Pregnant women in the United States are widely exposed to caffeine from consumption of coffee, soda, tea, and chocolate (Knight et al., 2004; Frary et al., 2005). Dietary caffeine is absorbed by the gastrointestinal tract and reaches almost all tissues of the body within 45 minutes (Marks and Kelly, 1973). In healthy adult women, the half-life of caffeine is about 2 to 6 hours (Patwardhan et al., 1980; Charles et al., 2008), which may be increased to 11 hours among those taking oral contraceptives (Patwardhan et al., 1980) or who are pregnant (Knutti et al., 1982). Also, caffeine crosses the placenta but the enzymes necessary to metabolize caffeine are absent in the fetus until several days after birth (Weathersbee and Lodge, 1977).

In addition, animal models have shown a teratogenic effect of maternal caffeine intake at high doses (Fujii et al., 1969; Nishimura and Nakai, 1960; Scott, 1983). Limb deficiencies (LDs), along with cleft palate, and neural tube defects (NTDs) are among the most frequent congenital malformations induced in animals by caffeine doses of 100 to 200 mg/kg (Fujii et al., 1969; Scott, 1983; Moriguchi and Scott, 1986). Synergistic teratogenic effects of subteratogenic doses of caffeine with certain agents, such as acetazolamide, treatment for glaucoma, and seizures (Ritter et al., 1982; Beck and Urbano, 1991), or mitomycin C, a hemotherapeutic agent (Fujii and Nakatsuka, 1983), have been observed in rodents.

In adult humans, caffeine intake has been associated with increased plasma cholesterol and homocysteine levels. Increased maternal plasma cholesterol level may be related to development of congenital vascular disease (Manderson et al., 2002), and increased maternal homocysteine level has been associated with congenital heart defects (Wenstrom et al., 2001) and NTDs (Mills et al., 1995). Although the literature has generally indicated that dietary caffeine is likely to be a weak teratogen at most for humans (Nelson and Forfar, 1971; Fedrick, 1974; Aro et al., 1982; Linn et al., 1982; Rosenberg et al., 1982; Kurppa et al., 1983; Furuhashi et al., 1985; Adams et al., 1989; Tikkanen and Heinonen, 1990; Olsen et al., 1991; Tikkanen and Heinonen, 1991; McDonald et al., 1992; Tikkanen and Heinonen, 1992a; Tikkanen and Heinonen, 1992b; Werler et al., 1992; Ferencz et al., 1993; Tikkanen and Heinonen, 1994; Fixler and Threlkeld, 1998; Samrén et al., 1999; Torfs and Christianson, 1999; Torfs and Christianson, 2000; Browne, 2006; Browne et al., 2007; Miller, 2008; Mongraw–Chaffin et al., 2008; Collier et al., 2009; Schmidt et al., 2009), considering the common use of caffeine by pregnant women, a slight risk elevation could have a significant impact at the population level. Given that LDs are the most frequent malformations induced by caffeine exposure in some animal studies (Fujii et al., 1969; Scott, 1983; Moriguchi and Scott, 1986) and that this relationship between LDs and maternal caffeine consumption has not been examined in humans, the current study explored the association between maternal caffeine consumption and LDs.

Lumping etiologically different groups of malformations together (Nelson and Forfar, 1971; Linn et al., 1982; Olsen et al., 1991) or into broad categories (Rosenberg et al., 1982; Kurppa et al., 1983; Furuhashi et al., 1985; McDonald et al., 1992; Samrén et al., 1999; Torfs and Christianson, 1999; Natsume et al., 2000) may have obscured identification of positive associations for a specific defect group. In addition, many studies assessed the dietary caffeine intake only from coffee and tea consumption.

This study attempts to address these limitations by measuring maternal caffeine intake from caffeinated coffee, tea, soda, and chocolate and by exploring the effect of caffeine exposure on the etiologically different subtypes of LDs using the National Birth Defects Prevention Study (NBDPS) data. Analyses previously conducted using NBPDS data examined associations between maternal dietary caffeine consumption and congenital heart defects (Browne et al., 2007), orofacial defects (Collier et al., 2009), NTDs (Schmidt et al., 2009), anorectal atresia (Miller et al., 2009), and bilateral renal agenesis or hypoplasia (Slickers et al., 2008), as well as other birth defect groups (Browne et al., 2011). The accumulated number of cases from NBDPS is now sufficient to examine subtypes of LDs. Potential effect modification by maternal smoking, alcohol consumption, and use of vasoconstrictive medication was also evaluated.

