ARCHIVED - Early primary school outcomes associated with maternal use of alcohol and tobacco during pregnancy and with exposure to parent alcohol and tobacco use postnatally

6. Discussion

Alcohol is well established as a teratogenic substance (Streissguth, Landesman-Dwyer, Martin & Smith, 1980). Experimental animal studies have manipulated alcohol use to cause malformations among offspring; cross-sectional or correlational studies with humans have correlated the presence of similar malformations among infants to retrospectively measured maternal alcohol use during pregnancy; and longitudinal studies with prospective measures of alcohol use in pregnancy have confirmed those same malformations.

Quite sensibly, the early human research literature has been focused on large doses of alcohol and striking malformations in the faces of children (Huizink & Mulder, 2006; Jacobson & Jacobson, 2002; Richter & Richter, 2001). More recent research has focused on the less visible teratogenic effects of alcohol in humans. Problems with executive function – notably attention, impulsive behaviour and hyperactivity – have been demonstrated in correlational studies, as have other behaviour problems including antisocial and delinquent behaviour. Prenatal exposure to alcohol has been associated with internalizing behaviour problem, such as depression and anxiety. Deficits in cognitive functioning and learning are also evident among children with prenatal alcohol exposure, including memory and information-processing difficulties, poor problem-solving skills, impaired planning and response inhibition, lower IQ scores and problems with linguistic, perceptual and motor development.

Recent debate has been focused on dose effects. From a public health perspective, the question of how much alcohol, if any, is safe to drink during pregnancy has sparked substantial debate (Gijsen, Fulga, Garcia-Bourmessen & Koren, 2008; Kelly et al., 2009; Sayal, 2009).

The results of the first set of analyses (ANCOVA) presented above indicate that the children of mothers who report higher-risk drinking during pregnancy manifest a range of compromised developmental outcomes in early primary school compared with the children of mothers who reported lower-risk or no drinking during pregnancy. Out of a total of 79 different measures, 11 (14%) were significant at the 1% level. These negative outcomes occurred most frequently in the domain of children’s behaviour problems (6 of 22 measures), more specifically in higher ratings of aggressive and hyperactive behaviours by Junior Kindergarten (JK) teachers when the children were 4 years old, and again by Grade 3 teachers when the children were 8 years old. In contrast, parent ratings did not indicate significant negative effects of prenatal alcohol exposure on children’s behaviour problems.

The second domain of children’s functioning in which negative associations with prenatal alcohol exposure were evident was cognitive development/academic performance (4 of 23 measures). Statistically significant negative effects were found in auditory and memory performance on the DISC developmental task administered by trained researchers at 4 years of age, and poorer ratings by Grade 3 teachers on measures of children’s school preparedness, attitudes toward academics, and general academic functioning.

Animal studies have confirmed that the alcohol, tobacco has teratogenic effects on the nervous system of the fetus. Although the effects of prenatal tobacco exposure of birth weight and infant growth are well established, the effects on cognitive, behavioural and social/emotional functioning into childhood are less well documented (Cornelius & Day, 2007; Huizink & Mulder, 2006; Richter & Richter, 2001). In our findings, maternal reports of smoking during pregnancy, collected when a child was 3 months old, were predictive of measurable problems in only one of five broad domains: child health (2 of 9 measures). Out of a total of 79 different measures, only 2 (2%) were significant at the 1% level. Since we would expect about 1% of the tests to be significant by chance alone; finding 2% of the tests to be significant is marginal evidence of an effect. Thus, for this approach to the analysis of data and given the large number of tests, we do not have conclusive evidence of smoking effects. A different statistical approach may give different results.

We also compared the children of women who reported both smoking and drinking during pregnancy with women who did neither (i.e. we left out the women who only smoked or only drank). The combined smoking and drinking was predictive in all five broad domains: general development (3 of 3 measures), cognitive development/academic performance (7 of 23 measures), social and emotional functioning (2 of 22 measures), behaviour problems (10 of 22 measures) and child health (2 of 9 measures). Out of a total of 79 different measures, 24 (30%) were significant at the 1% level. The apparent additive effect of the smoking and the drinking is intriguing, but must be interpreted cautiously given the possibility of selection bias. For example, women who smoke and drink during pregnancy may drink more than women who drink but do not smoke, or their nutritional status may be poorer, their body may already be coping with oxidative stress from smoking, or they may be living with a higher level of stress in their lives. In addition, people tend to smoke more when they are drinking.

Note that our use of statistical control techniques for 15 covariates would not have been sufficient to deal with such potential confounds.

