The Birth Certificate as an Efficient Means of Identifying Children Conceived With the Help of Infertility Treatment

Abstract

Recent work suggests that infertility treatment is associated with adverse child health outcomes. In exploring various methods of assembling a cohort of children conceived by infertility treatment, the authors conducted a validation study of the assisted reproductive technology and infertility drug use check boxes on the Massachusetts birth certificate. Using 2001 and 2002 data, the authors conducted telephone interviews with 399 women whose child's birth certificate had at least one of the boxes checked along with 185 women who were over age 42 years or who delivered twins or higher order multiples to compare the check box information with maternal report. Among the 579 women with available information, the birth certificate was fully concordant with respect to infertility treatment status for 271 (47%) women, partially concordant for 248 (43%) women, and discordant for 60 (10%) women. Agreement between the birth certificate and maternal report was good for singletons (weighted kappa = 0.66) but was found to be very poor among twins and higher order multiples (weighted kappa = 0.05). The authors concluded that birth certificates are an efficient means of locating children conceived with the help of infertility treatment but that they are not appropriate for identifying type of treatment.

Since 1983, when the first in vitro fertilization resulting in a livebirth occurred in the United States, the use of assisted reproductive technologies has increased markedly (1). In 2006, the most recent data available, there were 138,198 assisted reproductive technology procedures performed in the United States, resulting in 54,656 liveborn infants (approximately 1% of the infants born that year) (1).

The weight of scientific evidence supports that children conceived with assisted reproductive technologies are at increased risk of adverse outcomes. Results from a meta-analysis identified 5 significant adverse perinatal outcomes associated with such procedures—perinatal mortality (odds ratio (OR) = 2.19), preterm delivery (OR = 1.94), low birth weight delivery (OR = 1.77), very low birth weight delivery (OR = 2.69), and small for gestational age (OR = 1.56)—even after controlling for maternal age and parity (2). A key question is whether this increase in adverse outcomes is attributable to the assisted reproductive technology treatment itself or whether it is due to the women's underlying impaired fecundity. Schieve et al. (3) reported that singleton assisted reproductive technology pregnancies are at greater risk for low birth weight when born at term than nonassisted reproductive technology pregnancies (OR = 2.6), but no difference was observed for multiples (OR = 1.0). When they stratified by the use of donor versus nondonor eggs, an increased risk remained (OR = 1.9, 95% confidence interval (CI) = 1.5, 2.3), suggesting that assisted reproductive technology treatment itself may play a role. Further, a study that compared the outcomes of children born following treatment versus those who were conceived spontaneously by subfertile women still found a 3-fold excess of poor outcomes in the infertility-treated group (4).

Assisted reproductive technology procedures represent only a small fraction of the infertility treatments that are used in the United States. In the 2002 National Survey of Family Growth, while 11.9% of women of reproductive age reported having ever sought infertility services, only 0.3% reported ever having used assisted reproductive technologies (5). Among those women who reported having sought services, fertility drugs were used 12.7 times more often than assisted reproductive technologies, and intrauterine insemination was used 3.7 times more frequently (5). The percentage of these additional treatments that resulted in a livebirth is unclear, however, as they are not subject to the same reporting requirements as assisted reproductive technology procedures.

There is a dearth of information regarding whether ovarian stimulation is associated with an increased risk of adverse outcomes. Although fertility drugs are administered prior to conception, research suggests that they persist in the maternal circulation for several weeks and that they may accumulate in the body over time (6). Much of what we have gleaned thus far about the effect of these fertility drugs comes from small, uncontrolled descriptive studies. Two of these studies have suggested that infants born following ovulation induction may be at increased risk of preterm delivery or neural tube defects (6, 7).

Although prior research is certainly suggestive that the association between assisted reproductive technology therapies and ovulation-inducing drugs and increased perinatal outcomes may be causative, the available evidence is inconclusive (8). The ability to do population-based research of adequate size in the United States is impaired by the absence of a suitable registry or sampling framework for study, given that many of the outcomes of interest are not accurately reported on birth records. To address this methodological challenge, we evaluated the use of the birth certificate as a sampling frame to identify children conceived with the help of infertility treatment.

