Post-Marketing Sources of Data Regarding Reproductive and Developmental Safety of Prenatal Drug Exposures

POST-MARKETING SOURCES OF DATA REGARDING REPRODUCTIVE AND DEVELOPMENTAL SAFETY OF PRENATAL DRUG EXPOSURES

Once a medication is marketed, there are many resources that can provide observational data regard-ing drug safety in pregnancy.

1. Clinician case reports published in the medical literature can delineate a phenotype in an affected infant born to a mother with a specific prenatal medication exposure. However, these reports must be initiated spontaneously and therefore may involve investigator as well as publication bias. Furthermore, without a known denominator of exposed pregnancies that do or do not result in infants with the specific malformation, it is difficult to determine if the reported defect(s) represent an increase over baseline. If the phenotype is sufficiently unique, e.g. the isotretinoin embryopathy (Lammer et al., 1985), then a series of case reports can strongly suggest a hypothesis that can be confirmed using other methods.

2. Centralized adverse event reporting systems (AERSs) can provide a systematic method for the accumulation of case reports from a variety of resources. For example, under the U.S. FDA’s AERS, manufacturers and distributors of FDA-approved pharmaceuticals are mandated to report events such as congenital anomalies as they are reported to them or are published in the scientific literature, in associa-tion with prenatal exposures to their drugs. The FDA receives additional reports through the MedWatch program, an educational and promotional effort, which facilitates spontaneous reporting from health-care providers (Kessler, 1993; Goldman and Kennedy, 1998). And finally, consumers may provide informa-tion to the manufacturer or directly to the FDA.

One advantage of such systems is that reports can be accumulated from a variety of resources in a timely fashion. Although these systems have typi-cally not been fruitful in terms of identifying new human teratogens, once a possible teratogenic expo-sure has been identified through other methods, these systems have been useful resources for exploring the specific characteristics surrounding exposed and affected pregnancies. For example, the angiotensin II converting enzyme (ACE) inhibitor fetopathy, which includes a unique pattern of renal tubular dysplasia and hypocalvaria occurring in association with second or third trimester use of one of the drugs in the ACE inhibitor group, was first reported by a clini-cian (Pryde et al., 1993). However, the frequency of similar or related abnormalities in relation to gesta-tional timing of exposure and dose of the drug was identifiable through review of a series of 110 ACE inhibitor adverse event reports submitted to the FDA through 1999 (Tabacova et al., 2000). Similarly, case reports and cohort studies that identified the increased risk for a variety of neonatal complications with late pregnancy exposure to some antidepressants (Spencer, 1993; Chambers et al., 1996) have been confirmed and classified into possible pathogenetic subtypes using adverse event reporting data (Moses-Kolko et al., 2005; Sanz et al., 2005).

The primary limitations of such systems are simi-lar to those of case reports appearing in the medical literature. Reports must be initiated spontaneously, which may involve bias in the types and number of actual events that are reported as well as an erosion in the motivation to report these events the longer a product is on the market. Spontaneous reporting systems rely on the ‘prepared mind’ to make a link between medication exposure and preg-nancy outcome, a link more likely for outcomes normally rare and extremely severe and less likely for outcomes considered common or with subtle presentation. In addition, adverse event reports do not provide denominator information on the number of exposed, affected or unaffected pregnancies that could be used to develop a birth prevalence rate for purposes of comparison with baseline rates for a spec-ified outcome in the general population.

3. Pregnancy drug exposure registries have been one method of evaluating drug safety in pregnancy dating back to the Swedish lithium registry established in 1962 (Schou et al., 1973). Similar manufacturer-sponsored registries have been successfully completed for fluoxetine (Goldstein, Corbin and Sundell, 1997) and acyclovir (Andrews et al., 1992; Preboth, 2000), whereas several others are presently ongoing. A current listing is available on the U.S. FDA’s Office of Women’s Health website (http://www.fda.gov/ womens/registries/registries.html). All traditional pregnancy registries involve spontaneous reporting of exposed pregnancies. The collection of exposure and outcome data is usually accomplished through the healthcare provider who initiates contact with the registry; however, in some registry designs, exposure and outcome data are collected from the pregnant woman herself. Although pregnancy outcome reports can be collected retrospectively, most current drug registries also identify and follow exposed pregnancies prospectively, i.e. ascertain women during gestation, and collect exposure and other information before the known outcome of that pregnancy. In these cases, the registry may be considered a targeted follow-up study.