The finding that the combination of prenatal exposure to both alcohol and tobacco predicted the most negative long-term effects on children in primary school is, however, consistent with the research literature that has reported that the negative effects of prenatal exposure to alcohol are increased when combined with other potentially harmful substances, including tobacco or non-prescription drugs (Fried, O’Connell & Watkinson, 1992; Fried & Watkinson, 1990).

These findings also point to the importance of any future research to collect information about the use of multiple substances during pregnancy, in order to avoid inappropriate conclusions about the effects of one substance if no information on other substances is collected.

For example, if a study on maternal smoking during pregnancy does not collect information about the mothers’ drinking during pregnancy, and many of the smoking mothers also engaged in high-risk drinking, negative child outcomes may be attributed to prenatal exposure to tobacco when, in fact, they may be more strongly related to exposure to alcohol or the combination of the two substances.

Note also that the amounts of alcohol and tobacco use during pregnancy need not be large. Our criterion for smoking was “any smoking.” Our criterion for higher-risk drinking was a score of at least 1 on the CAGE scale. A score of 1 can be obtained by someone who feels badly about their drinking behaviour, who feels the need to cut back, or who has been criticized by others about their drinking.

Few studies of either prenatal alcohol use or smoking collect information on the use of both substances prenatally. For example, Martin, Dombrowski, Mullis, Wisenbaker and Huttunen (2006) recently reported results from the Helsinki Longitudinal Project indicating that smoking during pregnancy was associated with several negative effects on children’s development at ages 5 and 12 years of age. The authors acknowledged in their conclusions that, “Smoking, drinking and other forms of drug use are correlated, and some of the resulting effects may have been related to maternal drug use during pregnancy. This study was unable to control for pregnancy drug and alcohol use, which is a clear limitation” (p. 499). The study of either prenatal smoking or alcohol use that does not include information on the use of both substances runs the risk of forming conclusions on the effect of one substance while the effects may result either from the use of the other substance, or, as in the present study, the negative effects of using both substances prenatally (see, for example, O’Connor & Paley, 2006).

Further, few studies control for other parent or family variables such as parent education, single-parent status or family income. Several studies have reported that both prenatal alcohol use and prenatal smoking are strongly related to these variables, so if they are not controlled in analyses, the subsequent child outcomes may be more a function of the child’s socio-economic environment after birth, than the smoking or alcohol use prenatally. Consistent with this concern are the findings of several studies that have reported no or much-reduced effects on children’s development of maternal prenatal smoking when socio-demographic factors and postnatal environment were controlled in the analyses (D’Onofrio et al., 2008; McGee and Stanton, 1994).

Several of the negative outcomes of prenatal alcohol abuse and smoking on children’s cognitive development at 33 and 48 months of age were based on results of a standardized test of development administered to children individually by trained researchers. Nearly all of the other negative effects associated with maternal alcohol abuse and smoking during pregnancy occurred in ratings by the child’s teacher at 4 and 8 years of age. The fact that teachers would not have been aware of the mothers’ smoking or drinking behaviour prenatally or when the (now about 8 years old) child was 33 months old strengthens the confidence one can place on these results. The ratings collected from the children’s parents, on the other hand, showed virtually no association between the mother’s prenatal drinking or smoking and children’s later behaviour in the ANCOVA analyses.

The finding of larger effects in teacher report data than in parent report data is consistent with the literature (Brown et al., 1991). Several interpretations of the different results between teacher and parent ratings are possible. One of many is that teacher ratings of children’s behaviour and academic performance are generally considered to be more valid than those of parents, because teachers have extensive experience observing many children whereas parents’ experience is typically much more limited in this regard. Parents are not able to compare their child’s behaviour with those of many other children, while teachers are constantly making such comparisons. Also, since many of the child behaviour problems rated by both parents and teachers involve difficulties in relationships with peers, teachers would have more opportunity to observe a child’s peer interactions than parents. Finally, the negative outcomes in the area of academic performance were based on ratings of the child’s behaviour in the classroom setting, ratings that can be collected only from teachers since parents have no or extremely limited opportunity to observe such behaviours.

The fact that significant effects were present in the teachers’ rating when children were age 4 (Junior Kindergarten) and age 8 (Grade 3) but not age 6 (Grade 1) may reflect the fact that children face major developmental transitions at ages 4 and 8. At 4 and 5 years of age, individual differences in children’s school readiness are viewed as resulting from different levels of maturity in cognitive and social development. The challenges of formal school entry at this age accentuate individual differences in children’s social and cognitive maturity. The finding that children whose mothers reported higher-risk drinking during pregnancy, and particularly if they also reported smoking, showed compromised cognitive development and elevated levels of hyperactive and aggressive behaviours may reflect their delayed social and cognitive development and difficulty in adapting successfully to the challenges of formal school entry at age 4.