MATERIALS AND METHODS

In Massachusetts, birth certificate data are collected via the hospital birth certificate worksheet, which is completed by hospital personnel with the assistance of the mother's labor and delivery record. In the prenatal tests and procedures section of that worksheet, there are 2 infertility treatment-related check boxes, one for assisted reproductive technologies (alternative methods of conception) and one for fertility drugs. We selected all of the 2001 and 2002 birth certificates that indicated assisted reproductive technology use only, fertility drug use only, or both fertility drug and assisted reproductive technology use. If a woman was found to have contributed more than one pregnancy during the study period, only the first qualifying pregnancy selected was used (n = 4). Care was also taken to ensure that we did not include more than one infant from any given multifetal pregnancy (i.e., the unit of analysis was the woman).

We also selected a smaller sample of birth certificates from among births that are likely to have resulted from infertility treatment, but yet the relevant birth certificate check boxes were not checked. Specifically, we sampled among all 2001 and 2002 twin and higher order deliveries and infants born to mothers over age 42 years. The purpose of this targeted sample was to obtain a worst case scenario estimate of the degree of underreporting of infertility treatment on the birth certificate.

Selected mothers were sent a standard letter explaining the objectives of the research. The letter further indicated that a research assistant would be calling in approximately 2 weeks to explain the purpose of the study in more detail and to conduct or schedule a 15-minute interview. Should a woman not be interested in learning more, she was provided a postage-paid return postcard and a toll-free telephone number to call to opt out of the study. We sent a modified letter to women for whom we were unable to identify a reliable telephone number from the birth certificate and directory searches, to introduce the study and encourage participation. The modified letter asked women to call a toll-free number or return the postage-paid postcard and indicate whether or not they were willing to be contacted. The protocol was reviewed and approved by the appropriate institutional review boards. All women provided informed consent for participation.

Figure 1 provides a flowchart of the sample size and contact and participation rates. A total of 2,493 women were sought for participation, 1,728 (69%) whose child's birth certificate indicated fertility treatment and 765 (31%) whose child's certificate did not indicate treatment. Among those sampled women, we were able to contact slightly less than half of them, 46% in the check box-positive group and 44% in the check box-negative group (P = 0.34). Among those women we successfully contacted, roughly half agreed to participate, 50% in the check box-positive group and 55% in the check box-negative group (P = 0.14). This resulted in a final sample size of 584 mothers who were interviewed from 2004 to 2006.

Figure 1.

Sampling and participation flow chart, Massachusetts, 2001–2002.

Study participation and discordance of birth certificate data with maternal response were assessed for association with maternal and infant characteristics. Differences in the distributions of categorical variables were assessed by using chi-square tests. Differences in continuous variables were assessed by using nonparametric Wilcoxon rank-sum tests. Kappa and weighted kappa statistics were used to assess agreement. Multivariable logistic regression was used to examine the adjusted odds of discordance. Any variables found to be associated with discordance at the 0.10 level in the bivariate analyses were candidates for inclusion in the final model. All reported P values are 2 sided. Analyses were conducted in STATA, version 11, software (StataCorp LP, College Station, Texas).

RESULTS

Table 1 compares the selected maternal and infant characteristics of women whom we successfully contacted versus those we were unable to contact. Women we reached were more likely to be white, a college graduate or more, married, aged 40 years or older, and less likely to have their prenatal care paid by the government. It should be noted that the demographic profile of the women we reached closely mirrors that of women who report seeking infertility treatment in the United States (9).

Table 1.

Selected Maternal and Infant Characteristics by Contact Status, Massachusetts, 2001–2002

	Contacted (n = 1,137)			Not Contacted (n = 1,356)
	%	Median	Range	%	Median	Range
Fertility drug use reported on birth certificate	32.5			30.9
Assisted reproductive technologies reported on birth certificate	50.8			49.3
Plurality
Singletons	80.4			79.6
Twins	18.4			18.1
Triplets or higher	1.2			2.3
Maternal race*
White	93.1			84.4
Black	1.4			5.0
Asian	3.6			5.7
Other	1.9			4.9
Maternal education*
Less than high school graduate	1.8			4.2
High school graduate	24.0			30.9
Some college	12.0			9.6
College graduate or more	62.6			55.3
Maternal age*
<30 years	6.5			11.4
30–34 years	24.5			28.3
35–39 years	26.8			22.4
≥40 years	42.2			38.1
Mother married*	96.6			93.4
Prenatal care paid by the government*	2.6			9.3
Index infant was low birth weight	11.4			13.7
Gestational age at delivery of index infant, weeks		39	34–42		39	34–44
Parity		2	1–12		1	1–17

* P < 0.05.

