The authors conducted a search of the peer-reviewed literature using PubMed (National Center for Biotechnology Information, United States National Library of Medicine) with a filter for human studies in the English language published through February 4, 2014. Two distinct search strings were used, the results of which were combined and de-duplicated in a single EndNote X7 (Thomson Reuters, 1998-2013) library: (1)

antihistamines AND (pregnancy OR birth defects OR congenital defects OR congenital malformations)

Search string (2) was used to ensure that papers which had a primary focus on NVP, but reported analyses for antihistamine use were captured. Papers that only reported on the associations with NVP did not meet inclusion criteria and would be excluded. Two co-authors each initially screened one-half of the article titles and abstracts for relevance; two additional coauthors then independently re-screened the article titles and abstracts. If both co-authors who reviewed a title and/or abstract agreed that the article could be excluded, then it was excluded without further review. If one of the two co-authors determined that the screened title and/or abstract needed a review of the full text, either because it appeared to meet inclusion criteria or it could not be determined whether it met inclusion criteria, the full text was retrieved and reviewed. The reference lists of the selected articles were also searched for additional papers that were not ascertained through the PubMed search (see flowchart, Figure 1). Through personal communication with the author, one additional article that was electronically published in September, 2013 but was not indexed in PubMed until March 2014 was also included in the review (in flowchart box entitled “Articles identified through other sources”).

A study was included in this review if it: (1)

was written in English,

Studies that focused exclusively on doxylamine plus pyridoxine use and were published during the 1980s and 1990s (through the year 2000) were excluded post-hoc, as it was determined that re-reviewing this literature was unnecessary in light of the several previously published meta-analyses and commentaries on this medication. In addition, if other studies included doxylamine plus pyridoxine as one of several antihistamines under investigation, the results of the other antihistamines are included in this review and the doxylamine plus pyridoxine results are not discussed, though the table notes for Tables 1 and 2 indicate the papers with published results for doxylamine plus pyridoxine use. Recent papers (since January 2001) that included doxylamine plus pyridoxine exposures are abstracted and included in this review because these are subsequent to the last published meta-analysis.

Included studies were not required to report a measure of association (i.e. odds ratio, prevalence ratio, risk ratio) or the results of statistical testing. For selected studies in which adequate data were available but a measure of association was not reported in the original publication, we calculated a measure of association. When distinct publications included overlapping data, it was noted in the text and notes for Tables 1 and 2.

A total of 24 cohort studies reported findings with respect to histamine H₁-receptor antagonists (Table 1). Eight of these 24 cohort studies were based on data from a Teratogen Information Service (TIS). Briefly, TIS locations in the United States and in several countries provide evidence-based information to mothers, health care professionals, and the general public about medications and other exposures during pregnancy and while breastfeeding. This typically occurs when physicians or women call the service concerned about a chemical or medication exposure. During the call, information on pregnancy exposures is collected. For TIS conducting studies of pregnancy outcomes, within a year of delivery, a follow-up survey is sent to the women to gather information on the outcome of the pregnancy. In some services, the outcome of the pregnancy is verified with a health care provider. Additional data sources are sometimes linked to the information gathered during the initial telephone call or in the follow-up interview. Many of the studies using TIS data derive an “unexposed” group from women calling about exposures deemed to be “non-teratogenic”, such as dental examinations, x-ray exposures, and use of medications such as acetaminophen.

One cohort study analyzed antihistamines in the aggregate only ¹² though reported frequencies of selected birth defects by specific medication exposures. Källén explored a wide array of first and second generation H₁-receptor antagonists in relation to several pregnancy outcomes, but reported estimates of association by indication only (not specific antihistamines), and analyzed medications used to treat NVP separately from those used to treat allergies. Among the reported analyses of aggregated outcomes such as all birth defects, selected birth defects, all CHDs, specified CHDs, and all birth defects except CHDs, there were no significantly elevated associations with either antihistamines used to treat NVP or those used to treat allergies ¹².