The registry approach has many advantages includ-ing timely and centralized ascertainment of exposed pregnancies that can parallel prescribing practices for newly marketed medications. Particularly if the exposure is rare, this may be the most efficient method for collecting pregnancy outcome data as quickly as possible. Industry-sponsored registries can utilize the existing mechanism of pregnancy expo-sures and events that are reported to the sponsor’s medical information departments both nationally and internationally to more efficiently identify poten-tial registry participants (Shields et al., 2004). The registry approach when used to accumulate prospec-tive reports can provide good quality information about the temporal association between exposure and outcome. In addition, prospective registry designs provide a defined denominator of exposed women that facilitates comparisons of congenital anomaly rates to those of a reference group.

These registries generally have the ability to detect a meaningful increase in the overall frequency of major congenital anomalies that are evident at birth relative to the overall birth prevalence of major congenital anomalies in the general population (Koren, Pastuszak and Ito, 1998; White and Andrews, 1999; Shields et al., 2004). Especially for high-risk teratogens such as isotretinoin or thalidomide, such an approach is arguably the most efficient, cost-effective and timely method for identifying such agents quickly. For high-risk teratogens associated with a characteristic and frequently occurring pattern of major congenital anomalies recognizable at birth, only a small number of exposed pregnancies is necessary to infer potential teratogenicity (Koren, Pastuszak and Ito, 1998).

However, in the broader sense of pharmacovigilance for human teratogenicity, there are several limitations of the traditional pregnancy registry approach. As these studies depend on spontaneous reporting of exposed pregnancies, selection bias may be involved. It is also difficult to project sample sizes. Even with successful identification and recruitment of a high proportion of all exposed pregnancies occurring in the population, the absolute number of exposed pregnancies in the registry, and/or the specific timing of those exposures in gesta-tion, is unlikely to provide sufficient power to rule out or identify any but the most dramatic increased risks of specific congenital anomalies. This is of particu-lar concern in that most known human teratogens are associated with increased risks for specific patterns of birth defects and other adverse outcomes rather than an increase in all birth defects across the spectrum.

Thus, an important function of a typical pregnancy registry is to generate hypotheses on the basis of ‘signal’ detection when higher than expected numbers of specific malformations are reported, with additional studies required to confirm or refute the signal (Cham-bers et al., 2006).

Other limitations of traditional pregnancy registries include the difficulty in identifying an appropriate comparison group. Many registry designs do not include a registry-specific comparison group. Instead, outcomes in exposed pregnancies are frequently compared with externally derived reference rates for the general population. Depending on the character-istics of exposed pregnant women who are included in the registry, the use of external reference statistics, without the ability to adjust for possible confounding, may not represent the most appropriate comparison. Some registry designs do involve recruitment of an internal comparison group with collection of infor-mation on potential confounders so that comparisons can adjust for differences between groups (Scialli, 1999). Other registry designs that involve multiple drugs used for the same disease can address this prob-lem in part by comparing pregnancies with the expo-sure of interest to pregnancies with exposure to other medications used for the same disease (Scialli, 1999; Holmes, Wyszynski and Lieberman, 2004). Another approach is that used by the antiretroviral drugs in Pregnancy Registry. In this design, pregnancies with first-trimester exposure to the drugs of interest are compared with pregnancies in which exposure did not begin until the second or third trimester (Watts et al., 2004).

Finally, as registries typically rely on a wide vari-ety of individual healthcare providers and/or moth-ers themselves to report pregnancy outcome, there is a potential for the misclassification of outcomes such as major congenital anomalies with respect to accurate and complete diagnosis and/or suspected etiology (Honein et al., 1999). Furthermore, subtle or less easily recognizable teratogenic effects, such as the fetal alcohol syndrome or the minor struc-tural abnormalities that comprise the anticonvulsant embryopathy, are unlikely to be identified by the obstetrician or general pediatrician who is report-ing outcomes to a registry. In addition, especially when the healthcare provider is the primary source of registry information, there is concern that a substantial proportion of pregnancy exposure reports will be lost to follow-up, thereby potentially biasing conclusions that can be drawn from registry data.

In recent years, with the increasing number of preg-nancy registries established by industry sponsors as part of post-marketing commitments or initiated by other groups interested in generating pregnancy safety data, the U.S. FDA has produced a guidance docu-ment (U.S. FDA Office of Women’s Health, 2002) that establishes principles for the design and conduct of pregnancy registries. The guidance document is intended to improve and standardize the overall qual-ity and ultimate value of the data collected through pregnancy registry methods. In addition, a second recently approved Center for Drug Evaluation and Research (CDER) guidance document sets standards for reviewers who are evaluating human data on the effects of in utero drug exposure on the developing fetus (U.S. FDA Office of Women’s Health, 1999). Taken together, these guidelines provide a framework for the collection and interpretation of pregnancy exposure and outcome data that can contribute to consistency and improved quality in the collection and evaluation of safety data generated through pregnancy registries.

