Alcohol exposure before and during pregnancy is associated with reduced fetal growth: the Safe Passage Study

Study design

This study was embedded in the Safe Passage Study, a prospective cohort study, conducted between 2007 and 2016 [23]. The total cohort included 12,000 pregnant women, recruited from predefined communities at high risk for prenatal alcohol use (USA and South Africa). The detailed study protocol is described elsewhere [23]. Previous studies in the South African arm of the cohort showed a reduction in birth weight Z-scores in neonates prenatally exposed to alcohol [24, 25]. However, PAE as variable was categorized, causing difficulties in interpretation of PAE-effects. The current study, restricted to the South African arm of the study, investigated PAE as continuous variable in a longitudinal manner, focusing on timing and quantity, clarifying relationships more efficiently.

In a randomly selected subset (n = 1928) of the Safe Passage cohort recruited in South Africa, additional measures were collected. Women in this so-called embedded protocol were enrolled before 24 weeks of gestation [23]. After enrollment, three remaining visits occurred at 20–24, 28–32, and after 34 weeks of gestation. All women gave informed consent.

Study population

Within the sub-cohort (n = 1928), we excluded women with missing ultrasound or growth data (n = 91) and second or third participations in the study (n = 81). Pregnancies with congenital anomalies or different growth patterns due to other reasons than alcohol exposure were additionally excluded: twin pregnancies (n = 12, 24 fetuses), miscarriages (n = 3), congenital anomalies (n = 4), stillbirths (n = 25), and terminations of pregnancy (n = 3) (Additional file 1: Figure S1). The final study population included 1698 women with singleton pregnancies.

Data collection

Alcohol and tobacco exposure

Information on maternal alcohol consumption was collected through interviews performed by trained research staff, using an adapted version of the Timeline Followback method (TLFB) [26]. This method was contemplated as one of the most strict methods in self-reported exposure assessment, using calendar worksheets and visual prompts to support participants in recall about their alcohol consumption [27]. Maternal alcohol consumption periods were defined as follows: within the periconception period, defined as 15 days prior to and 15 days after the last menstrual period (LMP), first trimester was defined as gestational days 0 until 97 (equal to 13 + 6 weeks of gestation), second trimester as gestational days 98 until 195 (equal to 14 + 0 until 27 + 6 weeks of gestation), and third trimester as ranging from gestational day 196 until delivery (equal to 28 + 0 weeks of gestation until delivery). It should be noted there is a slight overlap between the periconception period and first trimester. Maternal alcohol consumption during the periconception period was collected at the enrollment visit. For each trimester, collected during the following prenatal visits, alcohol consumption data were collected based on the last reported drinking day and 30 days prior. For each drinking day, detailed information on the type of drink, the number and size of drinks, drinks including ice, and the duration of drinking was collected. The total amount of alcohol in grams was converted into standard drinks during the periconception period and per trimester separately [28]. Furthermore, the average amount of alcohol-use in grams per day during pregnancy as a whole and for the periconception period and each trimester were calculated. By definition, one standard drink contains 14 g of alcohol, and binge drinking is defined as drinking ≥ 4 alcoholic drinks per occasion [29, 30]. Maternal binge drinking was calculated as the total amount of binge-moments during pregnancy. The prevalence of total PAE in this study was 62%, and 27% of women reported binge drinking. Fetuses not exposed to maternal alcohol consumption were referred to as controls.

Since alcohol-using women were more likely to smoke cigarettes, which is known to reduce fetal growth, we also investigated maternal tobacco use during pregnancy [31, 32]. This was investigated using questionnaires with graduated frequency response options (e.g., none, monthly or less, 2–4 days/month, 2–3 days/week, 4–6 days/week, and 7 days/week) and the number of cigarettes smoked per day, covering a 30-day reference period prior to the last smoking day. The average amount of cigarettes per day was calculated.

Fetal growth, pregnancy, and pregnancy outcomes

Trained sonographers performed two-dimensional ultrasound examinations trans abdominally using a Voluson E8 ultrasound machine (GE Healthcare) with a RAB 4–8 3D transducer, applying internationally standardized protocols. Pregnancy dating occurred at the first prenatal visit. Between 6 + 0 and 13 + 6 weeks of gestation, the fetal crown-rump-length (CRL) was used. From 14 + 0 weeks onwards, fetal head circumference (HC), fetal biparietal diameter (BPD), fetal abdominal circumference (AC), and fetal femur length (FL) were used for pregnancy dating, as well as for fetal growth measurements [33]. Estimated fetal weight (EFW) was calculated based on HC, BPD, AC, and FL. All fetal growth outcome measures (including HC, BPD, AC, FL, EFW, and birth weight) were transformed into Z-scores using the INTERGROWTH-21^st project standard formulas, correcting for exact gestational age (GA) in weeks [34, 35]. Fetal growth Z-scores were treated as continuous longitudinal variables.

