Pregnancy registers (or registries , as they are sometimes termed) are intended to prospectively collect information about the relative teratogenicities of antiepileptic drugs. They are in essence observational studies that do not interfere with the actual existing managements of individual participants, so that no issue of unethical or doubtfully ethical experimentation on humans arises. They provide the best available practical and ethical alternative to the randomised, placebo-controlled studies which would in other circumstances be preferable to obtain strictly evidence-based results, but which would probably be judged ethically unacceptable if proposed in relation to pregnant women.

The register model of prospectively collecting pregnancy-related information that was established initially by Holmes in the USA was followed by the establishment of reasonably similar registers in several countries. Details of some of these registers have subsequently been reported (Beghi et al. 2001; Vajda et al. 2007). In 1999, Ettore Beghi convened a meeting that established collaboration between registers which had sufficiently similar features to enable their joining an international register called EURAP . Beghi (2012) summarised the situation concerning pregnancy registers as follows:

Pregnancy registries have been activated by collaborative groups of physicians in Europe (EURAP ), North America (NAREP), Australia and India (the latter two recently merged into EURAP), to enroll a large number of exposed women to be monitored prospectively with standardized methods, and by three pharmaceutical companies marketing lamotrigine , gabapentin and vigabatrin , as part of their post-marketing surveillance…… the implementation of a common database with information from the existing registries may provide valuable information in a shorter time period.

The European International Registry (EURAP ) was established following the meeting convened by Beghi (above). After 15 years of existence, this registry involves over 45 countries worldwide. Collaboration between it and additional registers continues to be explored. Individual collaborating registers may still choose to publish their own data, derived from a smaller database, but also feed the information into EURAP.

As one example of a register ’s content, the protocol for inclusion in the Australian Register of Antiepileptic Drugs in Pregnancy involves interviews with women with informed consent at their time of enrolment, at 28 weeks of pregnancy, soon after delivery and at 12 months postnatally. Information on general medical, social, epileptic seizure and treatment history, past pregnancies and outcomes, non-epilepsy-related medications and events is recorded, and access to patients’ medical records is obtained. Data are kept in a computerised database for subsequent analysis.

Other registers include the North American Epilepsy and Pregnancy Registry , the UK Epilepsy and Pregnancy Register , the Swedish Register, the Finnish National Drug Prescription Registry and the Danish Register. The main contemporary pharmaceutical company registers are the Prospective International Lamotrigine Registry and the Pregnancy Registry maintained by UCB to monitor levetiracetam .

There are differences between the registers. EURAP contains no internal control group of untreated women with epilepsy and endeavours to compare drugs and their effects on the foetus with each other. From the outset, the Australian Register has included antiepileptic drug-untreated women with epilepsy, comprising about ten per cent of the total, and also has attempted to recruit women receiving antiepileptic drugs for non-epileptic indications, such as pain or bipolar illness. The US Registry employs historical control groups. Although no single specific control group is ideal, the availability of sufficient numbers of women with untreated epilepsy probably represents the most appropriate comparator.

In the earlier days of the existence of such registers, Meador (2008a) commented that

The registries have different data collection timing (e.g. malformations at birth versus at 1 year), some lack of information about the mothers as well as a lack of follow up, which restricts usefulness of registry data. In addition, each of the individual registries has weaknesses ranging from missing data on key variables to low numbers.

In more recently developed registers, some of these criticisms from an earlier time have been addressed, at least to an extent. Even so, such registers, in which enrolment necessarily is voluntary, are likely to capture only a minority of the women with epilepsy in the community and, unless they draw on very large populations, probably will not accumulate sufficient pregnancy numbers for useful analysis in data collection periods of less than several years. Over such periods, therapeutic practice may change, possibly confounding interpretations. As well, women with epilepsy who are prepared to enrol their pregnancies may not necessarily be representative of women with epilepsy in the wider community. For instance, despite its best efforts over its period of existence, the Australian Pregnancy Register appears to have been able to collect only some 8.7 % of the pregnancies calculated to have occurred nationwide in women with epilepsy (Vajda et al. 2014). Registers are likely to prove more suitable for detecting differences between women with treated and untreated epilepsies in pregnancy (if sufficient numbers of the latter become available) and in comparing the effects of individual epileptic drugs, than in comparing malformation rates between women with epilepsy and those in other pregnant women in the general community.

