Association of Clomiphene and Assisted Reproductive Technologies With the Risk of Neural Tube Defects

Abstract

Clomiphene and assisted reproductive technologies (ART) are methods used to help subfertile couples become pregnant. ART has been reported to be associated with neural tube defects (NTDs) in offspring. To evaluate these associations, we studied mothers of 219 cases and 4,262 controls from the Slone Epidemiology Center Birth Defects Study (1993–2012) who were interviewed within 6 months after delivery about pregnancy events, including use of fertility treatments. We considered exposures to clomiphene (without ART) and ART during the periconceptional period. Logistic regression models were used to calculate adjusted odds ratios and 95% confidence intervals, controlling for education and study center. We observed elevated adjusted odds ratios of 2.1 (95% confidence interval: 0.9, 4.8) and 2.0 (95% confidence interval: 1.1, 3.6) for clomiphene and ART exposure, respectively. We performed a mediation analysis to assess whether the observed elevated NTD risk was mediated through multiple births. For clomiphene exposure without ART use, the direct effect estimate of the adjusted odds ratio (aOR_DE) was 1.7 and the indirect effect estimate (aOR_IE) was 1.4. Conversely, for ART exposure, the aOR_DE was 0.9 and the aOR_IE was 2.5. Our findings suggest that relatively little of the clomiphene-NTD association is mediated through the pathway of multiple births, while the ART-NTD association was explained by the multiple-births pathway.

Neural tube defects (NTDs) are birth defects which develop within the first 28 days after conception, when the neural tube fails to properly close (1). The spectrum of NTDs includes spina bifida, anencephaly, and encephalocele, which occur in approximately 6.4 per 10,000 live births in the United States (2). Consumption of folic acid (FA) during the periconceptional period has been demonstrated to decrease the risk of NTDs (3–6), which prompted the fortification of cereal grains with FA in the United States and Canada in 1998 (7). Although the prevalence of NTDs has decreased since then, NTDs continue to occur, even among infants whose mothers have consumed the recommended amount of FA (400 μg/day) (8). Thus, other causal mechanisms are likely to play a role in NTD development.

Clomiphene, a medication used to induce ovulation among women with subfertility, was raised as a potential risk factor when case reports of NTDs associated with clomiphene exposure were published in the 1970s (9, 10). Prompted by these case reports, epidemiologists investigated this potential association, but studies have yielded conflicting results (11–19). Risk estimates for observed positive associations ranged from 2.3 to 11.7, but studies were limited by an inability to separate the potential effect of the underlying subfertility from the treatment effects (13–15) and by small numbers, which did not allow for proper control of confounding (14). When ovulation induction with clomiphene fails to result in a viable pregnancy, the next line of treatment includes assisted reproductive technologies (ART) such as in vitro fertilization, intrauterine insemination, and intracytoplasmic sperm injection. Similar to clomiphene, ART has been implicated as a possible risk factor for NTDs, and findings for ART have also been inconclusive (20–26).

Due to the interconnectedness between subfertility and fertility treatments, it has been hypothesized that the underlying subfertility is the causal mechanism for the association of both clomiphene and ART with NTDs (12, 14, 21, 27, 28). This hypothesis has received support from studies comparing subfertile couples (defined as having a previous treatment or medical visit for infertility or a time to pregnancy of >12 months) who conceived without assistance to couples without subfertility; the former group was estimated to have an increased risk of central nervous system defects and NTDs (14, 21, 27, 28). However, conflicting evidence has been reported (12).

Utilizing data from the Slone Epidemiology Center Birth Defects Study, we evaluated the hypothesis that the risk of NTDs is increased by periconceptional exposure to clomiphene or ART in conjunction with maternal subfertility. We also investigated whether FA intake modifies these associations. Lastly, because fertility treatments are associated with multiple births, which are in turn associated with NTDs (29), we explored whether any observed clomiphene-NTD and ART-NTD associations were mediated through multiple births.