A total of 21 case-control studies that met inclusion criteria for this review reported findings with respect to histamine H₁-receptor antagonists (Table 2). Four studies analyzed antihistamines in the aggregate only ^36-39, the remainder reported aggregated analyses in addition to analyses of specific antihistamines.

In a paper focused on risk factors for infant craniostenosis (more commonly referred to as craniosynostosis), Källén and Robert-Gnansia ³⁶ compared prenatal use of medications among the mothers of 398 infants born with craniostenosis between 1995-2002 with the nearly 730,000 Swedish births during the same time period. Twenty-two case mothers had documentation in their prenatal record of first trimester exposure to an antihistamine compared with an expected frequency of 15.6 based on the larger population data (observed/expected ratio: 1.4; 95% CI: 0.9-2.1).

Two manuscripts based on data from the Slone Epidemiology Center Birth Defects Study (Slone BDS) also known as the Pregnancy Health Interview Study ^{37, 38} reported on the associations between any antihistamine use and gastroschisis; the more recent paper also reported on the association with small intestinal atresias ³⁸. In data from 1976-1990, there were 76 gastroschisis cases, among whom 10 (13.2%) were exposed to antihistamines (excluding antiemetics) and 2,142 controls affected by other, non-related birth defects, among whom 173 (8.1%) were exposed to antihistamines leading to an adjusted OR of 1.3 (95% CI: 0.5-3.1) ³⁷; in more recent Slone BDS data (1995-1999) based on 206 gastroschisis cases, the association was closer to the null (OR: 0.6; 95% CI: 0.3-1.2). There was also no association with small intestinal atresia (OR: 0.9; 95% CI: 0.4-1.8) (based on 126 small intestinal atresia cases) ³⁸.

Boneva and colleagues, using data from metropolitan Atlanta from 1982-1983, analyzed antihistamine use for the treatment of severe first trimester NVP among 998 mothers of infants with a CHD and 3,029 mothers of infants born without any major birth defect. The analysis focused on Bendectin use (not reported in this review) and antihistamines in general. Comparing women with the most severe level of NVP to those with no NVP, the association between CHD and using any antinausea medication was protective with a relative risk estimate of 0.74 (95% CI: 0.56-0.97). Considering specific CHD subtypes, all associations among women with NVP, comparing those who took medications with those who did not, were below the null (but not statistically significant), with the exception of tetralogy of Fallot, which was greater than 1.0 (OR: 2.19; 95% CI: 0.39-22.32), based on 8 cases, 6 of whom took medication to treat NVP, compared with 2 who did not take medication ³⁹.

The body of literature investigating the association between histamine H₂-receptor antagonists and birth defects is substantially smaller than that for H₁-receptor antagonists, with seven cohort studies ^59-65 and three case-control studies ^{53, 66, 67} meeting inclusion criteria for this review. Colin Jones and colleagues reported on 12-month post-marketing surveillance in Scotland of 9,809 users of cimetidine and a comparison population of 9,140 non-users. Over the surveillance, there were 20 pregnancies exposed to cimetidine and 22 unexposed pregnancies; among cimetidine users there was one birth defect reported – a case of Down syndrome. There were no birth defects among the non-users ⁵⁹.

Two studies used data from a TIS – the first based on data from the Motherisk Program ⁶² and the second based on pooled data from 18 members of the European Network of Teratogen Information Service (ENTIS) ⁶⁰. In the analysis of Motherisk data, there were 185 pregnancies exposed to histamine H₂-receptor antagonists (i.e. ranitidine [the most common], cimetidine, famotidine, nizatidine), 142 of them were first trimester exposed and resulted in a live birth. Three birth defects occurred among these pregnancies (2.1%) compared with 5 birth defects among 143 pregnancies in the comparison group (3.5%). The difference between the prevalence of birth defects among these two groups was not statistically significant (p=0.55) ⁶². The analysis of pooled ENTIS data (n=553 exposed to H₂-receptor antagonists; n=1,390 in comparison group exposed to “non-teratogenic” substances) had similar findings with a prevalence of birth defects of 2.7% among the exposed pregnancies and 3.5% among those unexposed to H₂-receptor antagonists ⁶⁰. In another analysis of the Motherisk data (combined with data from Italy and France TIS), first trimester exposure to H₂-receptor antagonists (not otherwise specified; n=113; n=98 with live births) was compared with exposure to non-teratogens (n=113; n=66 with live births) ⁶⁵. The prevalence of birth defects among those with live births was similar in the exposed and unexposed groups: H₂-receptor antagonists: 3.1%; non-teratogens: 3.0%.