4. Birth defects monitoring or surveillance systems are designed to provide population- or hospital-based identification of congenital anomalies to measure trends and to respond to unusual clusters of events. At this level of information gathering, if an upward trend in the birth prevalence of a certain defect or a time-related cluster of an unusual pattern of defects coincides with the widespread use of a new medication, then surveillance programs can function as an early warning system (Khoury et al., 1993). Because an unusual pattern of congenital anomalies may occur with extreme rarity within any one surveil-lance system, these efforts are enhanced by collabora-tions such as the International Clearinghouse of Birth Defects Monitoring Systems (ICBDMS), which has been in existence since 1974 (Erickson, 1991; Khoury et al., 1994).

5. Birth defects case–control studies can be classified into one of the two general approaches. The first group might be termed classical hypothesis-testing case– control designs, whereas the second involves ongoing case–control surveillance for drug-induced congenital malformations.

Using the first design, cases and controls are iden-tified with the specific intent to measure the asso-ciation between a risk factor and a specified birth defect or group of defects. This approach requires that a priori decisions be made regarding the research questions, selection of the appropriate control group and adequate power and sample size. For exam-ple, based on concerns raised in the literature, this design was successfully used to document an asso-ciation between congenital facial nerve paralysis, or Möbius’s syndrome, and first-trimester use of miso-prostol (Pastuszak et al., 1998).

The second approach, case–control surveillance, is not based on a pre-defined set of hypotheses but is instead focused on gathering a broad range of exposure and potential confounder information for malformed cases and controls over an indeterminate period to create a large repository of data suitable for testing multiple future hypotheses. This approach has been incorporated into some birth defects monitor-ing programs in the United States and is the general design of the U.S. National Birth Defects Prevention Study (Carmichael et al., 2006). These methods are also used on an ongoing basis in programs such as the Slone Epidemiology Center’s hospital-based surveillance study based at Boston Univer-sity (Mitchell et al., 1981; Hernandez-Diaz et al., 2000), the Latin American Collaborative Study of Congenital Malformations (ECLAMC) that involves over 70 hospitals in several South American coun-tries (Castilla and Peters, 1992) and the population-based Hungarian Congenital Malformation Registry (Czeizel et al., 2000). These programs usually involve ascertainment of malformed cases as well as system-atic sample selection of non-malformed infants who can be used as controls. Exposure and other risk-factor information is generally gathered by postnatal mater-nal interview either in person or by telephone and, in some cases, is supplemented by review of medi-cal records or pregnancy log books. In addition, some designs have incorporated DNA sampling and bank-ing from case and control children and their parents so that future hypotheses regarding genetic susceptibility or gene–environment interaction can be tested.

The primary advantage of any case–control approach in studies of rare events such as congenital anomalies is the enhanced power to detect or rule out a mean-ingful association for a given sample size. In contrast to pregnancy registries or other prospective designs, this method is often the only appropriate approach for detecting moderate or low-level teratogenic expo-sures associated with specific major malformations. Furthermore, to the extent that case–control surveil-lance studies collect comprehensive information on potential confounders, including vitamin use, tobacco and alcohol, this approach can provide reassurances that moderate effect sizes are not attributable to these other factors. Other advantages of case–control surveillance include, to a varying degree, relatively complete ascer-tainment of the congenital anomalies of interest within a defined population, concurrent selection of controls from the same population and the ability to validate the classification of diagnoses.

In addition, this approach provides flexibility in the ultimate use of the data, i.e. based on specific research questions, subsets of cases and controls can be selected from the entire data set to test or confirm specific hypotheses. For example, this method was useful in confirming the protective effect of antena-tal folic acid supplementation in reducing the inci-dence of neural tube defects (Werler, Shapiro and Mitchell, 1993) and in refuting a previous finding of an association between maternal loratadine use and the genito-urinary tract anomaly, hypospadias (CDC, 2004). Furthermore, case–control surveillance data are amenable to hypothesis generation. For example, these data were used to first raise the question of an association between pseudoephedrine and gastroschi-sis (Werler, Mitchell and Shapiro, 1992).