The approaches to obtaining sufficient pregnancy numbers mentioned above have other limitations. Because of their information-collecting and information-recording methods, as already noted, hospital records may not always contain relevant data from the foetal malformation point of view, official institutional databases tend to sacrifice data quality to achieve inclusiveness, while registers tend to sacrifice comprehensiveness for the advantage of relevance to the issue at hand.

One issue in relation to the various registers that does not seem to have received much discussion is the possibility that the material some of them contain may have been derived from pregnancies that have already been the subject of literature reports. Conceivably the same individual pregnancy may have been included more than once in the literature, without the replication of data being identifiable. This may lead to the impression that more information is available than is really the case.

Different studies have employed a variety of comparison data sets in assessing the significances of malformation rate values, e.g. known or expected malformation rates for the whole population or for particular and preferably matched population subsets, or malformation rates in siblings that were not exposed to antiepileptic drugs in utero (Dolk and McElhattan 2002). As already mentioned, probably the most satisfactory comparator would be the malformation rate in the offspring of women with untreated epilepsy . Unfortunately, the number of untreated pregnancies in women with epilepsy has usually proved to be substantially smaller than the number of antiepileptic drug-treated ones, generally about 10 % or 15 % of their number, though in the early series of Janz and Fuchs (1964), approximately one-third of all the pregnancies studied were not exposed to antiepileptic drugs. As noted above, the corresponding figure for the most recent analysis of the Australian Pregnancy Register was 8.7 % (Vajda et al. 2014). Little seems to have been published investigating whether such untreated pregnancies differ in important aspects from antiepileptic drug-treated ones, apart from the material in the paper of Vajda et al. (2008). This paper analysed the data from 68 pregnancies that were untreated in at least first trimester compared with the data from the remaining 709 simultaneously collected but drug-treated pregnancies in women with epilepsy then included in the Australian Pregnancy Register . More than 50 parameters were studied. The only statistically significant differences were that the untreated pregnancy group contained higher proportions of first pregnancies, pregnancies with seizures in the pre-pregnancy year, pregnancies that had been referred to the Register by neurologists and babies with slightly higher head circumferences at birth and slightly higher APGAR scores five minutes after delivery. There also were fewer pregnancies with postpartum seizures. Foetal malformations tended to be less frequent in the drug-unexposed pregnancies. Apart from the matter of taking or not taking antiepileptic drug treatment, the increased proportion of referrals from neurologists in the untreated group seemed to be the only difference that might suggest that the groups compared did not come from the same population. This difference in referral source raised the possibility that the epilepsies in the untreated group may have been managed by more skilled professionals. There were no significant differences in respect of family histories of birth defects in previous generations or in older siblings or pregnancies culminating in stillbirth . There also were no differences in preconception folate use or folate use in pregnancy, in illness in pregnancy or in smoking or alcohol intake in early pregnancy. Focal epilepsies were a little less frequent in the untreated group and generalised epilepsies more frequent, but the differences were not statistically significant. Seizure occurrence rates in earlier pregnancy were similar in the two groups. Overall, the untreated control and treated pregnancy groups may not have proved to be perfectly matched, but apart from the matter of antiepileptic drug intake, there did not seem to be any obvious difference between them that would have been expected to produce sizeably different foetal malformation rates . In 62 % of the pregnancies in the untreated group, antiepileptic drugs that had been taken previously had been withdrawn within a few months of the beginning of pregnancy. This suggests that the women who had done this had planned to be drug free in at least earlier pregnancy. There is published evidence that some women with epilepsy in the United Kingdom choose to adopt such a policy (Man et al. 2012).