In a 1998 publication of data from the Swedish Medical Birth Registry, Källén reported on the association between acid-suppressing drugs (e.g. proton pump inhibitors, H₂-receptor antagonists) and birth defects. The baseline birth defects prevalence was 3.9%; among users of H -receptor antagonists, the prevalence was 2.4% (6/255) (OR: 0.46; 95% CI: 0.17-1.20)⁶¹ Ruigómez and colleagues reported on two cohorts of cimetidine- and ranitidine-exposed pregnancies – one from Italy and one from the United Kingdom ⁶⁴. In the United Kingdom cohort, prevalence of birth defects among cimetidine-exposed pregnancies was 4.0% (9/227) and among ranitidine-exposed pregnancies was 7.4% (17/229), compared with the baseline prevalence of 5.7% among the unexposed population. In the Italy cohort, 2/10 (20%) cimetidine-exposed pregnancies and 3/101 (3.0%) rantidine-exposed pregnancies were compared with the unexposed baseline of 2.9%. Measures of association for each medication were calculated from pooled data: cimetidine (RR (95% CI)): 1.3 (0.7-2.6); ranitidine: 1.5 (0.9-2.6) ⁶⁴. In the most recently published cohort analysis, Matok and colleagues used data from the largest health maintenance organization in Israel to examine the safety of H₂-receptor antagonists during pregnancy ⁶³. The associations between any major birth defect and first trimester exposure to any H₂-receptor antagonist (n=1,148 exposed pregnancies) or famotidine (n=878 exposed pregnancies) were both null (OR: 1.03; 95% CI: 0.80-1.32 for any H₂-receptor antagonist and OR: 1.21; 95% CI: 0.92-1.58 for famotidine). No measure of association was calculated for ranitidine because there were fewer than seven birth defects noted (n=276 exposed pregnancies) ⁶³.

Anderka and colleagues, using NBDPS data, reported on H₂-receptor antagonists as a group, and where sample size permitted, ranitidine. There was no association found between H₂-receptor antagonists and CL/P (OR: 0.57; 95% CI: 0.19-1.67), CPO (OR: 1.04; 95% CI: 0.39-2.75), or hypospadias (OR: 1.07; 95% CI: 0.41-2.83). There was an elevated, though not statistically significant association with NTD (OR: 1.80; 95% CI: 0.80-4.05), though when restricted to ranitidine exposure only, the association did not persist (OR: 1.28; 95% CI: 0.43-3.82) ⁵³.

Data from the HCCSCA have been used in two recent case control analyses exploring H₂-receptor antagonists and birth defects. In the first study, Ács and colleagues explored the association among mothers with severe chronic dyspepsia. Severe chronic dyspepsia affected 148 case mothers and 214 control mothers; 9.5% (14/148) case mothers and 12.6% (27/214) control mothers took an H₂-receptor antagonist (i.e. cimetidine or ranitidine, not specified in the paper) (OR: 0.7; 95% CI: 0.4-1.4) ⁶⁶. In the second study, Bánhidy and colleagues, using data from HCCSCA restricted their analyses of the association between cimetidine exposure and major birth defects to mothers with definitive peptic ulcer disease (PUD). There were 20 mothers of cases (6 exposed) and 58 mothers of matched controls (7 exposed); the association was elevated (OR: 3.1; 95% CI: 0.9-10.8) ⁶⁷.