During the study period, 23,306 case and 8488 control infants were enrolled in the NBDPS. Among all case infants, 920 were identified with LDs, which included 871 live births, 19 stillbirths, and 30 induced abortions. After excluding infants with amniotic band syndrome (3 cases), LD case and control infants missing maternal caffeine exposure information (11 cases and 141 controls), mothers reported to have had chorionic villus sampling procedures during the index pregnancy (35 cases and 234 controls), and mothers diagnosed with type 1 or type 2 diabetes before pregnancy (28 cases and 51 controls), 844 LD cases and 8069 control infants remained.

As shown in Table 1, the LD subtypes with the largest numbers were transverse, longitudinal, and preaxial, of which 85%, 55%, and 38% were isolated LD cases, respectively.

Table 2 presents the distribution of maternal and pregnancy characteristics of the study participants. A larger proportion of case infants were boys compared to control infants. Compared to the control mothers, a higher proportion of case mothers tended to be younger in age and not to have children before the index pregnancy, to be of Hispanic ethnicity, to have 12 years of education or less, and to be overweight or obese. Case mothers were less likely to report nausea or vomiting during the first month compared to control mothers. Fever during pregnancy, active periconceptional smoking, and vasoconstrictive medication use were more common among case mothers compared to control mothers. Alcohol drinking, including binge drinking, was less often reported among case mothers compared to control mothers. Folic acid supplement use was slightly more common among mothers of all LD subtypes, except transverse LD subtype, compared to control mothers.

Among all eligible control mothers, 97% (n = 7806) reported caffeine consumption, with a mean intake of 129.4 mg/day. Of those control mothers who reported caffeine consumption, 47% reported coffee consumption, 48% reported tea consumption, and 68% reported soda consumption, with mean caffeine intakes of 139.5 mg/day, 34.0 mg/day, and 64.7 mg/day, respectively. Among mothers of infants with LDs, 96% consumed dietary caffeine (data not shown).

The associations between maternal caffeine consumption and odds of each LD subtype were adjusted for 0, 1, or 2 confounders, depending on the subtype. Including all three beverage types together produced very similar results to the models with individual beverages, indicating that exposure sources did not act as confounders to each other. Therefore, the final models presented only included the individual beverage type of interest. No covariables were adjusted for in the final model of isolated longitudinal LDs, for which crude ORs were reported. ORs are presented in Table 3 by LD subtype for isolated cases and in Table 4 for MCA LD cases for total dietary caffeine, coffee, tea, and soda consumption.

Increased odds for all isolated LDs combined and for isolated transverse LDs were observed for all total dietary caffeine intake categories compared to the referent category (Table 3). ORs were weakly to moderately elevated for all total daily dietary caffeine categories and isolated longitudinal LDs. Odds of isolated preaxial LDs were not associated with any caffeine consumption category compared to the no to low consumption category. No pattern of dose response was observed for any isolated LD subtype.

Coffee and tea consumptions were not associated with increased odds of any isolated LD subtype (Table 3). Tea consumption at 1 cup/month to 6 cups/week and 1 to 2 cups/day were inversely related to the odds of isolated longitudinal LDs (adjusted odds ratios [ORs], 0.6). Soda consumption was moderately associated with all isolated LDs (aORs, 1.2–1.4). The OR was marginally statistically significantly elevated for the association between isolated longitudinal LDs (aOR, 1.6; 95% CI, 1.0–2.5) and three or more servings of soda per day.

None of the MCA LD subtypes were associated with total caffeine, coffee, or tea consumption, except that consumption of 1 or 2 cups of coffee per day was marginally associated with decreased odds of MCA with longitudinal LDs and 1 cup per day was marginally associated with MCA with preaxial LDs compared to the referent (Table 4).