Table 2 presents the selected maternal and infant characteristics of women who agreed to participate versus those who refused among those whom we successfully contacted. Among successfully contacted women, 584 (51%) agreed to participate. Women who agreed to participate were more likely to be of higher educational attainment than nonparticipants.

Table 2.

Selected Maternal and Infant Characteristics by Participation Status, Massachusetts, 2001–2002

	Participated (n = 584)			Refused (n = 553)
	%	Median	Range	%	Median	Range
Fertility drug use reported on birth certificate	31.0			34.2
Assisted reproductive technologies reported on birth certificate	50.5			51.0
Plurality
Singletons	80.3			80.5
Twins	18.5			18.2
Triplets or higher	1.2			1.3
Maternal race
White	94.9			91.3
Black	1.4			1.5
Asian	2.4			4.9
Other	1.4			2.4
Maternal education*
Less than high school graduate	1.8			1.8
High school graduate	21.1			27.1
Some college	10.9			12.3
College graduate or more	66.3			58.8
Maternal age
<30 years	5.5			7.6
30–34 years	21.8			27.3
35–39 years	28.1			25.5
≥40 years	44.7			39.6
Mother married	95.6			97.7
Prenatal care paid by the government	2.6			2.5
Index infant was low birth weight	12.0			10.7
Gestational age at delivery of index infant, weeks		39	34–42		39	34–44
Parity		2	1–12		1	1–11

* P < 0.05.

Among the 584 women who agreed to participate, 399 (68.0%) were selected because of an indication of infertility treatment on the birth certificate. Among those whose child's birth certificate did not indicate treatment, 173 (30.0%) were selected because the women were over age 42 years, 3 (0.5%) because they delivered twins or triplets, and 9 (1.5%) because they were over age 42 years and delivered twins or triplets. Overall, 5 (0.8%) women refused to disclose their infertility treatment status and as such were excluded from further analyses.

Table 3 provides a detailed comparison of the infertility check box status on the birth certificate versus the mother's report of infertility treatment stratified by plurality. The birth certificates of singletons showed reasonably good agreement with maternal report (weighted kappa = 0.66; P < 0.01), while agreement among the birth certificates of twins, albeit based on a modest sample size, was poor (weighted kappa = 0.05; P = 0.11). As shown in Table 2, the birth certificates of singletons were fully concordant with maternal report for 245 (53%) women, partially concordant for 171 (37%) women (i.e., the certificate correctly indicated the use of infertility treatment, just not the correct type), and fully discordant for 44 (10%) women. The birth certificates of children resulting from multifetal pregnancies were fully concordant with maternal report among 26 (23%) women, partially concordant among 73 (63%) women, and fully discordant among 16 (14%) women. The most common error on the birth certificates of singletons and multiples was the underreporting of fertility drug use on the birth certificates of women whose children had been conceived with the help of both assisted reproductive technologies and fertility drugs. Among singletons, total discordance was related to both under- and overreporting of infertility treatment, and the predominant error seen on the birth certificates of multiples was underreporting of treatment.

Table 3.

Agreement Between Infertility Status on the Birth Certificate and Maternal Report of Infertility Treatment, Stratified by Plurality, Massachusetts, 2001–2002

Birth Certificate Status	Mother's Report
	Assisted Reproductive Technologies Only		Fertility Drugs Only		Both		Neither
	No.	%	No.	%	No.	%	No.	%
Singletons
Assisted reproductive technologies only	10	2.2	22	4.7	111	23.9	9	1.9
Fertility drugs only	6	1.3	45	9.7	24	5.2	11	2.4
Both	5	1.1	7	1.5	39	8.4	–a
Neither	–		7	1.5	13	2.8	151	32.5
Twins and triplets
Assisted reproductive technologies only	–		–		54	46.7	–
Fertility drugs only	–		5	4.4	11	9.6	–
Both	–		–		18	15.7	–
Neither	–		–		10	8.7	–

^a–, cell size < 5 (actual numbers suppressed according to the privacy policy of the Massachusetts Department of Public Health).

