We conducted a cross-sectional study of data from the National Survey of Family Growth (NSFG). The NSFG is an annual survey that began in 1973, which collects a combination of self-administered surveys and in-person interviews. This nationally representative sample includes civilian, non-institutionalized reproductive-aged U.S. women between the ages of 16 and 45 (Centers for Disease Control and Prevention, 2018). In 2006, the NSFG began to include questions measuring sexual identity and same-sex sexual behavior. The NSFG has a response rate of 69% for recent data releases, and includes detailed reproductive histories for all reported pregnancies and their outcomes. Survey data was collected using Computer-Assisted Personal Interviews (CAPI) methods, administered by interviewers but completed on their own. All information provided in the surveys was self-reported . Institutional review board approval was not obtained for this secondary analysis.
We restricted our analytic sample to pregnancies reported by NSFG participants between 2006 and 2017. Our total eligible sample was 53,751 pregnancies. For analyses restricted to live births, our total eligible sample was 36,374. We limited our analysis to singleton pregnancies of participants who were 1) currently not pregnant, and 2) who identified their race/ethnicity as NH Black, NH White, and Hispanic as distinct groups, as there were insufficient sample sizes for participants in the “other” category (n = 3,072, 5%, and describe which groups were represented in “other”, i.e. Asian, Native American, etc.).
The unit of analysis was pregnancy, and we applied the NSFG’s complex sampling frame, clustering by mother to account for multiple births reported to a single woman. All pregnancy history and birth outcomes were self-reported and collected retrospectively at the same time as demographic information by via CAPI.
Sexual Orientation was our primary predictor variable and was measured categorically, combining both sexual behavior and orientation measures as gender identity data were not available. Respondents were asked, "Do you think of yourself as heterosexual or straight; homosexual, gay or lesbian; or bisexual?" Respondents were also asked, "Have you ever had any sexual experience of any kind with another female?" From these questions, we created a four-category variable including heterosexual-identified with only male sexual partners (heterosexual-women who have sex with men (WSM)], heterosexual-identified with female partners (heterosexual-WSW), bisexual, and lesbian.
Race/ethnicity was constructed as a three-category variable, derived from two survey items that first asked respondents, “Are you Hispanic or Latina, or of Spanish origin?” followed by a survey item that asked “What is your race?” Respondents were recoded into mutually exclusive categories, NH White, NH Black, and Hispanic.
Maternal Age was derived from the participant's age at the time of the interview and the year the pregnancy ended.
Education was measured categorically, including less than a high school diploma, completed high school, some college, or a college degree.
Socioeconomic status was operationalized as Percent of Federal Poverty Line (FPL), which was measured as a categorical variable that captured whether household income adjusted for household size was < 100% of the FPL, [greater than or equal to] 100% and < 200% FPL, or [greater than or equal to] 200% FPL.
Nativity was measured as a dichotomous variable, with respondents who were born in the U.S. coded as 0 and those born outside the U.S. coded as 1.
Previous Preterm Birth was a variable created using the roster and birth information to capture whether a participant had previously reported a preterm birth prior to the index pregnancy.
Preterm birth was derived from a survey item that asked respondents “A preterm delivery is one that occurs at 36 weeks or earlier in pregnancy. As far as you know, did you have a preterm delivery?”.
Low birthweight was a NSFG recode derived from the item, “Did (she/he) weigh 5.5 pounds or more, or less than 5.5 pounds.” If the respondent stated that their infant was less than 5.5 pounds at birth, they were coded as being low birthweight. If the respondent stated that their infant was more than 5.5 pounds at birth, they were coded as not being low birthweight.
Birthweight was measured using survey items that asked respondents to self-report an infant’s birthweight in pounds and ounces. We then recoded this to represent birthweight in ounces alone.
Miscarriage was captured using the item, “In which of the ways shown on Card 13 did the pregnancy end?” Answer choices included miscarriage, stillbirth, abortion, ectopic or tubal pregnancy, live birth by c-section, and live birth by vaginal delivery. If the respondent reported that the pregnancy ended in a miscarriage, that pregnancy was dichotomously coded as a miscarriage.
Stillbirth was measured using the same question and responses as miscarriage. If the respondent reported that the pregnancy ended in a stillbirth or that the pregnancy ended after 20 weeks of gestation, that pregnancy was dichotomously coded as a stillbirth.
We controlled for a variety of covariates associated with our exposure and outcome measures. These included socioeconomic status, education, nativity, and age. To examine how multiple stigmatized identities may impact findings, we also stratified by race.
We first conducted bivariate analyses of preterm birth by sexual orientation. Next, we employed logistic regression models for the dichotomous outcome measures including preterm birth, low birthweight, miscarriage, and stillbirth. Linear regression models were used for continuous birthweight. In unadjusted analyses, we examined the association between sexual orientation [heterosexual- WSM; heterosexual-WSW), bisexual, or lesbian] and each birth outcome [preterm birth, low birthweight, miscarriage, and stillbirth]. We then conducted adjusted analyses controlling for maternal age, education, FPL, nativity, and previous preterm birth. We also include results stratified by race and ethnicity. All models were estimated using STATA Standard Edition version 14.2 and adjusted for NSFG population weights. Our unit of analysis was the pregnancy; therefore, we clustered on participants to account for the non-independence of pregnancies reported by each individual.