Odds of MCA with longitudinal LDs and MCA with preaxial LDs were elevated with soda consumption of 1 serving/month to <1 serving/day and 1 to 2 servings/day, with similar magnitudes between the two exposure categories (Table 4). However, the ORs decreased for the three or more servings/day category. Similarly, odds of MCA with transverse LDs were elevated among the 1 serving/month to < 1 serving/day (OR, 1.5; 95% CI, 0.8–2.6) soda consumption category, and then decreased in the higher exposure categories (Table 4).

To assess additive interaction for active maternal smoking, LD cases and control infants whose mothers were nonsmokers and had no to little dietary caffeine intake (0–<10 mg/day) served as the reference group. Compared to the reference group, the ORs across caffeine levels were similar among smoking and nonsmoking mothers. A pattern of less than additive interaction for caffeine consumption and active maternal smoking was observed for all isolated LDs combined, isolated transverse LDs, and all MCA LDs combined (Table 5). The ORs for smoking only (mothers who smoked and consumed little or no caffeine) were higher than the ORs for smoking and caffeine consumption for all intake categories. However, the RERI were not statistically significant. We did not observe additive interaction for environmental smoking exposure; neither did we observe effect modification for maternal alcohol consumption or vasoconstrictive medication after investigation.

Patterns of associations were not different from the main analyses when the analyses were restricted to mothers who completed the interview within 12 months of EDD (25 cases and 663 controls), or mothers reported no caffeine contained medication use (833 cases and 7966 controls). The results were also not different between those who had early recognition of pregnancy, planned pregnancy, or nausea/vomiting during early pregnancy (818 cases and 7890 controls) and those who did not meet those criteria (26 cases and 179 controls) (data not shown).

Results of the current study showed that maternal consumption of dietary caffeine was associated with a weak to moderate increased risk for all isolated LDs combined, isolated longitudinal LDs, and isolated transverse LDs. An elevated OR was observed for high soda consumption and isolated longitudinal LDs. The associations for isolated preaxial LD subtype were closer to the null compared to isolated longitudinal LDs. Coffee and tea consumption were not associated with any LD subtype. We did not observe a pattern of dose response for total dietary caffeine intake or for soda consumption. The observed results had no comparison to previous literature because no previous human epidemiology studies have examined caffeine and limb defects.

We suspected that maternal smoking might modify the effect of caffeine because smoking significantly increases the rate of caffeine metabolism by inducing CYP1A2 (Parsons and Neims, 1978; Vistisen et al., 1992). In our study, smoking and consuming caffeine together resulted in a lower risk than the sum of the risk for caffeine consumption alone and smoking alone. However, the ORs were similar across all levels of caffeine among both smokers and nonsmokers, and a higher OR was observed only for smokers who consumed little to no caffeine. Given that smoking status is often correlated with caffeine consumption, relatively few mothers smoked but consumed little to no caffeine and may differ with regard to potential confounding characteristics. Therefore, unmeasured or residual confounding and chance may have influenced the observed associations.

About 18% of women have total average caffeine intake of <10 mg per day (the referent group), meaning that on most days they drink no caffeinated coffee, caffeinated tea, or caffeinated sodas. It is possible that these women who avoid caffeine have healthier diets in terms of fruits and vegetables. That might explain the elevated ORs for those who have more usual intake of caffeine.

Besides estimating the dietary caffeine in mg/day, we categorized the frequency of consumption of coffee, tea, and soda and conducted the beverage-specific analyses to evaluate whether it was caffeine or other components in those beverages that might be playing a role in any observed associations. The association with soda and LDs is worth noting given the lack of association for coffee or tea. It might be explained by other constituents in these beverages, such as antioxidants present in coffee and tea that may be protective and that the sugar present in sodas may increase risk.

An important concern in this study is potential nondifferential error in the classification of caffeine consumption. The NBDPS measured caffeine intake in the year before pregnancy to better represent the actual exposure during the critical period of organogenesis (third–eighth weeks). According to previous studies, a large proportion of women change their caffeine consumption after pregnancy recognition or as the result of nausea or vomiting (Lawson et al., 2002), either of which can happen after limb development starts (Bayley et al., 2002). Retrospective assessment of women’s consumption in the first trimester may reflect that changed pattern of consumption. Caffeine intake in the year before pregnancy was considered to represent intake during the critical period which is the fourth to fifth gestational week (Sadler, 2009); however, women might start to change or stop their caffeine intake sometime before conception due to pregnancy planning or for other reasons. In such cases, the caffeine intake in the year before pregnancy might not reflect the actual consumption during the target period. We assessed this source of potential misclassification by comparing results for mothers who recognized pregnancy early, had an intended pregnancy, or reported nausea or vomiting in the first gestational month to those for mothers who did not meet these conditions. The results of the stratified analysis were not significantly different from the primary analysis.