In looking at the overall utility of the infertility treatment check boxes, we compared the treatment status of women (i.e., categorized as treated or not) with maternal report of treatment (data not shown). Of the 395 birth certificates of children whose mother was included because one or both of the infertility treatment check boxes indicated that she had received infertility treatment in the pregnancy, 367 (93%) reported actually receiving treatment. Among the 399 women in our sample who reported infertility treatment in the index pregnancy, 367 (93%) of the birth certificates indicated treatment. This yielded a kappa for treated versus not treated of 0.76 (P < 0.01), indicating good overall agreement between the birth certificate and maternal report.

A key issue related to the feasibility of using birth certificates as a sampling frame for the identification of children conceived with the help of infertility treatment is the question of how many treated pregnancies the birth certificates miss. To examine this issue, we examined the data of women who delivered twins or triplets and/or were over age 42 years (i.e., a select group of birth certificates that were checked as well as the supplemental sample of birth certificates that were not checked). Of the women whom we interviewed whose child's birth certificate did not indicate use of infertility treatment, 17.4% were incorrect. Given that we selected a sample of birth certificates with a high likelihood of error to examine this issue, this error rate clearly vastly overestimates the true amount of underreporting of infertility treatment use among all birth certificates.

Table 4 presents selected maternal and infant characteristics by infertility treatment status discordance between the birth certificate and the mother's report. Full discordance was more commonly found among women delivering triplets or higher order multiples, as well as women over age 42 years. There were no differences in discordance with respect to maternal race or age or any other maternal and infant covariate examined. Table 5 examines the adjusted odds of infertility treatment status discordance between the birth certificate and maternal report. There was a more than 18.5-fold increase (95% CI: 3.8, 89.5) in the odds of discordance found among women delivering triplets, while we found a 2.7-fold increase (95% CI: 1.5, 4.9) in the odds of discordance among women over age 42 years.

Table 4.

Selected Maternal and Infant Characteristics by Infertility Treatment Status Discordance Between the Birth Certificate and Maternal Report, Massachusetts, 2001–2002

	Concordant or Partially Concordant (n = 519)			Fully Discordant (n = 60)
	%	Median	Range	%	Median	Range
Fertility treatment status on birth certificate*
Assisted reproductive technologies only	39.1			21.7
Fertility drugs only	17.5			20.0
Both	14.1			5.0
Neither	29.3			53.3
Plurality*
Singletons	80.9			73.3
Twins	18.5			20.0
Triplets or higher	0.6			6.7
Maternal race
White	94.6			96.7
Black	1.5			0.0
Asian	2.3			3.3
Other	1.4			0.0
Maternal education
Less than high school graduate	36.2			31.7
High school graduate	20.2			26.7
Some college	41.4			41.7
College graduate or more	2.1			0.0
Maternal age*
<30 years	5.4			6.7
30–34 years	22.4			18.3
35–39 years	28.3			21.7
≥40 years	43.9			53.3
Mother married	95.6			91.7
Prenatal care paid by the government	2.7			1.7
Index infant was low birth weight	11.2			20.0
Gestational age at delivery of index infant, weeks		39	24–42		39	33–42
Parity		2	1–12		1	1–4

* P < 0.01.

Table 5.

Unadjusted and Adjusted Odds of the Birth Certificate and Maternal Report Being Fully Discordant With Regard to Fertility Treatment Status, Massachusetts, 2001–2002 (n = 579)

	Odds Ratio	95% Confidence Interval	Adjusted Odds Ratio	95% Confidence Interval
Plurality
Singleton	1.0	Referent	1.0	Referent
Twin	1.2	0.61, 2.4	1.7	0.8, 3.5
Triplet	12.7**	2.8, 58.7	18.5**	3.8, 89.5
Maternal age >42 years	2.2*	1.3, 3.8	2.7**	1.5, 4.9

* P < 0.05; **P < 0.01.

