Maternal Alcohol Use during Pregnancy, Birth Weight and Early Behavioral Outcomes

Abstract

Drinking during pregnancy represents a major public health problem and generates much public attention in many countries (O'Leary et al., 2007). In the USA, the Centers for Disease Control and Prevention (CDC) showed that in 2005 approximately one in every eight pregnant women reported that they had consumed alcohol during their pregnancy. Furthermore, this high rate of alcohol use during pregnancy has remained relatively unchanged in the USA since the 1990s (Centers for Disease Control and Prevention, 2009). Accordingly, approximately half a million infants are exposed prenatally to alcohol each year. Understanding the consequences of prenatal alcohol exposure on children is not only of academic interest, but is also important for those determining policy.

Prenatal alcohol exposure has been associated with various behavioral problems in children, including conduct problems (D’Onofrio et al., 2007), attention problems (Streissguth et al., 1984; Knopik et al., 2005) and criminal and delinquent behaviors (Streissguth et al., 1999; Fast and Conry, 2004). While these studies provide the best evidence, more methodologically rigorous studies are definitely needed in order to develop a comprehensive understanding of the impact of prenatal alcohol exposure on children's behavioral development, particularly the developmental consequences associated with low-to-moderate levels of alcohol exposure (Jacobson and Jacobson, 1999; Huizink and Mulder, 2005). Results from animal experiments show that light-to-moderate alcohol consumption during pregnancy may cause mental health and behavioral problems in offspring (Schneider et al., 1997; Kraemer et al., 2007; Cuzon et al., 2008). Additionally, experimental evidence further suggests that prenatal alcohol use has a greater influence on the behavioral outcomes of offspring than it does on infant birth weight. In two randomized experiments on rhesus monkeys, Schneider et al. (1997, 2001) found that moderate alcohol consumption throughout pregnancy was associated with irritability, attention problems and poor neurological functioning in offspring, in spite of the fact that the baby monkeys born during the study showed no significant difference in birth weight between the control and the experiment groups (Schneider et al., 1997). Animal experiments thus suggest a ‘differential susceptibility hypothesis,’ in which infant behavioral outcomes are more vulnerable to prenatal alcohol exposure than physical growth is.

Extant findings from observational studies of the consequences of low-to-moderate levels of alcohol consumption during pregnancy, however, remain somewhat contradictory. Several studies found that low-to-moderate levels of alcohol exposure were positively associated with children's problem behaviors (Streissguth, 1980, 1999; Olson et al., 1997; Sood et al., 2001). In contrast, two recent analyses showed that low-to-moderate drinking was not independently associated with an increased risk of either attention deficit hyperactivity disorder or behavioral problems in children (Knopik et al., 2006; Kelly et al., 2009). The mixed results of these studies may be due to the problem of identifying the causal effect of prenatal alcohol exposure (Linnet et al., 2003; Testa et al., 2003; Huizink and Mulder, 2005). In particular, many prior evaluations did not adequately control for confounders, such as family processes and maternal characteristics, which may correlate with both child behavioral outcomes and maternal alcohol use. The causal inference problem is further complicated by study design and the type of data collected. Many studies have relied on small, selective, clinical samples where measures of family processes and parental characteristics were often unavailable. Consequently, many prior studies involve a risk of confounding resulting from simple comparisons between children born to drinking mothers and children born to non-drinking mothers without controlling for unobserved differences between the subjects.

Quasi-experimental designs are useful alternatives to account for potential confounders, including factors that are not measured (Duncan et al., 2004; D'Onofrio et al., 2007). For studies where randomized trials are not feasible, quasi-experimental designs rely on natural experiments that pull apart processes that are typically confounded and provide better evidence for causal inference. One class of quasi-experimental designs draws inferences from siblings or children-of-twins comparisons (often referred to as a fixed-effects model in the social science literature) and is particularly useful for studies in human development. By contrasting children exposed to prenatal alcohol with their siblings or cousins who were exposed to less, this approach controls for the mothers' characteristics, including genetic liabilities that are shared by mothers and offspring. However, only a few recent studies have adopted such a design in examining the behavioral consequences of prenatal alcohol consumption (for example, Knopik et al., 2006; D'Onofrio et al., 2007; Knopik et al., 2009). Results from these studies suggest that while many of the prior associations may be spurious, maternal alcohol use during pregnancy may remain a risk factor for certain behavioral problems. More importantly, these studies demonstrate the potential of the sibling approach in studying the impact of low-to-moderate alcohol use while pregnant. Furthermore, the differential susceptibility hypothesis suggested by the animal model has not been adequately examined with observational data in humans. According to this hypothesis, it would be expected that low-to-moderate alcohol use would have a stronger negative effect on child behavioral outcomes than on infant birth weight. An empirical test of this hypothesis might also inform theories about the neurological development and physical growth of the fetus.

