Epidemiology of twinning in the National Birth Defects Prevention Study, 1997 to 2007

Introduction

In 2011, 1 in every 30 babies born in the United States was a twin (Martin et al., 2013). Twinning is associated with numerous pregnancy complications and adverse maternal and fetal outcomes (Helmerhorst et al., 2004; Boulet et al., 2008; Morcel et al., 2010; Yang et al., 2011). Mothers of multiples have twice the risk of pregnancy-related mortality and an increased risk of preterm labor and cesarean delivery (Conde-Agudelo et al., 2000). Compared with singletons, twins are at a higher risk of low birth weight, birth defects, prematurity, and infant mortality (Martin and Park, 1999; Li et al., 2003a; Tang et al., 2006; Chauhan et al., 2010; Zhang et al., 2011). The increased risk for adverse outcomes is particularly concerning given the increasing rate of twin births in the United States from 1980 (18.9 per 1000) to 2011 (33.2 per 1000) (Martin et al., 2013).

Recent studies have estimated that approximately 19% of U.S. twin births are attributable to assisted reproductive technology (ART) and an additional 19% of twin births are attributable to non-ART ovulation stimulation treatments (Sunderam et al., 2012; Kulkarni et al., 2013). Other factors that are associated with twinning include a family history of twinning, older maternal age, multiparity, obesity, and African-American race (Khoury and Erickson, 1983; Allen and Parisi, 1990; Martin and Park, 1999; Hoekstra et al., 2010). The relationships between twinning and oral contraceptives, folic acid-containing multivitamins, and maternal smoking have been examined, with conflicting results (Rothman, 1977; Olsen et al., 1988; Czeizel et al., 1994; Li et al., 2003b; Muggli and Halliday, 2007; Hoekstra et al., 2010).

Although twinning has been studied extensively, much of the work involving risk factors for twinning was conducted before fertility treatments were in widespread use. Given the steady increase in use of assisted reproduction, understanding the factors associated with multiple births in this population is important (Centers for Disease Control and Prevention et al., 2011). In addition, few studies have evaluated factors associated with twinning among mothers whose offspring have birth defects (Tang et al., 2006; Zhang et al., 2011). Our objectives were to use data from the National Birth Defects Prevention Study (NBDPS) to evaluate maternal demographic factors and periconceptional exposures associated with twinning, stratified by use of fertility treatments. We assessed these associations among mothers of unaffected controls. We also included mothers of infants or fetuses with a birth defect so that we might compare these findings with those of control mothers. Because twinning is associated with an increased risk of birth defects, we were interested in factors associated with twinning that were unique to the case group, as we hypothesized that these factors may be related to both the birth defect and twinning. To minimize heterogeneity, we selected one phenotype—isolated cleft lip with/without cleft palate or cleft palate only (cases)—as these are some of the more common birth defects that have previously been associated with twinning (Mastroiacovo et al., 1999; Tang et al., 2006; Zhang et al., 2011).

Materials and Methods

Results

Approximately 2.7% (n = 228/8470) of control mothers reported a twin gestation compared with 3.9% (n = 117/2976) of case mothers (i.e., fetus/infant with orofacial cleft). The interview participation rate was 68.4% for mothers of cases and 65.7% for mothers of controls. Approximately 31.1% of mothers of control twins and 29.9% of mothers of case twins reported use of fertility treatments (Table 1). Fertility treatment use was significantly associated with twinning for control mothers (OR, 14.5; 95% CI, 10.7–19.8) and case mothers (OR, 9.5; 95% CI, 6.2–14.8). ART use and clomiphene citrate use were significantly associated with twinning among control mothers (OR, 51.0; 95% CI, 31.8–81.7; and OR, 6.1, 95% CI, 3.5–10.7, respectively) and case mothers (OR, 23.6; 95% CI, 12.8–43.6; and OR, 5.3, 95% CI, 2.5–11.1, respectively).

Results were generally similar for case and control mothers (Table 1). Among control mothers, twinning increased with increasing maternal age: 58.8% of mothers of twins were age 30 years or greater, compared with 38.4% of mothers of singletons. Among control mothers, other factors significantly associated with an increased prevalence of twinning in bivariate analyses were higher education (> than a high school graduate), higher income (> $50,000 per year), and use of a folic acid-containing multivitamin. Hispanic control mothers were also less likely to have twins compared with non-Hispanic white mothers. Among mothers of cases with clefts, significant predictors of twinning were taller maternal height, higher maternal income, and use of a folic-acid containing multivitamin. A sensitivity analysis restricted to mothers who reported plurality during the interview that matched birth certificates (control twins: n = 187; case twins: n = 96) resulted in very similar findings (data not shown).