The limitations of case–control studies of any type generally relate to the use of retrospective data collection and the selection of controls. For example, maternal interviews may be conducted in some cases many months after completion of the pregnancy, which raises the possibility, although controversial, of limited recall of early pregnancy medication use (Tomeo et al., 1999). In addition, the potential for serious differential recall bias among mothers of malformed infants relative to mothers of non-malformed controls has been cited by some (Khoury, James and Erickson, 1994), whereas the potential bias associated with the use of malformed controls has been suggested by others (Prieto and Martinez-Frias, 2000). With respect to the use of appropriate controls, case–control surveillance studies have the advantage of flexibility in the selection of one or multiple control groups, malformed or not, from the larger data set as judged necessary for any specific analysis.

Because case–control surveillance programs are ongoing, they have the potential to recognize an asso-ciation with a newly marketed medication; however, they may have limited sensitivity in this regard. These studies may miss an association if the medication of interest is related to a relatively unusual or uncommon congenital anomaly and/or that specific defect is not included in the range of selected anomalies for which maternal interviews are conducted. In addition, if new medications are infrequently used among pregnant women, then weak or moderate associations may be difficult to detect. However, for medications that are more commonly used – e.g. by 1% or more of preg-nant women – given the rarity of congenital anoma-lies in general, these approaches provide a relatively powerful method of hypothesis testing and hypothesis generating and can be effectively used alone and in conjunction with other methods.

6. Large cohort studies can involve open cohorts that are population-based and ongoing or can be hospital-or health insurer-based and/or of limited duration. For example, the Swedish Registry of Congenital Malfor-mations in combination with the Swedish Medi-cal Birth Registry encompasses nearly all births in Sweden and utilizes exposure interviews conducted by midwives during the first trimester of pregnancy as well as data recorded prospectively in medical records (Ericson, Kallen and Wiholm, 1999). The Collaborative Perinatal Project conducted in the 1960s was a study involving over 50 000 mother–child pairs identified at multiple sites throughout the United States (Chung and Myrianthopoulos, 1975). Simi-lar large longitudinal cohort studies, each to some extent addressing risk factors for congenital anoma-lies, have recently been initiated in other countries such as Denmark (Olsen et al., 2001) and are in the process of being organized in the United States under the auspices of the National Children’s Study (http://www.nationalchildrensstudy.gov/).

These studies have the advantage of large and repre-sentative sample sizes, prospective ascertainment of exposure information as well as data regarding a vari-ety of potential confounders and ability to collect outcome information over a long term of follow-up. In addition, women with and without the expo-sure of interest are concurrently enrolled as members of the cohort, facilitating the identification of one or more appropriate reference groups. Like ongoing case–control designs, studies of this type can address multiple hypotheses that need not be formulated a priori (Irl and Hasford, 2000).

However, even in large cohort studies, issues of sample size can be a limitation. For example, the Collaborative Perinatal Project had inadequate power to detect weak to moderate associations with any but the most common major congenital malforma-tions and the most commonly used drugs due to the relatively small numbers of women exposed to most specific medications of interest. By contrast, the Swedish Registry with approximately 120 000 annual births, accumulated over more than a 25-year span, has enhanced power to identify these asso-ciations, assuming the frequency of exposure in pregnant women is sufficient to test such hypothe-ses. For example, using the Swedish data, Kallen, Rydhstroem and Aberg (1999) were able to iden-tify over 2000 first trimester–inhaled corticosteroid (budesonide)-exposed pregnancies and rule out with acceptable confidence an increased risk in overall rate of major congenital anomalies. However, the numbers of exposed and affected infants were too small even in this relatively large cohort to address the hypothesis of an increased risk for oral clefts, which is the specific type of major congenital malformation that has previ-ously been associated with maternal systemic corti-costeroid use and is therefore of theoretical concern.

7. Small cohort studies focused on specific medications have been conducted by Teratology Information Services (TIS) both in North America and in Europe. These studies draw on a base of callers who contact a TIS-seeking counseling regarding the safety of a medication used in pregnancy. Follow-up of pregnancy outcome is obtained for selected exposures. These studies have strengths similar to the registries described above with respect to the potential for rapid identification of exposed women, particu-larly for a new drug, as well as prospective collection of exposure and other risk factor information. Tera-tology Information Services studies usually employ a concurrently enrolled unexposed control group, often both a disease-matched and a non-diseased group, which may provide the most appropriate reference groups in this context.

Similar to traditional pregnancy registries, the primary limitation of TIS studies relates to sample size. Individual TIS sites either independently or in collaboration have published studies typically involv-ing between 100 and 200 exposed subjects (Pastuszak et al., 1993; McElhatton et al., 1999). Also, similar to pregnancy registry designs, TIS studies rely on spon-taneous callers for the recruitment of subjects that may result in selection bias.