METHODS

The Boston University Slone Epidemiology Center Birth Defects Study was initiated in 1976 as an ongoing (40-year) North American case-control study focused on the risks and safety of antenatal medication exposure in relation to specific birth defects; it has been described in detail elsewhere (8, 30). For this analysis, cases consisted of infants born with anencephaly, encephalocele, or spina bifida, excluding those with an accompanying conjoined twin, a chromosomal anomaly, a Mendelian inherited disorder, a known syndrome, amniotic bands, or a body wall defect. Cases were ascertained from participating birth hospitals and tertiary-care centers in the greater metropolitan areas of Boston, Massachusetts; Philadelphia, Pennsylvania; San Diego County, California; and Toronto, Ontario, Canada. Research staff identified cases by reviewing admission and discharge lists and clinical and surgical logs, and by contacting newborn nurseries and labor and delivery rooms. Mothers of cases in New York State and in Massachusetts (since 1998) were identified from statewide birth-defect registries. The study protocol has been approved by the institutional review board at Boston University and other institutions, as appropriate. Cases include live births, fetal deaths occurring after 20 weeks' gestation, and therapeutic abortions performed after 12 weeks' gestation; however, fetal deaths and therapeutic abortions were not systematically ascertained. Beginning in 1993, nonmalformed liveborn infants were ascertained from the same birth population as the cases to serve as controls.

Maternal interviews were conducted by trained study nurses within 6 months of delivery, fetal loss, or termination; interviews were conducted in person prior to 1998 and by telephone thereafter. The interview includes questions on sociodemographic factors, reproductive history, illness during pregnancy, behavioral risk factors (e.g., smoking or alcohol use), and medication use at any time from 2 months before pregnancy through the end of pregnancy. Between 1988 and 1998, dietary habits during the period from 6 months before pregnancy to the last menstrual period (LMP) were assessed using the long version of the Willett food frequency questionnaire; subsequently, it was replaced with a modified shortened version (31).

The present study included mothers interviewed between 1993 and 2012, during which time the study questionnaire contained questions pertaining to subfertility and related treatments. Cases and controls eligible for analysis were restricted to study hospitals in which both a case and a control were identified. Cases and controls were also restricted to mothers who indicated that the index pregnancy was planned.

In the interview, mothers were asked whether they had ever received a medical evaluation, medication, or treatment for difficulty becoming pregnant and, if so, whether it was related to the index pregnancy. Affirmative responders were asked what treatments, procedures, or medications they or their partners had received anytime from 2 lunar months before the LMP through the end of pregnancy. Mothers who reported subfertility for the index pregnancy—defined as a woman's reporting that she had sought a medical evaluation for difficulty becoming pregnant—and indicated taking clomiphene anytime between 1 month before the LMP to 1 month after the LMP, and who reported not taking any other fertility medications or utilizing any ART during the same period, were assigned to the “subfertile and clomiphene only” group; those who reported uncertain dates of clomiphene use were excluded. ART exposure was defined as treatment in which the sperm and/or the egg/embryo were handled medically, such as intracytoplasmic sperm injection, intrauterine insemination, or in vitro fertilization. Mothers who reported subfertility for the index pregnancy and indicated having undergone ART treatment during the period from 1 month before the LMP to 1 month after the LMP were considered exposed to ART, regardless of exposure to clomiphene or other fertility drugs. Mothers who reported no use of clomiphene, other fertility drugs, or ART during the period from 1 month before the LMP to 1 month after the LMP and reported subfertility were assigned to the “untreated subfertility” group, while those who did not report subfertility for the index pregnancy were considered unexposed and constituted the reference category.