After asking women various questions related to their infertility treatment status in the index pregnancy, we asked women who indicated that their index child was conceived with the help of infertility treatment a series of questions related to their willingness to participate in research related to the health and developmental follow-up of their children. Overall, 387 (97%) women indicated that they would be willing to participate in such a follow-up study. Five (1.3%) women said that they were unsure about participation, while 7 (1.7%) women said that they would not participate. Among the women who were unsure about participation, the most commonly cited reason was that they needed more information. Among those who did not want to participate, the reasons included not wanting to think about infertility treatment anymore, privacy concerns, and not being sure how the child would feel about it. Table 6 presents the willingness to partake in various research activities among the 387 women who were interested in participating in a developmental follow-up of their child. Women were most likely to be willing to complete questionnaires about themselves or their child and least likely to grant access to medical records.

Table 6.

Mothers’ Willingness to Participate in Various Activities Related to the Follow-up of Their Children Among Women Who Would Be Interested in Participating in a Study, Massachusetts, 2001–2002 (n = 387)

	Yes, %	No, %	Don't Know, %
Would complete a questionnaire about self	97.2	0.5	2.3
Would grant access to personal medical records	48.6	21.4	30.0
Would provide a urine sample	57.4	21.9	20.7
Would provide a blood sample	54.3	23.0	22.7
Would complete a questionnaire about child	86.1	2.8	11.1
Would grant access to child's medical records	40.2	30.6	29.3
Would allow in-home evaluation of child's development	65.0	18.1	16.8

DISCUSSION

To our knowledge, this is the first study to examine the validity of the infertility treatment check boxes on the birth certificate in relation to maternal report of treatment. Overall, we found the birth certificate to be an efficient means of identifying children conceived with the help of infertility treatment, but that it was not a valid source for information regarding type of treatment. Furthermore, we found pronounced underreporting of infertility treatment in the supplementary sample of women that we included, namely, women who delivered twins or higher order multiples and/or women over age 42 years whose child's birth certificate did not indicate that the mother had received infertility treatment. As such, any researchers attempting to use birth certificates as a sampling frame to identify children born to infertility-treated women should consider contacting all older women (particularly those over age 42 years), as well as all of those delivering twins or higher order multiples.

In 2003, Roohan et al. (10) performed a small study validating a sample of New York State birth certificates against medical records. For this project, they selected a random sample of birth certificates (n = 440) from maternity hospitals in 4 counties in New York (2 downstate counties and 2 upstate counties). They then obtained the clinical records pertaining to each of these pregnancies and compared the information reported on the birth certificate with that available in the medical record. They reported the sensitivity, specificity, positive predictive value, and negative predictive value of the in vitro fertilization check box to be 80%, 100%, 80%, and 100%, respectively. As previously discussed, infertility treatment is a very rare exposure among liveborn infants and, as such, the assessment of the validity of the infertility treatment check box requires a targeted sampling strategy such as the one that we used.

Zhang et al. (11) recently published results of a study in which the data from the 1997–2000 Massachusetts infertility check box were compared with the data from the National Assisted Reproductive Technologies Surveillance System (known as “NASS”). This system includes data reported by US infertility clinics on assisted reproductive technology cycles that are reported to the Centers for Disease Control and Prevention as mandated by the 1992 Fertility Clinic Success Rate and Certificate Act. These data comprise only assisted reproductive technology procedures and do not include information regarding cycles that involve only fertility drug use and/or intrauterine insemination.

These authors found the sensitivity of the birth certificate check boxes to be lowest among singletons (24.3%), with slightly higher sensitivity among twins (31.8%) and triplets (43.0%). In contrast, the specificity among singletons and twins was found to be quite high (99.8% and 95.6% respectively), while the specificity for triplets was much lower (66.2%). Overall, they found that women with birth certificate evidence of preexisting conditions, pregnancy complications, and preterm delivery were more likely to have infertility treatment correctly noted. This finding may simply be due to a higher quality or more complete maternal record being available for these women at the time of data abstraction. Although twins and triplets were also more likely to have infertility treatment correctly indicated, there were no other clear demographic differences in the odds of having infertility treatment correctly noted. Therefore, although there are certainly concerns about the low sensitivity of the assisted reproductive technology data reported on the Massachusetts birth certificate, it appears to be a relatively efficient method of identifying children who were conceived with the help of treatment.