The distinction between population-based studies and studies based on a convenience sample is important. The population-based case-control studies included in this review ^{3, 39, 49, 53, 55, 66, 67} have the advantage of deriving cases from surveillance systems that, in general, attempt to capture all major birth defects cases occurring in a population. Most of the significant findings were reported in case-control studies which are typically better able to assess individual types of birth defects rather than being limited to an assessment of all birth defects as an aggregate group. Given the rarity of major birth defects in the population (approximately 3% overall ⁶⁸; specific birth defects are less common) population-based cohort studies for birth defects are less frequent in the literature; however, cohort studies in both Sweden ^{14, 19, 30, 31} and the United States ^{17, 69} have been conducted making important contributions to the literature. The Swedish cohorts were built from extensive linkages of national registries; while the Collaborative Perinatal Project gathered data through maternal interviews of pregnant women enrolled at medical centers around the United States ^{70, 71}. Cohorts derived from health maintenance organizations are designed to simulate population-based cohorts ^{18, 25-27, 63} and in some situations do capture the vast majority of the population in a given geographic area. These cohorts have access to rich clinical information as well as prescription claims. Cohorts based on a convenience sample, however, such as those derived from the TIS around the United States and internationally, while likely having acceptable internal validity, have questionable external generalizability. The TIS serve a critically important role – providing counsel to concerned women and providers about chemical and medication exposures during pregnancy and might provide important initial signals of teratogenicity. However, TIS study cohorts are comprised exclusively of individuals who choose to inquire about an exposure with a TIS ^{20-22, 29, 32, 33, 35, 60,65}. Those who inquire about an exposure are likely to be qualitatively different from those who do not; those who do not inquire cannot, by definition, be in a TIS cohort. In addition, the TIS studies, by necessity, must select a comparison group of women exposed to “something” – since this is the entirety of their cohort. The studies tend to choose a comparison group of women exposed to “nonteratogenic agents” that is consisting of women exposed to dental x-rays or acetaminophen (see Table 2).

A second important issue to consider is the ascertainment and analysis of antihistamine exposures. Medication exposure during pregnancy is typically ascertained in one of two ways: (1) self-report by the mother, either prospectively or retrospectively and reported either in prenatal care records or during a maternal interview for the purposes of the study; or (2) linkage with pharmacy records or pharmaceutical claims. Both approaches are represented in the 54 papers included in this review though self-report by the mother was the most common; over 30 studies used the approach either alone or in combination with one of the other approaches. Given that many of the antihistamines of interest are currently sold over-the-counter, self-reported exposure assessment is a critical tool. In addition, given concerns about the safety of medication use (including antihistamine use) during pregnancy ⁷², it is not uncommon for women to fill a prescription, but decide not to take it during pregnancy ⁷³. However, retrospective self-reporting has limitations – the most concerning is the potential for biased or inaccurate recall ⁷⁴. The Slone BDS and NBDPS attempt to limit inaccurate recall by asking about specific indications of medications, categories of medications, and specific medications, including both generic and brand names. Study participants in the BDS are also provided with picture booklets of medications as well as a calendar to highlighting key dates and events to aid recall ^{2, 37, 38, 54}. Earlier BDS analyses also used a “malformed control” group ^{2, 37, 38}; this design was chosen under the assumption that mothers of control infants born with birth defects other than the birth defect of interest might have less biased recall than mothers of control infants born without any major birth defects. In addition, NBDPS has approximately a 70% response rate among cases and controls; analyses have suggested that NBDPS participants are representative of the populations from which they were selected ⁷⁵. However if mothers of cases and controls have differential participation with respect to pregnancy exposures, study findings could be subject to selection bias.

