National birth defects prevention study

The NBDPS is an ongoing, population-based, case-control study comprising data collected by 10 birth defects surveillance systems throughout the United States (Arkansas, California, Georgia/Centers for Disease Control and Prevention [CDC], Iowa, Massachusetts, New Jersey [through 2002], New York, North Carolina [beginning 2003), Texas, and Utah [beginning 2003]). Cases in the study had 1 or more of more than 30 eligible birth defects and were liveborn, stillborn, or electively terminated. Control infants were not matched with case infants and were liveborn infants without birth defects who were randomly selected either from birth certificates or hospital birth records.

The NBDPS enrolls all eligible cases and approximately 300 controls per study center, per year. This sample size goal was based on the assumption of etiologic heterogeneity among birth defects and that it would always be more informative to analyze specific birth defect phenotypes compared with controls, rather than all birth defects in the aggregate.

Mothers were interviewed by telephone in either English or Spanish using a computer-based questionnaire 6 weeks to 24 months after the estimated date of delivery. Interviewers obtained information on maternal demographic characteristics and exposures (eg, nutritional, behavioral, occupational) and medication use both before and during pregnancy. The participation rate for mothers of control infants was 67% and for mothers of CHD cases, 69%. The NBDPS was approved by the institutional review boards of the CDC and the participating study centers.

Clinical review and classification of congenital heart defects

All CHD cases were confirmed by echocardiography, cardiac catheterization, surgery, or autopsy. The medical records of case infants were reviewed, and each case was classified by a team of pediatric cardiologists and clinical geneticists on 2 distinct axes. The first axis of classification focused on the heart itself. “Simple” cardiac defects were anatomically discrete or a well-recognized single entity (eg, hypoplastic left heart syndrome or tetralogy of Fallot). “Associations” were common, uncomplicated combinations of cardiac defects. CHDs that included 3 or more distinct defects were considered “complex.” The second axis of classification considered the case infant as a whole. Cases with no major extracardiac defects were considered “isolated”; those with major extracardiac defects were considered “multiple.” The systematic review of all cases by clinical geneticists resulted in the exclusion of those with recognized or strongly suspected single-gene conditions or chromosome abnormalities from the NBDPS.

Inclusion criteria

Case and control infants delivered on or after Oct. 1, 1997, who had an estimated date of delivery on or before Dec. 31, 2004, were eligible for this study. Mothers with self-reported PGDM (type 1 or type 2) or missing prepregnancy BMI were excluded. We included simple CHDs and CHD associations (ie, CHD combinations: coarctation of the aorta with ventricular septal defect, ventricular septal defect with atrial septal defect, and pulmonary valve stenosis with atrial septal defect) if there were at least 50 infants with no extracardiac defects. Also included were case infants with complex CHDs of heterotaxia (n = 169) or single ventricle (n = 174).

Because of the high prevalence of muscular ventricular septal defects, these lesions were captured by the study centers in California, Georgia (CDC), Iowa, Massachusetts, New York, and Texas only through Oct. 1, 1998, and by the Arkansas and New Jersey study centers only through Jan. 1, 1999. Because North Carolina and Utah joined the study in 2003, these centers did not provide any case infants with muscular ventricular septal defects. In addition, California began ascertaining cases with pulmonary valve stenosis according to NBDPS criteria for births on or after Jan. 1, 2002; therefore, any case infants with pulmonary valve stenosis from that study center prior to this date were excluded from the analyses. Control infants were similarly restricted by ascertainment dates and study center.

Exposure and covariate definitions

All variable information was self-reported during the maternal telephone interviews. Self-reported prepregnancy height and weight were converted to metric units, and maternal BMI was calculated as weight in kilograms divided by height in meters squared. Five weight status groups were then formed using BMI: underweight (<18.5 kg/m ² ), normal weight (18.5-24.9 kg/m ² ), overweight (25.0-29.9 kg/m ² ), moderately obese (30.0-34.9 kg/m ² ), and severely obese (≥35.0 kg/m ² ). For some analyses, the categories of overweight and moderate and severe obesity were combined.

Eight covariates were considered in the study based on the literature and exploratory data analyses of the associations between the covariates and BMI and CHDs: maternal age (<20, 20-24, 25-29, 30-34, and ≥35 years old), maternal race or ethnicity (non-Hispanic white, non-Hispanic black or African American, Hispanic, and other race or ethnicity), maternal education (less than high school, completion of high school, and more than high school), parity (first live birth and >1 previous live birth), hypertension during the index pregnancy, and study center. Maternal smoking and folic acid supplement intake, 2 potentially time-varying covariates, were categorized on the basis of use during the month before pregnancy or the first month of pregnancy. GDM was explored as a potential effect measure modifier but not as a confounder.

Statistical analysis

Frequency distributions and ORs were calculated to determine the associations between BMI and CHDs and between BMI and covariates of interest among control infants. We conducted conditional logistic regression using the 8 covariates described previously, grouping on study center. For the analyses of main effects, we restricted our case groups to simple and associated (combinations of) CHDs with no extracardiac defects. The presence of effect measure modification on both the additive and multiplicative scales was assessed separately for GDM, maternal race or ethnicity, and periconceptional folic acid supplement intake.

To determine whether the interactions led to greater than multiplicative effects, we calculated a P value for the likelihood ratio test comparing saturated multiple logistic regression models (main effects plus interaction terms) with reduced models (main effects only). Because raising the type 1 error rate for this test recently has been shown to be inappropriate in most settings, we used a P value cut point of .05 to determine whether multiplicative interactions were present.

Then to assess whether the interaction was a departure from additivity of effects, we calculated the relative excess risk caused by interaction (RERI) based on a model with disjoint exposure categories (eg, normal BMI and no GDM, normal BMI and GDM, overweight and no GDM, overweight and GDM, obese and no GDM, and obese and GDM). Statistically significant RERI estimates greater than 0 suggest greater than additive effects or evidence of a synergistic interaction. For these effect modification analyses, we collapsed the categories of moderate and severe obesity into a single group and did not limit our analyses to simple or associated CHDs without extracardiac defects but rather included multiple and complex CHDs to increase the available sample size.

We conducted 2 additional analyses to assess the sensitivity of our results to the effects of 2 strong but relatively uncommon risk factors for CHDs: multiple gestations and a first-degree family history of CHDs. All analyses were conducted in SAS version 9.1 (SAS Institute, Cary, NC).