In considering using the birth certificates to identify children for a study of this type, one must carefully consider potential biases. It is possible that the reporting of infertility treatment status may be differential with respect to birth or child health outcome. For instance, women who deliver healthy children who appear to be developing typically after infertility treatment may be less likely to be forthcoming about their treatment status. However, we feel that the implication of this bias is less an issue of the potential for differential misclassification with regard to exposure and more an issue of selection bias. We assert that, if a woman does not wish to admit her infertility treatment status, she is more likely to be unwilling to participate in a follow-up study of her child rather than to participate and misreport her treatment status. Therefore, if women of children who do well are less likely to participate in studies of this type, any estimate of the association between infertility treatment status and adverse outcomes would likely be an overestimate of the true association.

Our study has several limitations that should be considered. First, we had a somewhat low contact and participation rate. Specifically, we were able to contact only slightly less than half of the targeted mothers. Part of this low contact rate could certainly be due to the fact that 2–4 years had elapsed between the index delivery and our attempt to contact the woman and, as such, we were more likely to contact women who were less likely to have moved from the address listed on the birth certificate. This rate of contact could certainly be improved by attempting to contact mothers sooner after delivery.

Further, women of a minority race and of lower educational attainment were less likely to have participated in our study. Although it is clearly very important to understand the effects of infertility treatment among racial and ethnic minorities, infertility treatment remains a therapy that is typically accessed only by families with significant financial resources due to the lack of coverage by many insurance companies. Despite the fact that at least 15 states have mandated some form of coverage for infertility treatment by insurance companies, a recent report indicated that there is no evidence that such mandates have lessened the racial, ethnic, and educational disparities associated with who is able to access infertility treatment (12).

In addition to the concern about participation rates, there is always a question as to whether or not all women in our study were forthcoming with regard to their treatment status. We have no reason to believe that there are any concerns in this regard. As noted above, there is always the concern that women who do not want to disclose their infertility treatment status will be less likely to participate in studies of this type, which would certainly limit the generalizability of study findings. The fact, however, that the overwhelming majority of the infertility-treated women in our study indicated that they would agree to participate in at least minimally invasive follow-up of their child (i.e., questionnaires) is encouraging and suggests that follow-up studies of children conceived with the help of infertility treatment as identified on the birth certificate (with maternal confirmation of treatment) are feasible. Investigators undertaking this type of research must simply pay close attention to the potential for selection bias and interpret their results accordingly.

Finally, the wording of the infertility treatment check boxes on the Massachusetts birth certificate differs somewhat from the recommended check boxes on the 2003 US Standard Certificate of Livebirth. The Standard Certificate is worded as follows:

Pregnancy resulted from infertility treatment—if yes, check all that apply:

• Fertility-enhancing drugs, artificial insemination, or intrauterine insemination

• Assisted reproductive technology (e.g., IVF [in vitro fertilization], GIFT [gamete intrafallopian transfer])

Notably, there is no evidence that the Centers for Disease Control and Prevention examined the validity of the infertility check box wording before issuing the 2003 version of the Standard Certificate (13). Given, however, that any errors in reporting are more likely a function of the information available to hospital personnel in the prenatal care record at the time the birth certificate worksheet is completed rather than the wording of the infertility treatment check boxes themselves, we feel that our findings are likely generalizable to the data available on all US birth certificates on which such check boxes are included.

As an increasing number of children born in the United States are conceived with the help of infertility treatment, it will become even more important for investigators to conduct large-scale population-based follow-up studies of health and development in this population. The infertility treatment check boxes that are currently being phased in by states on birth certificates across the nation offer promise for identifying these cohorts of children.

Glossary

Abbreviations

CI

confidence interval

OR

odds ratio