In the analytic phase of a study, steps can be taken to enhance the accuracy of the exposure data. One of the most common approaches in studies of birth defects is to limit the medication exposure to the first trimester (sometimes also including the 30 days prior to conception to account for the fact that women who are prescribed a medication prior to conception may take it during after conception as well). Since the primary period of susceptibility to human teratogens is the first eight weeks of pregnancy (approximately 10 weeks when counting from the date of last menstrual period) ⁷⁶, a focus on medication exposure during the first trimester (since medication exposure data are often too imprecise to accurately focus on specific weeks) is appropriate. While most studies limited the period of exposure to the first trimester, a handful explored the association with exposure any time during pregnancy ^{13, 60}, potentially leading to an underestimate of the association with birth defects. In another analytic approach, the Slone BDS has developed and utilized an exposure classification algorithm based on the certainty of recall to help categorize individuals as “likely” or “possibly” exposed to a medication ⁷⁷. Furthermore, the investigators of several case-control studies ^{39, 51, 53, 66, 67} have restricted analyses to individuals with a particular indication for medication use (e.g. NVP, peptic ulcer disease and severe dyspepsia in the papers noted above) as a method of minimizing confounding by indication ⁷⁸. This is an advantage in this body of literature, especially since a lack of NVP is considered a risk factor for a poor pregnancy outcome. In addition, because women often take multiple medications during pregnancy⁷⁹ analyses can exclude women who took medications during pregnancy known to be teratogenic or adjust for other medication use.

Among the studies included in this review, the data sources for the ascertainment of birth defects varied widely with respect to data quality and specificity. In studies using birth defects data derived from state or national surveillance programs, the quality and accuracy of the data were likely to be high ^{3, 12, 19, 30, 31, 39, 46, 49, 51, 53, 55-58, 61}. In datasets developed exclusively through linkages between national birth registry data and hospital discharge data with cases of birth defects identified by International Classification of Diseases (ICD) coding, the quality might be somewhat lower than data that are actively abstracted from medical records, however, the statistical power that comes with larger sample sizes as well as the ability to create small, relatively homogeneous categories of birth defects is beneficial. Additionally, in some studies ^{39, 54, 55} rigorous clinical review of every case ^80-82 was undertaken to verify reported diagnoses and determine whether the case met the study's inclusion criteria. Similarly, a few studies in the review are based on data from health maintenance organizations ^{18, 25-27, 42, 63} with medical records forming the foundation of the birth defects ascertainment and in some of these studies, additional clinical review of potential cases was also undertaken.

In contrast, in other studies, such as those based on the TIS included in this review ^{20-22, 29,32, 33, 35, 60, 65}, the pregnancy outcome data were self-reported by mother who initially called the TIS, up to a year or more after their initial inquiry. This delay could lead to a substantial loss to follow-up, if mothers could not be located in order to report on the pregnancy outcome. While the reported pregnancy outcome was verified with the child's pediatrician in some of the included studies, this was not universally done across all TIS-based analyses.

A final methodological issue pertaining to birth defects outcome ascertainment is follow-up – the time period after birth during which birth defects could be identified and reported. In most state and national surveillance systems, this period of time is at least one year, accounting for the fact that not all birth defects are identified prenatally or immediately after birth. Some studies ascertained birth defects over a shorter period of time, for example, at delivery ²³ or before age 5 months ^{37, 38, 54}. In the BDS, the 5-month cut-off was intentional, to help ensure the conduct of maternal interviews no later than 6 months after delivery.

Despite these methodological issues, the body of literature on the risk of birth defects from antihistamine use in early pregnancy remains reassuring, particularly for the first generation H₁-receptor antagonists. There is still a need for larger studies with clinical verification of birth defect subtypes, validation of maternal recall of medication exposures, and appropriate selection of study populations to follow-up on some of the signals found in previous studies, as well as a need to enrich the body of literature on H₂-receptor antagonists, which are currently relatively understudied.

Finally, as stated earlier, this review was limited in scope to only structural birth defects; other potential adverse pregnancy outcomes such as fetal loss, preterm delivery, or low birth weight, or longer-term neurodevelopmental outcomes were not included. To have a complete understanding of the safety of medication for use during pregnancy, an understanding of the full spectrum of adverse outcomes associated with that medication is required. At this time the body of literature investigating other adverse pregnancy or developmental outcomes associated with prenatal use of antihistamines is much smaller than that focused on birth defects.