There is also the related question of the denominator against which malformation rates are expressed. Often it has been live births, but doing this omits intrauterine deaths and stillbirths which may be due to antiepileptic drugs and also excludes therapeutic abortions which have been carried out because of detected major foetal malformation. The ideal denominator, conceptions, is impractical because of the difficulty in obtaining accurate data for spontaneous abortions , especially ones that occur early in pregnancy. Malformation rates have sometimes been expressed relative to the number of pregnancies studied, raising the question of how the data from multiple offspring from a single pregnancy have been handled. If the malformation rates are expressed relative to numbers of live offspring, pregnancies yielding multiple offspring will be counted more than once. Some of the differences in foetal malformation rates , and in types of malformation, recorded in different studies in the literature probably arise from such differences in the denominators used in the rate calculations.

Because of difficulty in agreeing when minor anatomical aberrations or unusual physical appearances would warrant the designation of ‘malformation’, investigators have often restricted themselves to the statistics of major congenital malformations . Even then, the boundaries between what is designated ‘major’ and what is considered ‘minor’ may be somewhat elastic. Mothers of newborn infants may not consider ‘minor’ what a medical observer does, and mothers are often the primary source of register information. In various studies, there have been differences between the time points when the presence or absence of malformations has been determined. The cut-off could be immediately after birth or, more commonly, during the postnatal few days, but sometimes has been after the first three months of postnatal life (e.g. in the North American Registry for some of the studies in which it has been utilised) or, in the case of the Australian Register of Pregnancies in women with epilepsy, at the end of the first year of postnatal life. This time factor may also contribute to differences between the results of the published analyses of data based on different registers. It is much less likely to apply for differences in findings between various studies based on analysis of retrospective data drawn from large institutional or national databanks, though these studies are likely to underestimate true malformation occurrence rates, particular rates of minor malformations, because their data are usually collected close to the time of birth, a stage at which less severe malformations may not have been noticed. Vajda et al. (2007) examined the effect of the time after birth in determining the presence of malformations in the Australian Pregnancy Register material. Approximately 21 % of the malformations that were ultimately included, based on the data available one year after the end of pregnancy, would have escaped inclusion if only assessment during the postnatal period had been used. Moreover, there were also pregnancies which could not be traced one year after birth. Had these been included, and assuming they would have had similar malformation rates to those offspring who had been followed to the end of the postnatal year, as many as 29 % of all offspring with malformations might have been missed in assessments made close to childbirth. Admittedly, a considerable majority of the late recognised malformations in the Australian Register were relatively minor, mainly genitourinary, cardiac and cranial ones, and changes in the appearances of the digits. Battino and Tomson (2007) observed that 40 % more malformations were found with an extended postnatal follow-up than would have been detected at birth.

The larger data sets that have become available in recent times that relate foetal outcomes in women with epilepsy to antiepileptic drug exposure during pregnancy fall into several types, viz. (i) Large scale, often whole pregnant female population data, almost always assembled retrospectively for the relevant age group over a designated period of time. Their data can provide indications of the malformation risk relative to that for the general population, but the roles of antiepileptic drug therapy in the situation may be difficult to define because therapy-related information is deficient. (ii) Usually smaller-sized malformation rate data collections from pregnant women with antiepileptic drug-treated epilepsy compared with matched and simultaneously collected data from pregnant women not taking antiepileptic drugs, at least during the first trimester of pregnancy, the time when foetal malformations probably develop. If the untreated women have epilepsy, this type of material permits the effects of pregnancy itself to be distinguished from the effects of the administered antiepileptic drugs, a matter of some scientific interest. (iii) Smaller-scale studies comparing malformation rates associated with individual antiepileptic drugs. In the latter type of study, the comparator employed usually is the antiepileptic drug associated with the lowest malformation rate, which it is assumed will not be less than the rate for the general population of pregnant women. Such information may identify increase malformation rates associated with a drug other than the comparator one. (iv) Small-scale studies in which malformation rates associated with particular drugs are investigated, and where, if any comparator is available, it is likely to be an assumed malformation rate for the general population (generally around 2–3 %).