Several factors were found to be significantly associated with twinning among control mothers who reported unassisted conception (Table 2). For these women, the association between non-Hispanic black race and twinning was of borderline significance after adjusting for study site (adjusted odds ratio [aOR], 1.6; 95% CI, 1.0–2.4). Increasing parity was modestly associated with twinning, with ORs of borderline statistical significance after adjusting for maternal age, race/ethnicity, education level, annual household income, and study site (one previous live birth: aOR 1.4, 95% CI 0.9–2.1; two+ previous live births: aOR 1.6, 95% CI 1.0–2.5). Maternal tobacco smoking was significantly associated with twinning after adjusting for maternal age, race/ethnicity, education level, annual household income, parity, study site, and year of EDD (aOR 1.6, 95% CI 1.1–2.4).

Among the control mothers who reported any use of fertility treatments, twinning was more frequent among mothers aged 30 to 34 years compared with mothers aged 25 to 29 years (Table 2). The proportion of non-Hispanic white mothers was slightly higher among mothers of twins (91.6%) than among mothers of singletons (83.9%). The odds of twinning were 4.3 times higher for mothers with an annual household income ≥ $50,000 compared with those with an income between $10,000 and $50,000 (95% CI, 1.9–9.9). Twinning was less common among mothers 159 to 163 cm tall than among mothers <159 cm tall.

Among case mothers who did not report use of fertility treatments, non-Hispanic black women were more than twice as likely as non-Hispanic white women to have a twin pregnancy (aOR, 2.3; 95% CI, 1.1–4.6) (Table 3). Compared with women with no previous live births, women with at least two previous live births had a modestly increased risk of a twin pregnancy that was of borderline statistical significance (aOR, 1.7; 95% CI, 0.9–3.1). Mothers from New York were also significantly more likely to have a twin pregnancy than their counterparts from Georgia, although this association was of borderline significance after adjusting for year of EDD. Overall, the associations for case mothers were similar to the findings from the controls analysis, with the notable exception that there was no association observed for smoking and twinning among case mothers reporting unassisted conception. We did not observe any significant associations with twinning among case mothers who reported use of fertility treatments, although the small sample size limited our ability to detect associations in this group.

Associations with twinning for the unlike-sex control twins and estimated dizygotic twins were similar to the associations observed for all control twins (Table 4). Fewer associations were observed for the like-sex and estimated monozygotic (MZ) control twins.

Discussion

Twin pregnancies occurred in approximately 2.7% of control mothers in the NBDPS, similar to the U.S. twin birth rate of 3% during the same time period (Martin et al., 2013). As expected, fertility treatment use was a significant predictor of twinning, both for mothers of controls and of cases with isolated orofacial clefts. Our results support previous reports of an association between fertility treatments and MZ twinning (Aston et al., 2008). Given the magnitude of the associations for ART and clomiphene citrate, it may appear that our results indicate a greater contribution of ART births to the overall twinning rate compared with non-ART treatments, in contrast to previous studies (Kulkarni et al., 2013). However, when comparing the proportion of twins attributable to IVF with non-IVF treatments, without adjusting for maternal age, our results (16.7% of twins and 14.5%, respectively) were similar to those of Kulkarni et al. (2013). We did not find any evidence to suggest a difference in the factors associated with twinning for unaffected controls and cases with orofacial clefts.

Historically, maternal age has been cited as a strong predictor of twinning (Bulmer, 1970; Bortolus et al., 1999). A recent report found the natural dizygotic twinning rate was relatively stable between 1969 and 2009 after controlling for maternal age (Derom et al., 2011). We found an association between maternal age and twinning for mothers reporting assisted conception but not among mothers reporting unassisted conception. One explanation for the discrepancy may be that we could not account for zygosity in our primary analyses, and previous studies have suggested that the association with maternal age may be limited to mothers of DZ twins (Bulmer, 1970). In our zygosity sub-analysis, we observed a trend of increased frequency of twinning with increasing maternal age for MZ and DZ twins, although the association was not statistically significant for MZ twin mothers. Our finding that the association between maternal age and twinning was attenuated after controlling for fertility treatment use, maternal race/ethnicity, education level, annual household income, study site, and year of EDD may suggest that this association is more complex than previously described.