Multiple logistic regression models were used to estimate relative risks with crude and adjusted odds ratios (cORs and aORs, respectively) and 95% confidence intervals. When logistic models failed to converge due to quasi-separated data caused by sparsely distributed potential confounders, resulting in at least 1 zero cell when data were stratified by the potential confounders, Firth logistic regression was used (32–34), since it can overcome model nonconvergence due to reliance upon a penalized maximum likelihood rather than a maximum likelihood. Potential confounders were identified through the use of directed acyclic graphs (35). Factors that were considered included: maternal race/ethnicity (non-Hispanic white, non-Hispanic black, Hispanic, or other); maternal age (<20, 20–24, 25–29, 30–34, or ≥35 years); maternal education (<12 years, 12 years, or >12 years); study center, by year (Boston/Massachusetts 1993–1997; Boston/Massachusetts 1998+; Philadelphia, Pennsylvania 1993–1997; Philadelphia, Pennsylvania 1998+; Toronto, Ontario, Canada 1993–1997; Toronto, Ontario, Canada 1998+; San Diego County, California 1998+; New York State 1998+); body mass index (weight (kg)/height (m)²; underweight (<18.5), normal-weight (18.5–<25), overweight (25–<30), or obese (≥30)); periconceptional FA intake (<400 μg/day or ≥400 μg/day); parity (0 or ≥1); number of previous miscarriages (0 or ≥1); pregestational diabetes mellitus (yes or no); history of convulsions or seizures (yes or no); pregestational hypertension (yes or no); and smoking during the exposure period (yes or no). The factors were added into the model one at a time, and if a variable changed the cOR estimates by more than 10%, it was retained in the final model. For women who had missing body mass index data, multiple imputation methods utilizing 10 data sets were used to impute body mass index (36). The following variables were used to impute body mass index: study center, height, weight, maternal age, race, education, pregestational diabetes status, gestational diabetes status, birth year, maternal nulliparous status, and exposure status.

FA intake was determined from the summation of the average daily FA intakes from fortified foods from 6 months before conception to LMP and supplements taken from 1 month before the LMP to 1 month after the LMP. Natural folate was included in the average but discounted by 30% because of its lower bioavailability (37). Using the residual energy adjustment method (38), we adjusted all dietary variables for total caloric intake. Women were categorized according to their total FA intake (<400 μg/day or ≥400 μg/day). Participants with extreme caloric intakes (<500 kcal/day or >4,000 kcal/day) or incomplete food frequency questionnaires (≥3 missing items) were excluded from analyses involving FA intake, with 2 exceptions: Women who reported ≥400 μg of FA per day from vitamin supplements or who reported not taking any supplements containing FA were retained in the analysis. The former group was retained because women taking FA-containing vitamins would almost always consume at least 400 μg of FA per day regardless of dietary intake. The latter group was retained on the basis of previous studies indicating that nonusers would probably not reach the ≥400 μg/day stratum from diet alone (7); therefore, they were placed in the <400 μg/day category. Analyses stratified according to periconceptional FA intake were performed to assess potential effect modification. Finally, we estimated the joint effects of low FA intake and each exposure—clomiphene and ART—on NTDs.

We then performed a mediation analysis treating multiple births as a potential mediator because both clomiphene and ART can cause multiple births and multiple births are associated with NTDs (29). We utilized a mediation analysis method for a dichotomous outcome within the counterfactual framework to estimate the direct effects of treatments on NTDs and those which operate through multiple births (indirect effects) (39). With this method, we estimated the natural direct effect (DE) aOR (aOR_DE) and the natural indirect effect (IE) aOR (aOR_IE) (39). All analyses were performed using SAS 9.3 software (SAS Institute, Inc., Cary, North Carolina) (40).

In order to assess the impact of unmeasured confounding of the potential mediator-outcome relationship, we simulated the presence of a binary confounder, U, causally associated with multiple births and NTDs under 2 different confounding scenarios, moderate and strong. Each simulation was performed 1,000 times. We utilized the ranbin SAS (40) function, which uses a binomial distribution to randomly assign U to individual subjects. For both the moderate and strong confounding scenarios, regardless of exposure status, a probability of having U of 10% was assigned to non-NTD singleton births. Under the moderate scenario of confounding, a probability of having U of 20% was assigned to non-NTD multiple births and to NTD singleton births. Lastly, NTD multiple births were assigned a probability of 40%. Under the strong confounding scenario, a probability of having U of 20% was assigned to non-NTD multiple births. A probability of 40% was assigned to NTD singleton births, and NTD multiple births were assigned a probability of 80%. After each iterative simulation, the aOR_DE and aOR_IE were calculated for both clomiphene and ART exposure, controlling for U as well as the other identified potential confounders.