The present study was designed to address the methodological limitations and research gaps in the literature. Specifically, this study investigated the association between maternal drinking during pregnancy and infant behavioral outcomes as related to the level of alcohol use. The study employed a sibling fixed-effects model to account for unobserved heterogeneity among the mothers of the children in the sample. In addition, the study tested the differential susceptibility hypothesis with observational data. Finally, additional analyses were performed to check the fixed-effects identification assumption and to investigate whether the associations between prenatal drinking and infant behavioral outcomes, if any, were moderated by social status. Given that the majority of women who drink while pregnant are considered light or moderate drinkers (Centers for Disease Control and Prevention, 2009), the findings from this study have significant implications for public health campaigns and prenatal problem prevention efforts that promote healthy pregnancies and child well-being.

The focus on infant behavioral outcomes is further driven by two reasons. First, studying infant outcomes provides additional opportunities to identify the causal effects of prenatal alcohol exposure. As children grow up, they are increasingly subjected to the influence of environmental forces (Shonkoff and Philips, 2000). Studies that focus on child outcomes that investigate children at older ages may not be able to isolate the effects of prenatal alcohol use from the child's changing environment. By examining behavioral outcomes during the very first periods of life, the present study minimizes the influence of potential confounders from postpartum environmental factors. Second, recent interdisciplinary evidence has shown that early behavioral outcomes, which have often been referred to as non-cognitive skills in economic and sociological literature (e.g. Farkas, 2003; Cunha et al., 2010), are strong predictors of later psychological well-being (Caspi et al., 2003), problem behaviors (Caspi et al., 1995) and labor market success (Caspi et al., 1998). At the same time, research suggests that the investments made in interventions later in life appear to be less efficient than earlier investments (Carneiro and Heckman, 2003; Coneus et al., 2012). In order to safeguard children's development, policies should emphasize early intervention, and, if possible, beginning during pregnancy (Doyle et al., 2009). Inferences about the relationship between alcohol use while pregnant and the behavioral development of offspring in early life may thus shed light on the origins of social inequality and future policy solutions for reducing social inequality.

METHODS

Sample

The current study investigated children born to women who participated in the National Longitudinal Survey of Youth (NLSY). The NLSY is a nationally representative sample of 12,686 participants, who were between 14 and 21-year-old at the start of the survey in 1979. Since 1986, the NLSY has biennially interviewed and gathered information on the children of the women in the original NLSY sample. However, the NLSY stopped collecting information on infant behavioral outcomes for children born after 2000. The following statistical analyses were thus restricted to children born between 1986 and 2000. The NLSY data include extensive information on maternal behaviors during pregnancy for every birth since 1986. Moreover, further data on the mothers can be obtained from the NLSY master files and linked to the corresponding child records on a year-by-year basis.

In the NLSY data, there were ∼6700 children born to the participating women between 1986 and 2000. The final analytical sample included only children aged 23 months to 4 years who had at least one sibling born between 1986 and 2000, thereby eliminating very young infants, whose behaviors are more difficult to assess (Rothbart, 1981; Shonkoff and Philips, 2000). This left 2132 children. Cases with missing values for infant behavioral outcomes and prenatal alcohol use (n = 371) and time-varying covariates were excluded (n = 143), producing a sample of 1618 children. Most of the sibling groups in the sample included only two children; however, sibling groups that included three or four children were not uncommon. The maximum number of children in a sibling group was five.