Information on maternal hypertensive disorders (systolic blood pressure > 140 mmHg or diastolic blood pressure > 90 mmHg), birth weight, and preterm birth (< 37 weeks of gestation) were collected from medical records.

Covariates

Potential covariates were selected a priori based on previous studies [16, 36, 37]. Self-reported maternal characteristics (at enrollment) included age, parity, obstetric and medical history, years of education, and monthly income in South African Rand [23]. Also, maternal anxiety was collected at enrollment, using the state-anxiety subscale from the Spielberger state trait anxiety scale [38, 39]. Maternal depression, collected at the enrollment visit, was scored using the Edinburgh postnatal depression scale (EPDS) [40]. Maternal mid-upper arm circumference (MUAC) and self-reported other substance use (e.g., marijuana and methamphetamine) were collected each visit, of which procedures were described previously [23, 41].

Statistical analysis

To evaluate non-response and investigate representativeness of the study sample for the entire study population, baseline characteristics between included and excluded women were compared (Additional file 2: Table S1). Baseline characteristics of the alcohol exposed group were compared to controls (Table 1). Continuous variables were analyzed using Student’s t-test (normal distributed) or Kruskal–Wallis test (non-normal distributed), categorical variables with χ² tests.

Very few variables were missing, except monthly income (26% missing), for which multiple imputation by chained equations was used (25 datasets, Additional file 2: Table S2) [42].

To investigate the association between PAE and fetal growth or birth weight Z-scores, linear mixed models were applied. We first investigated the association of PAE as accumulative parameter over the course of pregnancy and fetal growth Z-scores, using a model in which average alcohol consumption (grams/day) during pregnancy and GA were added as predictors (model 1, accumulation model). Second, we built a model in which PAE per trimester was separated, to investigate whether exposure in any specific trimester was crucial in fetal growth. In this model, four different exposure periods (periconception period, trimester 1 (T1), trimester 2 (T2), and trimester 3 (T3)) and GA were added as predictors (model 2, trimester-specific model). Since third trimester alcohol exposure cannot influence second trimester growth, we modeled that third trimester PAE could only influence third trimester growth measurements. First and second trimester PAE can influence both second and third trimester growth measurements. Similar methods were applied to investigate associations between PAE and birth weight Z-scores. Our secondary analysis examined the association of binge drinking with fetal growth, applied as bingers compared to non-drinkers and to drinkers, but non-bingers (model 3, binge drinking model). Predictors in this model were PAE as accumulative parameter and total binge drinking moments during pregnancy. To explore the association of smoking with fetal growth, a fourth model was built adding smoking as single predictor, as well as simultaneous tobacco and alcohol exposure (model 4, tobacco co-use model). In all analyses, growth trajectories of alcohol exposed fetuses were compared to those of controls using linear mixed models, by calculating estimated Z-score differences per growth measurement. Lastly, the associations between PAE and hypertensive disorders or preterm birth, both dichotomous outcomes, were investigated using logistic regression analysis.

All models, except for model 3, were adjusted for fetal sex, maternal age, MUAC, parity, years of education, monthly income, other drug use (marijuana and/or methamphetamine), anxiety, and smoking (in models 1 and 2). Model 3 was only adjusted for fetal sex. For interpretability purposes, we estimated Z-score differences (ED) with 95% confidence intervals (CIs) using preset situations, all calculated for 20, 30, and 36 weeks of gestation, except for the binge drinking model, in which we calculated beta’s (β) and 95% CIs. Model 3 was built to investigate if binge drinking has an additional association with fetal growth. For interpretability purposes, we calculated uncorrected βs, indicating the additional increase or decrease in fetal growth per unit increase in alcohol consumption.

In these preset situations, variables were included as follows: alcohol exposure = 1 standard drink/day, mean tobacco exposure = 10 cigarettes/day (only in model 4), fetal sex = “male”, maternal age = 25 years, MUAC = 277 mm, parity = 1, education = 10 years, monthly income = 910 South African Rand, other drug use = “yes”, anxiety score = 31, average cigarette exposure/day (to adjust for smoking in models 1 and 2) = 3.17.

In all analyses, the SPSS statistical software version 25.0 and R statistical software version 4.1.0 for Windows were used. A p-value ≤ 0.05 was considered statistically significant.