To assess whether incomplete inclusion of therapeutic abortions might bias our observed results, we applied hypothetical exposure prevalences for nonincluded cases. The hypothetical scenarios appear in the Web Appendix (available at http://aje.oxfordjournals.org/), along with their corresponding corrected odds ratios (Web Table 1).

RESULTS

Initially, 9,593 participants were identified for this study: 610 cases and 8,983 controls (Figure 1). A total of 1,372 infants (95 cases and 1,277 controls) were excluded because they could not be matched to a birth hospital, and 3,601 (272 cases and 3,329 controls) were excluded because their mothers had not planned their pregnancy. An additional 43 infants (4 cases and 39 controls) were then excluded due to unclear information on timing of clomiphene or ART use. Finally, we excluded 86 infants (10 cases and 76 controls) with missing FA data and 10 cases with missing multiple-birth status data. Thus, the final analyses (n = 4,481) included 219 cases and 4,262 controls. Among all mothers who were eligible and were asked to participate, 69% of case mothers and 67% of control mothers agreed to be interviewed (41).

Figure 1.

Selection of the final study population for an analysis of associations between clomiphene and assisted reproductive technologies (ART) and neural tube defects, Slone Epidemiology Center Birth Defects Study, 1993–2012.

The distributions of sociodemographic factors are presented in Table 1 by case status. Cases were more likely to be Hispanic, less educated, and obese, to have lower FA intake, to have had a previous miscarriage, and to have smoked during the periconceptional period. When compared with unexposed controls, both clomiphene- and ART-exposed control women were more likely to be non-Hispanic white, older, more educated, and nulliparous, to have higher FA intake, to have had a previous miscarriage, and to have not smoked during the periconceptional period (Table 2), while untreated subfertile control women were more likely to be Hispanic, older, more educated, and nulliparous, to have a higher FA intake, and to have had a previous miscarriage in comparison with unexposed controls.

Table 1.

Demographic Characteristics of the Mothers of Neural Tube Defect Cases and Controls, Slone Epidemiology Center Birth Defects Study, 1993–2012

	Cases		Controls
	No.	%	No.	%
Total	219		4,262
Maternal race/ethnicity
Non-Hispanic white	162	74.0	3,262	76.5
Non-Hispanic black	10	4.6	184	4.3
Hispanic	34	15.5	511	12.0
Other	13	5.9	301	7.1
Missing data	0	0.0	4	0.1
Maternal age at conception, years
<20	6	2.7	81	1.9
20–24	22	10.0	380	8.9
25–29	76	34.7	1,197	28.1
30–34	75	34.2	1,677	39.3
≥35	40	18.3	917	21.5
Missing data	0	0.0	10	0.2
Maternal education, years
<12	25	11.4	208	4.9
12	41	18.7	563	13.2
>12	153	69.9	3,489	81.9
Missing data	0	0.0	2	0.1
Study center (by year)
Boston, Massachusetts (1993–1997)	38	17.4	86	2.0
Boston, Massachusetts (1998+)	24	11.0	1,638	38.4
Philadelphia, Pennsylvania (1993–1997)	16	7.3	102	2.4
Philadelphia, Pennsylvania (1998+)	44	20.1	922	21.6
Toronto, Ontario, Canada (1993–1997)	31	14.2	188	4.4
Toronto, Ontario, Canada (1998+)	22	10.1	381	8.9
San Diego, California (1998+)	23	10.5	748	17.6
New York State (1998+)	21	9.6	197	4.6
Body mass indexa
Underweight	13	5.9	231	5.4
Normal-weight	112	51.1	2,732	64.1
Overweight	50	22.8	804	18.9
Obese	36	16.4	416	9.8
Missing data	8	3.7	79	1.9
Folic acid intake, μg/day
<400	113	51.6	1,510	35.4
≥400	106	48.4	2,752	64.6
Nulliparous
Yes	98	44.7	1,925	45.2
No	121	55.3	2,331	54.7
Missing data	0	0.0	6	0.1
No. of miscarriages
0	142	64.8	3,168	74.3
≥1	76	34.7	1,091	25.6
Missing data	1	0.5	3	0.1
Multiple birth
No	200	91.3	4,146	97.3
Yes	19	8.7	116	2.7
History of diabetes mellitus, seizures, or hypertension
No	209	95.4	4,184	98.2
Yes	10	4.6	78	1.8
Maternal smoking
No	192	87.7	3,926	91.4
Yes	27	12.3	370	8.6