Outcome measures

The measures of the infant behavioral outcomes were derived from the NLSY child temperament instruments, which were a modified version of the Rothbart Infant Behavior Questionnaire (Center for Human Resources Research, 2006). Children of different ages received different sets of temperament items. The present study used 11 items administered to all of the children under 23 months to construct infant behavioral scales. In addition to these common items, six other items were administered to children under 11 months. These items were excluded from the scale construction, because they focused more on measuring very young children's activity levels and behavioral routines (i.e. sleepy and hungry at the same time), which are not the primary interest of this study and were administered only to part of the age range of the children in the sample.

For each of the 11 items, the mothers rated their infant's behavior using a 5-point scale, where a higher value indicated a higher frequency of occurrence (i.e. 1, almost never, 2, less than ½ the time, 3, about ½ of the time, and so on). To construct the infant behavioral scales, this study followed the procedures described in previous studies and the guidelines from the NLSY User's Guide (Center for Human Resources Research, 2006; Lahey et al., 2008), which recommend that researchers differentiate three dimensions of infant behavioral outcomes: (a) positive mood (i.e. the child's frequency of smiling or laughter in any situation); (b) fearfulness (i.e. the child's level of distress for novel people and situations) and (c) difficultness (i.e. the child's level of fussiness and ability to be comforted). A total score was obtained by summing across all the items in each dimension. These totals were then standardized to facilitate interpretation and comparison across different behavioral ratings.

Maternal drinking during pregnancy

The NLSY asked every woman to report her drinking behavior during each pregnancy between 1986 and 2000 retrospectively. The mothers were first asked whether they had consumed any alcohol in the 12 months before the birth of the child. Next, if the answer was ‘yes,’ they were asked to report the frequency of alcohol consumption during the pregnancy: never (33.8%), less than once a month (32.0%), about once a month (17.3%), 3 or 4 days a month (7.8%), 1 or 2 days a week (6.6%), 3 or 4 days a week (1.5%), nearly every day (0.7%) and every day (<0.1%).

This information was coded in two ways. First, this study created a dummy variable to indicate whether a mother drank any alcohol during each pregnancy. Second, the frequency of prenatal drinking was separated into three levels: no drinking, light-to-moderate drinking (defined as less than 3 or 4 days a month) and heavy drinking (defined as more than 1 or 2 days a week). These two variables were the key explanatory variables in the following statistical analyses.

Control variables

In contrast to studies relying on clinical samples involving the local recruitment of subjects, the NLSY database includes rich information on maternal characteristics, demographic variables and family socioeconomic status. Time-invariant control variables include race and ethnicity, maternal education and grandmother and grandfather's education. This study also controlled for several time-varying variables that are known to be related to child outcomes, including maternal characteristics, such as poverty, marital status, age and the use of prenatal visits in the first trimester of the pregnancy, as well as child characteristics, such as gender, birth order and the child's age at the time of assessment. Finally, with the availability of extensive measures of maternal cognitive and behavioral skills in the NLSY, this study also included the mothers' Armed Force Qualification Test (AFQT) test scores, locus of control (from the Rotter Locus of Control Scale), self-esteem (from the Rosenberg Self-Esteem Scale) and self-rated sociability as covariates.

Statistical analyses

To estimate the relationship between maternal alcohol use and infant behavioral outcomes, this study first evaluated the data using ordinary least squares (OLS) regressions, and controlled for a wide range of maternal characteristics, including measures of maternal cognitive and behavioral skills. Nonetheless, the incidence of women who participated in drinking during pregnancy was not randomly distributed in the population of pregnant women. The impact of maternal alcohol use during pregnancy on an infant's behavioral outcomes could reflect characteristics not often measured in surveys. Traditional OLS regression, however, cannot account for such unobserved variable bias; a sibling fixed-effects model is a better alternative to deal with this unobserved variable bias. A sibling fixed-effects model compares siblings with the ‘treatment’ of interest with siblings without the treatment in the same family. This strategy allows anything that is shared across births to be factored out. If, among otherwise similar families, children who were exposed to prenatal drinking show higher levels of behavioral problems than those with no exposure to alcohol, a causal relationship between prenatal drinking and behavioral outcomes is plausible. The rationale of sibling fixed-effects can be expressed in the following equations:

1

2

Here, the subscript i refers to a child and subscript j refers to the family (mother) of the child. Y is the outcome of interest. D represents maternal drinking while pregnant. X_k is the vector of n covariates, which serve as control variables. In addition, µ refers to the family-level, time-invariant unobserved factors. If these unobservable factors are simultaneously correlated with infant behavioral outcomes and prenatal drinking, traditional OLS estimates (as presented in the first equation) would yield a biased estimate of β. The sibling fixed-effects estimator removes bias from the time-invariant maternal and family components of µ by subtracting the average for all siblings in a given family from each child's value (represented in the second equation) and provides better estimates of the effects of maternal alcohol use during pregnancy on infant behavioral outcomes.

RESULTS

Descriptive statistics

Descriptive statistics are presented in Table 1. These statistics are reported at the maternal and child levels, because each mother in the sample had multiple children. Less than half of the sample was non-drinking mothers, while mothers who drank during all of their pregnancies and mothers who drank during at least one pregnancy constituted 17% and 29% of the sample, respectively. The mean age of the mothers in 1986 was 24.5 years, and, on average, each had two children. Mothers who drank during their pregnancies were more likely to be white and better educated, and they were less likely to be poor. In contrast, mothers who did not drink at all or who drank only during some of their pregnancies tended to be African-American and Hispanic, with lower levels of education. Furthermore, drinking mothers showed higher levels of marital stability, but were more likely to smoke while pregnant. Mothers who changed their drinking behavior were less likely to live in a stable environment and were more likely to have experienced short-term poverty, and 24% had a change in marital status between pregnancies.

The bottom half of Table 1 shows selected characteristics of the children. Infant behavioral outcomes varied with maternal drinking behavior during pregnancy. Mothers who drank while pregnant tended to have children with slightly lower levels of positive mood and higher levels of difficultness. Children born to non-drinking mothers showed higher levels of fearfulness compared with children born to mothers who drank while pregnant.

Effects of maternal alcohol use on infant behavioral outcomes

Estimates of the effects of alcohol use on infant behaviors are reported in Table 2. Table 2 shows the estimates of the overall effects of maternal drinking during pregnancy, regardless of the level of alcohol used. In the basic OLS model (Model 1), drinking while pregnant was associated with a lower level of positive mood of 0.25 standard deviations (P < 0.001). Children whose mothers drank while pregnant scored 0.20 standard deviations higher (P < 0.001) on the difficultness rating. Prenatal drinking appeared to have no effect on children's fearfulness. None of the maternal demographic variables significantly correlated with any infant behavioral outcomes examined. However, some maternal cognitive and behavioral skills were strong predictors of infant behavioral outcomes. For example, maternal self-esteem had a significant positive effect on the children's difficultness. An additional unit increase in the maternal self-esteem rating was associated with an increase of 0.02 standard deviations (P < 0.001) in children's positive mood and with a decrease of 0.03 standard deviations (P < 0.001) in child difficultness. The mother's sociability was also negatively correlated with fearfulness among children.

Model 2 in Table 2 employed the sibling fixed-effects model. Compared with the estimates obtained from Model 1, the negative effect on positive mood disappeared. This suggests that the previously identified negative relationship between prenatal drinking and children's positive mood might be spurious (i.e. the result of other unobserved variables). With respect to difficultness, the negative effect remained sizable. The results showed that drinking while pregnant is positively associated with an infant's level of difficultness. Infants whose mothers drank while pregnant scored 0.26 standard deviations higher (P < 0.001) on the difficultness rating compared with infants whose mothers did not drink. In contrast, maternal alcohol use during pregnancy appeared to have no effect on the positive mood or fearfulness of the infants.