Other factors associated with both fertility treatments and twinning such as education, prepregnancy weight, and use of a folic acid-containing multivitamin were associated with twinning when not controlling for fertility treatment use but were no longer associated after the models were stratified by fertility treatment use. The possibility of an association between folic acid and twinning has been particularly controversial (Werler et al., 1997; Li et al., 2003b). Our study found little difference in folic acid intake during the pregnancies of twin and singleton mothers reporting unassisted conception. In the ART stratum, almost all mothers reported taking folic acid. Although more twin mothers reported taking folic acid, this association was not statistically significant.

Among mothers who reported unassisted conception, a moderate increase in the frequency of twinning was observed for mothers who were non-Hispanic black, who reported two or more previous live births, and who smoked cigarettes between 1 month before conception through the first month of pregnancy, which is consistent with previous findings (Bulmer, 1970; Nylander, 1981; Hoekstra et al., 2010).

The associations we observed for mothers reporting fertility treatments likely reflect the cost of fertility treatments or choices regarding treatment options, such as implanting multiple embryos, rather than causal factors. The association between income and twinning may be due to treatments such as IVF being cost-prohibitive to lower income families and many insurers do not provide coverage for these treatments (American Society for Reproductive Medicine, ). Cost has been previously cited as a factor related to the socioeconomic and ethnic disparities among women using fertility treatments (Smith et al., 2011). Similarly, the large proportion of twin births in MA are likely due to state mandates that insurers provide coverage for fertility treatments, including IVF (Martin et al., 2011). For women receiving IVF, choices regarding the number of embryos implanted are an important consideration, although not one that we were able to analyze.

Our study has several limitations. First, our study was limited by the relatively small sample size after stratifying by fertility treatment use and estimated zygosity. We were unable to conduct multivariable analyses for mothers reporting fertility treatment use. We relied on self-reported data from a retrospective maternal interview, so we cannot rule out exposure misclassification due to inaccurate recall. The average age at interview was 9 months for controls (singletons: 9.3 months; twins: 10.1 months) and 11 months for cases (singletons, 10.7 months; twins, 11.3 months). The study participation rate was slightly higher among mothers of twins than mothers of singletons for both controls (singletons, 65.4%; twins, 68.7%) and cases (singletons, 68.2%; twins, 71.6%). We were not able to verify the accuracy of the responses for reported periconceptional exposures or reported use of fertility treatments. For mothers who reported fertility treatment use, we were unable to determine the number of embryos implanted, a strong predictor of twinning (Templeton and Morris, 1998). Another limitation is that a comparison of the plurality reported by the mothers in the interview with information recorded on the infant's birth certificate or mother's hospital record revealed some discrepancies, with mothers reporting higher rates of twin gestation. Although previous studies have found a high degree of concordance between plurality from maternal report and on birth certificates, some variability exists and may be due to clerical errors or early fetal demise of a co-twin (Querec, 1980; Schoendorf et al., 1993). Ascertainment of early fetal demise has likely increased because these previous studies due to increased use of early ultrasonography. A sensitivity analysis restricted to control infants without plurality discrepancies produced results similar to those from the primary analysis.

We were also unable to directly determine zygosity. This limitation is important because many factors have only been associated with DZ twinning (Bulmer, 1970; Hoekstra et al., 2010). Zygosity is particularly difficult to obtain as the gold-standard for determining zygosity is genetic analysis (Chen et al., 1999). However, our simulation modeling approach may be useful as it allowed us to estimate risk factor associations for MZ and DZ twins with accompanying uncertainty intervals. Although we attempted to assess factors associated with zygosity of twins, there are other features of twinning, such as chorionicity, that we were not able to assess using these data.

Strengths of this study include use of a demographically diverse, population-based sample from 10 different states across the United States. We were able to control for a variety of factors associated with twinning. This analysis assesses potential risk factors for twinning among a modern sample of mothers that reported fertility treatment use in addition to mothers that reported unassisted conception of twins.

Fertility treatment use was a strong predictor of twinning, as expected, but the factors associated with twinning among women receiving fertility treatments are likely due, in part, to the type of treatment, number of embryos implanted, and the availability of insurance coverage. We observed an increase in twinning with increasing maternal age, but factors such as maternal race/ethnicity and socioeconomic status may contribute and should be examined further. Given the increased risk of adverse outcomes associated with a twin pregnancy, understanding the factors associated with twinning, be it unassisted or assisted, is of public health importance.