The study is part of a larger prospective cohort study on maternal stress, birth outcomes and infant health. The original cohort consisted of women who received prenatal care at health centers in Philadelphia, PA, and who were recruited consecutively from February 2000 to November 2002. These health centers consisted of Federally Qualified Health Center Look Alikes (FQHC-LAs) and FQHCs and have been described previously.35
The ability to speak English or Spanish and having an intrauterine pregnancy were the two major enrollment criteria. This research was approved by the Thomas Jefferson University, the University of Pennsylvania, and Drexel University institutional review boards.
For this study, we use data from three surveys. The first survey was administered to women at their first prenatal care visit (mean gestational age ± standard deviation = 13.8 ± 6.3 weeks), and the second and the third were respectively conducted at 3 ± 1 months and 11 ± 1 months postpartum in their homes. Among the 2,374 eligible women, 42 declined participation, 190 had an unknown birth outcome, and 158 had a miscarriage, abortion, ectopic pregnancy or stillbirth. The remaining 1,984 women were known to have a live birth and completed the prenatal interview, and both postpartum surveys were completed by 1,482 (75%) women. Of the outstanding 502 participants, 126 moved, 129 refused to participate, 28 were excluded because they were unable to participate or no longer lived with their child, 10 had children who died, and 209 did not complete the postpartum interviews. Six additional women who completed the postpartum surveys were dropped from the study due to missing information for a final sample of 1,476 women. When compared to the final sample of 1,476, the 508 women, who lacked or had incomplete postpartum information, did not differ with respect to education, marital status, and parity; however, they were slightly older, more likely to be Latina and less likely to be white, and more likely to have lower annual personal income (data not shown). Trained female interviewers surveyed the women using standardized questionnaires in either English or Spanish. Maternal sociodemographic characteristics were assessed in the initial prenatal interview. Risky health behaviors were assessed in the first two surveys. Childhood experiences, prior to the age of 16, were assessed by 32 items in the third survey at 11 ± 1 months postpartum.
The dependent variables consisted of four risky health behaviors. Smoking in pregnancy, assessed in the prenatal survey, was defined as an affirmative response to the question, “After you found out that you were pregnant this time, have you smoked at all?” Alcohol, marijuana and other illicit drug use were assessed during the second survey. Alcohol use during pregnancy was defined any “beer, wine, 40's, coolers, liquor, or other alcoholic beverages when you were pregnant with [CHILD]?” Marijuana use during pregnancy was defined as any “marijuana, pot, joints or blunts when you were pregnant with [CHILD]?” Other illicit drug use during pregnancy was defined as any “other drugs, such as cocaine, heroine, crack, crank, LSD, uppers, downers, or similar drugs you can get on the street when you were pregnant with [CHILD]?”
Independent variables consisted of sociodemographic variables and adverse childhood experiences (ACEs). The following sociodemographic characteristics were obtained prenatally: maternal age, education, race/ethnicity, marital status, and annual personal income.
ACEs, based on events occurring before age 16, consisted of 7 variables: physical abuse, sexual abuse, verbal hostility, domestic violence, having witnessed a shooting, having a guardian in trouble with the law or in jail, and having a guardian with substance use. All questions were preceded by the clause, “Before you were 16 years old…”
Measures that overlapped with the ACE Study included assessments of verbal hostility, physical and sexual abuse, household violence, household substance abuse, and a household member being incarcerated. Measures differed from the ACE Study in that the 7 ACEs were assessed before age 16, whereas the 10 ACEs in the ACE Study were assessed before age 18. The ACE Study, however, assessed living with household members that were substance abusers or incarcerated rather than specifying substance abuse in or incarceration of a parent or guardian. ACE Study variables not assessed were emotional and physical neglect (measured in less than 50% of the ACE Study sample), household mental illness, and parental separation or divorce. One measured variable not found in the ACE Study was witnessing a shooting.
The seven ACE variables used in our study were defined as follows: (1) Physical abuse was defined by “rarely, sometimes, or often” as opposed to “never” to a combination of survey items, separated below by semicolons, and summarized by, “How often did you experience slapping you in the face; punching, pushing, kicking, or beating you with fists; hitting you with an object such as a belt, spoon; or burning you with an object such as a cigarette or iron by the person or people who raised you when you did something wrong?” Spanking was a separate question not included in our analysis. (2) Sexual abuse was based on a confirmatory answer to the question, “Were you ever sexually abused? (3) Verbal hostility was defined by “often” as opposed to “sometimes, rarely or never” to the question, “How often did you experience yelling by the person or people who raised you when you did something wrong?” We believe that considering “sometimes or rarely” yelling as “verbally hostile” would overestimate verbal hostility.
(4) Domestic violence was based on an affirmative answer to, “Was anyone in your house being hit or beaten up?” In the ACE Study, “mother treated violently” was measured rather than any “domestic violence.” (5) Witnessing a shooting was based on an affirmative response to, “Did you ever see someone get shot?” (6) Having a guardian in trouble with the law or in jail was based on an affirmative answer to two items summarized by, “Did your main parent/guardian ever have trouble with the law or ever spend any time in jail?” (7) Having a guardian with substance use was based on an affirmative answer to the question, “Did your main parent/guardian ever have a drug or alcohol problem?
Frequencies were calculated for categorical variables while means, medians and standard deviations were calculated for continuous variables. For each participant, we totaled the number of ACEs to create a summary score, which ranged from 0 to 7. Due to the low prevalence of reported illicit drug use other than marijuana during pregnancy (2%), marijuana use was combined with other drug use to record any illicit drug use in pregnancy. Bivariate analyses were performed comparing the prevalence of each of the risky health behaviors among those with and without each of the ACEs, using the Chi-square test for categorical variables and the Student's t test for continuous variables to test for statistical significance.
In addition, bivariate analyses were performed to determine if there was a dose response for each risky health behavior based on the total number of ACEs. For increasing ACEs, there was a plateau effect for the risky behaviors after ACEs reached 3 or more; therefore, we combined those with 3 or more ACES into a category of “3 or more.” Collapsing the higher categories also increased the number of subjects in cells of 2 × 2 comparison tables and increased the strength of comparisons between groups.
For each dependent variable – smoking, alcohol and any illicit drug use during pregnancy -- further analyses were performed to adjust for confounding variables. For each of the dependent variables, we conducted a logistic regression analysis to adjust for potential confounding variables and to derive maximum likelihood estimates of combined relative odds with 95% confidence intervals (95% CIs). Risk factors and confounders for inclusion in our final models were identified a priori
based on our review of the literature, which is summarized above. Each of the models contained the following independent variables: age, race/ethnicity, marital status, education, annual personal income, and total ACEs. The Hosmer-Lemeshow goodness-of fit χ2 statistic was calculated for each final model to assess model fit.36