Table 3 shows the sibling fixed-effects estimates by levels of alcohol use. As Table 3 suggested, infants' positive mood and fearfulness were not affected by heavy drinking during pregnancy. However, there were some interesting findings regarding infant difficultness. Estimates suggested that the negative effects of prenatal drinking on infant difficultness were not limited to women who drank heavily during pregnancy. Although infants whose mothers drank heavily while pregnant scored 0.36 standard deviations higher (P < 0.05) on the difficultness rating than infants of non-drinkers, light-to-moderate drinking also increased infants' difficultness ratings by 0.20 standard deviations (P < 0.01). Thus, it appears that the adverse behavioral consequences of light-to-moderate drinking are not trivial. In addition, the estimates from Table 3 provided suggestive evidence that the negative impact of prenatal drinking increased linearly with the amount of alcohol consumed. Further explorative analysis by using the original categories of frequency of alcohol use found partial support for the hypothesis. However, the data do not allow a closer examination of this linear relationship among heavy drinkers, because the sample size in each heavy drinking category was too small to render significant results. Finally, the association between heavy drinking and child fearfulness was sizable. Even though the association was not statistically significant, the sizable effect suggested that heavy drinking may be a risk factor for fearfulness as well. More studies are definitely needed to investigate the relationship between prenatal drinking and fearfulness.

To provide further evidence of the effect of alcohol and infant difficultness and test the sensitivity of behavioral development over physical growth as it relates to prenatal drinking, the subsequent analysis expanded the previous sibling fixed-effects model to include another important outcome resulting from prenatal consumption of alcohol: low birth weight. The fourth column of Table 3 shows these results. The estimates showed no association between light-to-moderate drinking and low birth weight, but mothers who drank heavily during pregnancy had an increased chance of 10% (P < 0.05) of giving birth to a child with a low birth weight. Maternal smoking during pregnancy was also associated with a higher chance of having a child with a low birth weight. These findings are consistent with the conclusions of prior research: smoking is a strong risk factor for low birth weight. Heavy drinking seems to reduce children's birth weights as well, whereas low-to-moderate levels of alcohol use appear to show no effect on birth weight.

Results from these tests, therefore, offer a more comprehensive picture of the impact of prenatal drinking on child development. These results suggest that light-to-moderate drinking may convey little harm to an infant in terms of birth weight. By contrast, light-to-moderate drinking while pregnant has a substantial negative impact on a child's difficultness. These findings suggest that infant difficultness is more susceptible to the influence of maternal alcohol consumption during pregnancy. Results from Table 3 thus provide partial support for the differential susceptibility hypothesis that has been suggested in previous animal studies.

Finally, this study performed two additional sets of analyses in order to check the robustness of these findings. First, this study estimated models that included interaction terms for maternal alcohol use and gender, birth order, race and maternal education. This was done in order to examine whether the negative consequences of alcohol use on infant difficultness differed by child and/or maternal characteristics. The results are presented in Table 4. Interestingly, none of these interaction terms was statistically significant. Therefore, there was no strong evidence that the impact of prenatal drinking on infant mental health differs across gender, birth order, race or maternal education status. Second, this study examined whether the results were altered by the use of different selection criteria in sampling. The sibling fixed-effects model relies on the assumption that unobserved characteristics are time invariant. Intuitively, however, it seems that the longer the time between births, the more likely it is that this assumption may fail. To deal with this issue, the new analysis excluded children who were born >60 months apart. Results are presented in Table 5. Although the magnitudes of the estimated coefficients did change slightly, the patterns were quite similar to those in the previous analyses. Prenatal drinking remained positively correlated with the infants' difficultness, and the harmful effects of light-to-moderate alcohol consumption remained sizable and statistically significant. Thus, the results held, regardless of the potential problems implicit in the assumptions of fixed-effects modeling.

DISCUSSION

The present study provides additional evidence for the impact of prenatal alcohol exposure on early developmental outcomes using data from the NLSY. The results show a negative effect of maternal alcohol use on infant difficultness, even for light-to-moderate drinking. By contrast, there is no significant association between prenatal alcohol exposure and positive mood and fearfulness in children. Furthermore, this study suggests that prenatal alcohol exposure has a greater effect on infant difficultness than it does on infant birth weight. Although occasional drinking during pregnancy does not have an adverse effect on birth weight, infrequent alcohol consumption produces a statistically significant positive association with infant difficultness.