In an effort to increase sample size and to shorten the time needed to identify a given number of exposed pregnancies, collaborative projects among networks of TIS sites in North America are conducted through the Organization of Teratology Information Special-ists (OTIS) (Scialli, 1999) and in Europe through the European Network of Teratology Information Services (ENTIS) (Vial et al., 1992; Schaefer et al., 1996). These formal collaborations can add to the vari-ability and possibly the representativeness of subjects in the sample and increase the obtainable sample size by drawing on a larger population of potentially exposed women. However, even these studies, similar to other cohort studies with moderate sample sizes, usually are only sufficiently powered to detect or rule out very large increased risks of specific major congenital anomalies associated with exposures.

The primary strength of TIS studies is the abil-ity to evaluate a spectrum of pregnancy outcomes following a given exposure, including major congen-ital anomalies, spontaneous abortion and stillbirth, preterm delivery, pre- and postnatal growth deficiency and, in some cases, longer term child development. In this context, although underpowered to evaluate rare outcomes, these studies can be useful for gener-ating hypotheses that can be tested using other meth-ods. Furthermore, in some OTIS and individual TIS designs, exposed and comparison children are system-atically evaluated for a pattern of both major and more subtle minor congenital anomalies. This additional level of scrutiny can increase the sensitivity of this approach for the identification of a unique pattern of effects on fetal development, e.g. a pattern analogous to the anticonvulsant embryopathy, that might not be detectable through any of the other study methods available (Jones et al., 1989; Chambers et al., 2001; Lyons Jones, Johnson and Chambers, 2002).

8. Database linkage studies, as technological advances permit, can offer many of the advantages of large cohort studies at potentially far less cost. Early efforts along these lines utilized the Michigan Medicaid database, a government health insurance program within which maternal prescription records could be linked to pediatric billing records to identify children born with and without congenital anoma-lies (Rosa, 1999). Similar approaches have been used successfully elsewhere in North America and Europe. For example, investigators in Denmark have linked prescription database records to hospital discharge and medical birth register records for children with and without congenital anomalies to investigate the safety of a widely used antibiotic (Larsen et al., 2000).

In countries where there is universal and stan-dardized healthcare delivery and record keeping, or in countries where healthcare maintenance organi-zations or other large membership-based providers serve a significant proportion of the population, linked prescription and birth records provide an attractive alternative method for testing hypotheses regard-ing drug safety in pregnancy. For example, hospital discharge data across the Canadian population have been used to evaluate adverse outcomes of pregnan-cies complicated by asthma (Wen, Demissie and Liu, 2001). This approach has also been used successfully to evaluate pregnancy exposure to clarithromycin using longitudinal claims data for members from 12 geographically diverse United Health Group-affiliated insurance plans (Drinkard, Shatin and Clouse, 2000). Similarly, information from the Group Health Cooperative of Puget Sound in the United States has been used to examine the association between topical tretinoin (Retin-A) and major birth defects (Jick, Terris and Jick, 1993). The General Practice Research Database in the United Kingdom is another potentially fruitful resource (Jick and Terris, 1997; Jick, 1999). Recent efforts to develop algo-rithms for accurately identifying pregnancies, expo-sure windows, gestational timing and pregnancy outcomes utilizing this database hold promise for increased utilization off these existing resources to address hypotheses related to pregnancy exposures (Hardy et al., 2004).

The primary advantages of large-linked databases are the availability of large numbers of subjects, the ability to establish temporal relationships between exposure and outcome by constructing an histori-cal cohort and relative ease of access to previously collected medical, administrative or claims data. This approach also avoids some of the biases involved in studies that rely entirely on maternal report to classify exposure, especially if that information is collected retrospectively.

These strengths must be weighed against the limita-tions inherent in a study design that does not involve subject contact. For example, these studies usually cannot insure that the medication prescribed was actu-ally taken by the mother, taken in the dose prescribed or taken during the period critical for the develop-ment of any specific birth defect. To remedy this limitation, some database analytic designs involve the validation of a subset of records through other methods such as chart review or maternal interviews. There are also issues related to the misclassification of outcome depending on the quality of records used to determine or exclude the diagnosis of a congenital anomaly. Again, this limitation is not insurmount-able if it is possible to incorporate some level of validation.

In addition, similar to large cohort studies, even databases containing hundreds of thousands of patient records may have limited power to test drug-specific hypotheses due to relatively small numbers of pregnant women exposed to any particular drug. Furthermore, for low to moderate risk teratogens, large-linked databases often do not have immedi-ate access to information on potentially important confounders such as maternal exposure to tobacco, alcohol, vitamins and over-the-counter medications. However, databases can be a relatively efficient method for surfacing and testing hypotheses related to prescription medications, and therefore, these studies hold significant promise for the future.