^a Weight (kg)/height (m)².

Table 2.

Demographic Characteristics of the Mothers of Controls According to Fertility Treatment Exposure Status, Slone Epidemiology Center Birth Defects Study, 1993–2012

	Fertility Treatment Exposure
	Unexposed		Untreated Subfertility		Subfertile and Exposed to Clomiphenea		Subfertile and Exposed to ARTb
	No.	%	No.	%	No.	%	No.	%
Total	3,700		250		69		243
Maternal race/ethnicity
Non-Hispanic white	2,785	75.3	199	79.6	63	91.3	215	88.5
Non-Hispanic black	167	4.5	11	4.4	1	1.4	5	2.1
Hispanic	475	12.8	22	8.8	4	5.8	10	4.1
Other	269	7.3	18	7.2	1	1.4	13	5.3
Missing data	4	0.1	0	0.0	0	0.0	0	0.0
Maternal age at conception, years
<20	79	2.1	2	0.8	0	0.0	0	0.0
20–24	348	9.4	24	9.7	3	4.3	5	2.1
25–29	1,098	29.7	52	20.8	18	26.1	29	11.9
30–34	1,454	39.3	102	40.8	27	39.1	94	38.7
≥35	715	19.3	68	27.2	20	29.0	114	46.9
Missing data	6	0.2	2	0.8	1	1.4	1	0.4
Maternal education, years
<12	200	5.4	6	2.4	1	1.4	1	0.4
12	516	14.0	23	9.2	6	8.7	18	7.4
>12	2,982	80.5	221	88.4	62	89.9	224	92.2
Missing data	2	0.1	0	0.0	0	0.0	0	0.0
Study center (by year)
Boston, Massachusetts (1993–1997)	82	2.2	4	1.6	0	0.0	0	0.0
Boston, Massachusetts (1998+)	90	2.4	1	0.4	23	33.3	151	62.1
Philadelphia, Pennsylvania (1993–1997)	174	4.7	8	3.2	3	4.4	8	7.8
Philadelphia, Pennsylvania (1998+)	1,356	36.7	108	43.2	17	24.6	45	18.5
Toronto, Ontario, Canada (1993–1997)	804	21.7	56	22.4	6	8.7	0	0.0
Toronto, Ontario, Canada (1998+)	346	9.4	20	8.0	7	10.1	8	3.3
San Diego, California (1998+)	679	18.4	45	18.0	9	13.0	15	6.2
New York State (1998+)	169	4.6	8	3.2	4	5.8	16	6.6
Body mass indexc
Underweight	208	5.6	13	5.2	3	4.3	7	2.9
Normal-weight	2,375	64.2	153	61.2	42	60.9	162	66.7
Overweight	694	18.8	53	21.2	15	21.7	42	17.3
Obese	349	9.4	29	11.6	8	11.6	30	12.3
Missing data	74	2.0	2	0.8	1	1.4	2	0.8
Folic acid intake, μg/day
<400	1,404	38.0	72	28.8	15	21.7	19	7.8
≥400	2,296	62.0	178	71.2	54	78.3	224	92.2
Nulliparous
No	2,127	57.5	95	38.0	28	40.6	81	33.3
Yes	1,572	42.5	155	62.0	41	59.4	158	65.0
Missing data	1	0.0	0	0.0	0	0.0	4	1.6
Previous miscarriage
No	2,799	75.1	167	67.3	45	65.2	157	64.6
Yes	899	24.9	83	32.7	24	34.8	85	35.0
Missing data	2	0.1	0	0.0	0	0.0	1	0.4
Multiple birth
No	3,662	99.0	246	98.4	63	91.3	175	72.0
Yes	38	1.0	4	1.6	6	8.7	68	28.0
History of diabetes mellitus, seizures, or hypertension
No	3,631	98.1	246	98.4	69	100.0	238	97.9
Yes	69	1.9	4	1.6	0	0.0	5	2.1
Maternal smoking
No	3,400	91.0	226	90.4	65	94.2	235	96.7
Yes	300	9.0	24	9.6	4	5.8	8	3.3