The present study has several limitations. First, by restricting the study subjects to children born between 1986 and 2000, the sample consists of infants born to relatively older mothers. Women that have more education and come from relatively higher socioeconomic strata tend to give birth at older ages (Rindfuss and John, 1983). As a result, this sampling strategy may provide a more conservative estimate of the impact of maternal drinking during pregnancy than many epidemiological surveys that sampled high-risk populations. Second, results from the sibling-based analyses may not generalize to children without siblings. However, families with a single child may not be the contemporary norm in the USA. The total fertility rate per woman in the USA is approximately two and the predominant ideal family size in the USA is made up of two to three children (Hagewen and Morgan, 2005). Results from sibling comparison are therefore still relevant with regard to a large portion of the US population. Third, the results from these analyses may not generalize to different countries and societies. Fourth, the assessment of maternal alcohol use during pregnancy in the NLSY was based on frequency, which did not include information on the amount of alcohol consumed on each occasion. However, the magnitude of the association and the results are consistent with previous research. Furthermore, explorative analysis of the NLSY substance use data measured during the non-pregnant period suggested a strong positive association between frequency of alcohol use and amount of alcohol consumption on each occasion. In fact, binge drinkers are among the most frequent drinkers. Prior episodes of binge drinking also strongly relate to frequent alcohol consumption while pregnant. Therefore, frequency of alcohol use remains a meaningful indicator of the levels of maternal alcohol use during pregnancy. Finally, maternal ratings of infant behaviors are subjective, and thus may vary from mother to mother. However, the sibling fixed-effects model compared siblings' behaviors that were rated by the same mother, and therefore eliminated (or at least minimized) the problem of different evaluative standards. Furthermore, some child psychologists argue that maternal reports generally provide valid and reliable measures of infant behaviors in everyday life, and this may capture the infant's behaviors better than lab assessments that measure only a snapshot of the child's behaviors in a contrived environment (Fox and Henderson, 2000).

Limitations notwithstanding, this study makes several significant contributions to the literature. Most importantly, the sibling fixed-effects modeling technique removed potential biases due to unmeasured time-invariant family-specific characteristics. While the sibling model cannot fully account for all unobserved heterogeneities, it provided better estimates on the relationship between prenatal alcohol exposure and temperamental outcomes than the traditional OLS model. In addition, this study examined infant behavioral outcomes. Longitudinal evidence has demonstrated a strong link between difficultness and childhood problem behaviors, which suggests a continuity of early problem behaviors. For example, using data from a prospective longitudinal study, Guerin et al. (1997) found that infants who scored high on difficultness when they were 1.5 years old had higher ratings of aggressive behaviors during school age, as rated by parents and teachers. Another study by Keenan et al. (1998) also found that a high level of difficultness measured between 1 and 2 years of age was significantly associated with boys' internalizing problem behaviors and externalizing problem behaviors by age 3. Thus, infant difficultness is a strong predictor of behavioral problems in childhood, and recent reviews of the literature show that infant difficultness is an indicator of early behavioral problems, which can be used to identify children at high risk for problem behaviors (Bates et al., 1985; Frick and Morris, 2004). Findings from this study thus broaden the scope of those obtained from prior studies, and provide further evidence of a link between prenatal drinking and offspring behavioral problems in early periods of life. Moreover, this study investigated the differential susceptibility of prenatal drinking in infant behavioral outcomes and birth weight. Results provide the very first evidence, and partial support, of the relative sensitivity of infant behavioral outcomes over birth weight using observational data involving humans. Finally, this study's findings are more generalizable than those obtained in prior clinical studies, because the present results are based on a sample of the offspring of women in a large national study, including women from diverse social and economic backgrounds. In terms of public health practices, the present study also provides information that may have significant implications for public policy. Findings imply that interventions that aim to reduce maternal drinking while pregnant may generate improvements on children's behavioral well-being. In addition, given that a substantial proportion of Americans do not consider light drinking during pregnancy to be hazardous (MacKinnon et al., 1995), these findings also send an important message to the general public. Nevertheless, additional studies are definitely needed to inform theories about the etiology of child behavioral problems and provide necessary information for public health campaigns.

References