measures of developmental delay (see above), behavioural disorder, fetal anticonvulsant syndrome,6, 7 or minor anomalies.4, 7 Studies may differ in how conditions are grouped together for analysis or inclusion criteria for study and control children, such as restricting study to children with normal intelligence3 or to cases without family history. Where the general population is used as a comparison, it can be difficult to find truly comparable statistics. For example, population studies of congenital malformation prevalence may ascertain cases of malformation more or less completely than a study of a cohort of epileptic women, especially with regard to anomalies diagnosed after the neonatal period. Ten of the 36 children with major malformations in the study of Dean et al had an inguinal hernia, a condition which is on the exclusion list for many congenital anomaly registers and therefore does not appear in general statistics.8 Population statistics concerning later developmental outcomes tend not to be available or are based on poorly standardised assessments. Thus, true differences can be masked and false differences observed. Dean et al chose not to compare their findings with a cohort of non-epileptic women or the general population, but to identify a cohort of untreated epileptic women as the comparison group.

Finally, results of studies may differ because of true differences between the populations of the study. Women with epilepsy in different populations may differ with regard to the proportion following optimal treatment regimens or in genetic susceptibility to drug effects.9 If the study is not truly population based, there may be a tendency to enter into the study only women with severe epilepsy who attend particular types of clinic.

In the treatment of the pregnant epileptic woman, the potential harm to both mother and fetus of the epilepsy itself must be weighed against the potential harm of the drug treatments. Maternal epilepsy may affect fetal and child health through the effects of seizures during pregnancy, or through the genetic background associated with epilepsy, or through differences in maternal behaviour or family circumstance, pre- or postnatally. One problem in studies which set out to distinguish between the effects of the drugs and of the epilepsy itself is that few epileptic women do not have treatment and these women usually differ from those receiving treatment in other ways, including the characteristics of their epilepsy. In the study of Dean et al, for example, the non-exposed comparison cohort were taken to be 38 sibs of exposed cases who were not themselves exposed in utero, either because they were born before epilepsy developed or because the mother took no treatment. Apart from being unrepresentative, this comparison group is also very small, reducing the statistical power of the study to find differences between the exposed and unexposed cohort. The average age of the unexposed children was higher than that of children exposed to most drug treatments, posing potential problems in assuring that comparable medical and neurodevelopmental information was available throughout the study period.

To distinguish drug effects from disease effects, there are two main epidemiological arguments, as discussed by Dean et al. The first is to look for specificity of individual drug malformation associations, that is, if a type of malformation is associated with only one type of drug, it is more likely to be a drug than a disease effect. For example, a specific association between spina bifida and valproic acid and perhaps carbamazepine is suggested by some published reports.10, 11 However, specific indications for different drugs may confound this assessment of specificity. For example, phenytoin has been used for more severe epilepsy. The second argument is to look for a relationship between risk of adverse pregnancy outcome and dose or monoversus polytherapy, again an indicator of the drug itself as causal agent, although the severity of the epilepsy may again confound this association. A fairly consistent finding, supported by the results of Dean et al, is that the risk of polytherapy is greater than monotherapy, and this has led to current guidelines advising monotherapy during pregnancy. International or multicentric studies can be a useful tool to try to dissociate drug and disease, because of the wide variations existing in treatment practice.12, 13 Studies over long time periods can also track changes in treatment independently of epilepsy.14 The study of Dean et al covered a 15 year time period, differences in mean age of offspring of different drug treatments reflecting changes in drug treatment, also probably tending to dissociate drug and disease characteristics. On the other hand, the overall results relate to drug regimens that are in part no longer current, for example, phenobarbitone and phenytoin are now used less frequently.

A third way of distinguishing drug from disease effects is to introduce a further comparison group, those who receive antiepileptics for conditions other than epilepsy.7, 15 Finally, studies of children of fathers with epilepsy can shed light on the importance of the genetic background in causing an excess risk of malformations.1

It is perhaps surprising that there is not more information available to evaluate accurately the health risks of maternal epilepsy and antiepileptic treatment for the fetus and child. In the study of Dean et al, nearly one third of the exposed cohort of children had either a major malformation or developmental delay, emphasising the importance of the problem. Further research with follow ups into childhood is clearly needed. It could even be argued that collecting basic information on long term outcome should be part of clinical care. The way forward should involve multicentric studies of large numbers of exposed pregnancies, with suitable control groups of unexposed (non-epileptic and epileptic non-treated) pregnancies, standardised assessments of outcome, and detailed information regarding seizure history and drug treatment. Single centre studies should design and report on their studies with a view to the need for meta-analysis or combined assessment of results across studies. Furthermore, as Dean et al conclude, it is likely that future research will want to focus on why, while the majority of children of epileptic women on antiepileptic drugs experience “normal” health and development, some experience significant problems, and thus how we can predict more precisely those at risk and target therapies more effectively.9

While further research is clearly needed, there is also evidence that existing knowledge is not being effectively incorporated into health care in all communities.16 In the UK, national guidelines are needed for treatment of epilepsy in women of child bearing age. These should address a range of issues, including whose responsibility it is to speak to teenagers/young women about potential problems should they become pregnant.

Antiepileptic medication in pregnancy


View Abstract

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Linked Articles