Abbreviation: ART, assisted reproductive technologies.

^a Women classified as exposed to clomiphene were not exposed to any ART or other fertility drugs.

^b Women classified as exposed to ART could also have been exposed to fertility drugs.

^c Weight (kg)/height (m)².

Of the 219 cases, 3.2% were exposed to clomiphene only, 6.4% were exposed to ART with or without use of other fertility medication, and 5.5% reported untreated subfertility (Table 3). Lower proportions of controls were exposed to clomiphene (1.6%), while similar proportions were exposed to ART (5.7%) and subfertile (5.9%). Among cases with ART exposure, 1 infant was conceived via intracytoplasmic sperm injection, 9 were conceived via intrauterine insemination, and 4 were conceived via in vitro fertilization. Among controls, 6 infants were conceived via intracytoplasmic sperm injection, 94 via intrauterine insemination, and 143 via in vitro fertilization (data not shown). Furthermore, cases had a higher proportion of multiple births than controls (8.7% and 2.7%, respectively).

Table 3.

Associations Between Neural Tube Defects and Subfertility, Clomiphene Use, and ART Use During the Period From the First Month Before the LMP to 1 Month After the LMP, Slone Epidemiology Center Birth Defects Study, 1993–2012

Exposure Status	Cases (n = 219)		Controls (n = 4,262)		cOR	95% CI	aORa	95% CI
	No.	%	No.	%
Untreated subfertility	12	5.5	250	5.9	1.0	0.5, 1.7	1.2	0.6, 2.2
Subfertile and clomiphene use only	7	3.2	69	1.6	2.0	0.9, 4.5	2.1	0.9, 4.8
Subfertile and ART use	14	6.4	243	5.7	1.2	0.7, 2.0	2.0	1.1, 3.6
Unexposed	186	84.9	3,700	86.8	1.0	Referent	1.0	Referent

Abbreviations: aOR, adjusted odds ratio; ART, assisted reproductive technologies; CI, confidence interval; cOR, crude odds ratio; LMP, last menstrual period.

^a Adjusted for maternal education and study center (by year).

For our analysis, maternal education and study center by year met the criteria for confounding. In the adjusted model, we observed elevated aORs for the associations between NTD risk and clomiphene use (aOR = 2.1, 95% confidence interval (CI): 0.9, 4.8) and ART exposure (aOR = 2.0, 95% CI: 1.1, 3.6) (Table 3). Furthermore, no association was identified among women reporting subfertility and no treatment (aOR = 1.2, 95% CI: 0.6, 2.2).

The estimated joint effect of low FA intake (<400 μg/day) and clomiphene use was strongly associated with NTD risk in comparison with women who were fertile and used no fertility treatment and had a higher FA intake (≥400 μg/day) (aOR = 8.0, 95% CI: 3.0, 21.3) (Table 4). With the same comparison group, we observed an elevated aOR for the combination of low FA intake and ART exposure in risk of NTDs (aOR = 2.2, 95% CI: 0.5, 10.6); however, the 95% confidence interval included 1, and there were only 2 case mothers who were doubly exposed. Additionally, untreated subfertility remained unassociated with NTDs in both FA strata.

Table 4.