Another source of potential misclassification is the estimation of caffeine dose. The actual caffeine content was not measured directly in our analysis. Instead, standardized caffeine content was assigned to each unit size of coffee, tea, soda, and chocolate based on previously published guidelines. However, each serving of the beverage might differ widely in caffeine level or dose of intake. For example, caffeine levels in each cup of coffee may vary by brand, type of coffee bean, brewing time and method, serving size, seasonal variation of intake, or patterns of intake.

A complication with the soda classification concerns the reporting of energy drink consumption, which has increased in recent years (Reissig et al., 2009). These drinks were recorded in response to a question about consumption of sodas or soft drinks in the maternal interview for the years of this analysis. Women were not prompted to include energy drinks when they were asked to provide frequency of soda consumption, therefore, if women did not think to report energy drinks as sodas or soft drinks, misclassification of total caffeine intake occurred. With the growing energy drink market (Reissig et al., 2009), it will be of interest to ask specifically about energy drink consumption to ensure adequate recording of caffeine from this source.

The study did not observe a dose-response either for total caffeine intake or for the individual caffeinated beverages. Due to the potential misclassification of caffeine levels, the lack of a dose-response relation for elevated LD risk with increasing total caffeine intake should be interpreted with caution. Theoretically, a ‘dose-response fallacy’ (Selevan and Lemasters, 1987) may exist if a very high dose of caffeine exposure caused infertility or early fetal loss, resulting in that the lower dose of exposure was observed among the fetuses that survived to delivery. Without complete data on early loss and termination from all centers, it is possible that, if congenital anomalies within these outcome categories are related to high doses of caffeine, we may have missed associations for high caffeine consumption and LDs. However, results of animal studies did not show increased risk of embryonic death associated with oral administration of caffeine (Brent et al., 2011). We were not able to conduct separate analysis of postaxial or split hand/foot due to small numbers.

Recall bias might occur if case parents underreported or overreported exposures due to guilt or concern. Given that March of Dimes recommendations are to not consume more than 200 mg/day of coffee (or caffeinated beverages) during pregnancy (March of Dimes, 2011), case mothers might be less likely to report consuming over 2 cups/day, which could be another explanation for no observed dose-response.

The NBDPS reported a slightly higher response rate among cases versus controls (Cogswell et al., 2009). Case and control parents may have different motives for participating in the study; however, the possibility that participation was associated with caffeine intake seems small given that our control mothers had similar distributions for demographic characteristics compared to the base population (Cogswell et al., 2009).

In conclusion, the current study explored the effect of maternal dietary caffeine on the risk of congenital limb deficiencies overall, as well as in subtypes including longitudinal, preaxial longitudinal, and transverse limb deficiencies. We observed a moderate increase in the risk for limb deficiencies overall and for transverse LDs with maternal dietary caffeine consumption. Maternal soda consumption of one to two servings per day was associated with an elevated risk of MCA preaxial longitudinal LDs, and soda consumption of three or more servings per day was associated with elevated risk of isolated longitudinal LDs. No dose-response pattern was observed and no risk increase was observed for maternal coffee or tea consumption. Risk of isolated LDs overall, isolated transverse LD subtype, and MCA LDs overall associated with caffeine consumption above the lowest intake level among active tobacco smokers did not differ from that among nonsmokers. Future studies might improve exposure assessment by collecting more detailed information on change in consumption in early pregnancy, more accurately measuring caffeine contents in beverages, and measuring energy drink intake in addition to intake of soda and other soft drinks. Genetic epidemiologic studies including gene-caffeine and gene-gene interaction in the CYP1A2 and NAT2 genes would be of interest to further explore any relationship between caffeine and LDs.