Interdependence Between Subfertility, Clomiphene Use, ART Use, and Folic Acid Intake in the Risk of Neural Tube Defects, Slone Epidemiology Center Birth Defects Study, 1993–2012

Exposure Status and Folic Acid Intakea	Cases		Controls		aORb	95% CI
	No.	%	No.	%
No fertility treatment and not subfertile
High	93	42.5	2,296	53.8	1.0	Referent
Low	93	42.5	1,404	32.9	1.1	0.8, 1.5
Untreated subfertility
High	8	3.6	178	4.2	1.2	0.6, 2.7
Low	4	1.8	72	1.7	1.1	0.4, 3.4
Subfertile and clomiphene use
High	0	0.0	54	1.3	—c	—
Low	7	3.2	15	0.4	8.0	3.0, 21.3
Subfertile and ART use
High	12	5.5	224	5.2	2.0	1.1, 3.8
Low	2	0.9	19	0.5	2.2	0.5, 10.6

Abbreviations: aOR, adjusted odds ratio; ART, assisted reproductive technologies; CI, confidence interval.

^a Low folic acid intake: <400 μg/day; high folic acid intake: ≥400 μg/day.

^b Adjusted for education and study center (by year).

^c Not calculated because no cases were observed in this stratum.

In our mediation analysis, we controlled for maternal education and study center by year. For clomiphene exposure, we estimated an aOR_DE of 1.7 (95% CI: 0.7, 4.0) and an aOR_IE of 1.4 (95% CI: 1.0, 2.0) (Table 5). When we considered ART exposure, we observed an aOR_DE of 0.9 (95% CI: 0.4, 1.8), while the aOR_IE was 2.5 (95% CI: 1.5, 4.0).

Table 5.

Estimates of Direct and Indirect Effects of Clomiphene and ART Use on Risk of Neural Tube Defects, Mediated Through Multiple Births, Slone Epidemiology Center Birth Defects Study, 1993–2012

Exposure Status	Singleton Birth		Multiple Birth		Direct Effect		Indirect Effect
	No. of Cases	No. of Controls	No. of Cases	No. of Controls	aORa,b	95% CI	aORa,b	95% CI
Subfertile and clomiphene use only	6	63	1	6	1.7	0.7, 4.0	1.4	1.0, 2.0
Subfertile and ART use	4	175	10	68	0.9	0.4, 1.8	2.5	1.5, 4.0
Unexposed	178	3,662	8	38	1.0	Referent	1.0	Referent

Abbreviations: aOR, adjusted odds ratio; ART, assisted reproductive technologies; CI, confidence interval.

^a Adjusted for education and study center (by year).

^b Calculated using Firth logistic regression.

In our sensitivity analysis, under the scenario of a moderate confounder, U, the cORs for the association between U and multiple births were between 1.2 and 4.6 (Web Figure 1), while the cORs for the association between U and NTDs were between 1.2 and 4.2 (Web Figure 2). Under the scenario of a strong confounder, U, the cORs for the association between U and multiple births were between 1.6 and 5.3 (Web Figure 3), while the cORs for the association between U and NTDs were between 4.5 and 10.5 (Web Figure 4).

For clomiphene exposure, under the moderate confounding scenario, the aOR_DE 95% simulation interval ranged from 1.5 to 1.9 (Web Figure 5), and the aOR_IE 95% simulation interval ranged from 1.2 to 1.4. Under the strong confounding scenario, the aOR_DE 95% simulation interval ranged from 1.3 to 2.2, and the 95% simulation interval for the aOR_IE ranged from 1.1 to 1.3 (Web Figure 6).

For ART exposure, under the moderate confounding scenario, for aOR_DE we observed a 95% simulation interval of 0.8 to 0.9 (Web Figure 7). For estimates of aOR_IE, we observed a 95% simulation interval of 2.0 to 2.4. Under the strong confounding scenario, for aOR_DE we observed a 95% simulation interval of 0.7 to 1.0 (Web Figure 8) and for aOR_IE a 95% simulation interval of 1.6 to 2.2.

DISCUSSION

In this large, geographically diverse case-control study spanning the years before and after food fortification with FA, we confirmed previous reports of a positive association between fertility treatments and NTDs. Exposures to both clomiphene alone and ART remained modestly associated with NTD risk after controlling for potential confounders. These associations were strongest for women who used fertility treatments and did not consume the recommended amount of FA.

Other observational studies have also found increased risks of NTDs in association with clomiphene use; these risk estimates ranged from 1.8 to 11.7 (13–15, 18). However, null associations have been observed in other studies (12, 16, 17, 19). In our analysis, women who used clomiphene in combination with low FA intake had the greatest odds of having an NTD-affected infant. Despite our finding and other observed associations between clomiphene and NTDs, a specific causal mechanism is not clear.

As in the present analysis, other observational studies have found increased odds ratios for the relationship between NTDs and ART, ranging from 2.9 to 12.9 (20, 23–25). However, null findings have also been reported (21, 22, 26).

Our findings regarding spontaneous conception among subfertile women conflict with the majority of the current literature (14, 21, 27, 28); however, when we compared the methods with which we defined subfertility to those of the other studies, we noted that we did not use a time criterion, which may have resulted in lowered specificity. We would expect this misclassification mechanism to have been nondifferential; however, if our misclassification were only “approximately nondifferential” (42), it would be possible that our true association would be closer to previously reported associations.

When we considered the direct and indirect ART-NTD associations, we observed that the estimated total effect was fully explained by the multiple-births pathway. When we assessed these associations for bias from unmeasured confounding, we observed little deviation from our observed results. Furthermore, comparisons of the estimated indirect effects of clomiphene and ART suggested that these exposures operate through different pathways—ART through multiple births and clomiphene through another pathway. However, there may remain pathways by which ART is independently associated with NTDs and NTD subtypes. In particular, a wide range of procedures and medications may be involved in ART, and associations with birth defects may well vary for specific regimens such as freezing and thawing of embryos (43) or combinations of ovulation inducers (44). However, we were unable to evaluate specific ART procedures because of small numbers.

Based upon the results of this study and others, clinicians should discuss the potentially increased risk of a fetal NTD with women considering fertility treatments, noting that even a doubling in risk would translate to a low absolute risk. In our study, we did not find any women who reported clomiphene exposure and FA intake of ≥400 μg/day. This could indicate that taking the recommended amount of FA may minimize the associated risk. As for ART exposure, we observed that the association was explained by the multiple-births pathway, indicating that the observed association may weaken if fewer embryos are transferred.

Our study had both strengths and weaknesses. A strength of our study was our ability to assess the potential effect of unmeasured confounding within our mediation analysis through the use of simulations. A limitation of our study was the potential for maternal recall bias. However, the use of fertility treatments preceding the index pregnancy would presumably be an important life event, resulting in accurate reporting of exposures and reporting being similar between cases and controls. Further, interviews were conducted within 6 months of birth by a trained nurse utilizing a highly structured and standardized questionnaire that asked about specific fertility drugs and treatments. Another limitation is the possibility of improper control selection. We believe this was minimized by matching of cases and controls on birth hospital so that the controls represented the same source population as that which gave rise to the cases. Also, incomplete ascertainment of terminated cases was probable. However, as shown in the Web Appendix, even under extreme conditions, differential exposure among nonincluded cases probably would not explain the observed associations. Finally, although our study included over 219 NTD cases and was sufficiently robust to permit evaluation of clomiphene and ART associations overall, it was too small to allow evaluation of different ART methods.

A limitation of all observational studies of subfertility treatments is the inability to separate the possible effect of the underlying subfertility on risk of NTDs. If previous findings showing that untreated subfertility itself increases NTD risk (14, 28) are true, our observed elevated risks could have been driven, at least partly, by the underlying subfertility; however, the clomiphene and ART associations would remain for treated women.

In our investigation of the association between NTDs and the use of clomiphene and ART during the periconceptional period, we observed elevated NTD risks associated with exposure to both clomiphene alone and ART, although most of the association with ART appeared to be mediated via the indirect NTD risk among multiple births. Further, women who did not consume the recommended amount of FA (≥400 μg/day) and were exposed to either clomiphene alone or ART had the greatest risk of NTDs, but these odds ratios were based on small numbers and were unstable.