At age 7, another major transition begins in normal cognitive development, namely the transition to conceptual thinking or, in Piagetian terminology, concrete operational thought (Piaget, 1964). Delays in cognitive development at this age mean that children cannot successfully adapt to academic tasks requiring the use of concepts in mathematics and reading, resulting in poor school performance and possible frustration and conflict with more mature peers. Consequently, the negative effects on children’s cognitive and social development of prenatal exposure to alcohol and tobacco may be particularly noticeable by teachers at this age.

In the second set of analyses, we used a structural equation modelling technique to address some of the issues not dealt with in the more coarse-grained ANCOVA analyses reported above. We selected a subset of variables measured in Grade 3 that are related to problems in social and emotional function for dependent variables, and we included measures of drinking and smoking exposure at two times in the children’s development – in utero and when the child was 33 months old. The use of the intermediate measures of smoking and drinking behaviour act as a general control for the “third variable problem.” If something that we have not measured is related to both our predictor and our outcome measure, we can get a spurious relationship mediated by that unseen variable. However, if such a variable exists, the spurious effect ought to be more powerful when the smoking or drinking was measured at 33 months of age than when measured at 3 months of age. The more recent measurement ought to “carry” the third variable effect more strongly than the more distant measure.

When we examine the measures of child behaviour collected from teachers, there is clear support for the hypothesis that drinking during pregnancy leads to problems in social and emotional functioning in elementary school, with significant paths from reported drinking during pregnancy and both internalizing behaviours (r = .19) and externalizing behaviours (r = .21). The more proximal measure of maternal drinking when the child was 33 months old does not predict either externalizing or internalizing behaviour. Smoking during pregnancy does predict externalizing behaviour problems (r = .13) and exposure to second-hand smoke at 33 months does predict internalizing behaviour (r = .15). If we interpret this at face value, it suggests that the mother’s smoking behaviour during pregnancy can have effects that are evident 8 years later in a child’s classroom behaviour and that those effects are over and above the effects of more recent (albeit 4 or 5 years ago) exposure to second-hand smoke. Given that the effects of paths are additive, the effect of smoking and drinking combines for an essentially doubled effect on teacher-rated externalizing behaviour.

As with the first set of analyses (ANCOVA), the same analysis using measures of child behaviour collected from parents shows fewer effects. Only smoking during pregnancy is a significant predictor of externalizing behaviour (r = .17).

Structural equation modelling is a correlational technique. While it is a truism that correlation does not prove causation, the findings of our two models interact with the existing literature in a powerful manner. Drinking and smoking during pregnancy are significant predictors of problems in externalizing behaviour noted 8 years later by both teachers and parents and of internalizing behaviour noted by teachers. These predictors are significant even when “competing” for covariance with related measures collected much closer in time to the behaviour data collection. In the context of recent animal and human findings, the most responsible interpretation of these findings is that smoking and drinking during pregnancy cause some problems in Grade 3 and the predictive relationships observed in the SEM is due to the causal impact of tobacco and alcohol use during pregnancy.

How big is the effect we are looking at? Is it merely statistical, or is it of a magnitude that people would notice?

One way to approach this problem is to look at comparative effect sizes. Meyer et al. (2001) presented an array of effect sizes from meta-analyses that allow a researcher to fit a finding onto a scale. In the table below, our own findings have been embedded among other effect size findings, using the r statistic to make them more compatible with the results of SEM. For example, when people with allergic reactions use antihistamines for runny nose and sneezing, the effect size averages 0.11. Prenatal smoking has an effect of .13 on externalizing ratings. If our data are accurate, then abstinence from smoking and drinking during pregnancy ought to have an effect on internalizing behaviour about as powerful as taking non-steroidal anti-inflammatory drugs (NSAIDS) for pain or taking anti-histamines for allergies. Given that the effects of prenatal smoking and drinking are both evident in teacher reports, the effect of the double abstinence may be twice as large. Moreover, there may be another independent relationship with second-hand smoke. Keep in mind that causal inference from a SEM of this type must be cautious in the absence of manipulation of the independent variable.

Data (effect sizes) from the first statistical analysis of this report (i.e. the ANCOVA results) are reported in terms of the d statistic rather than the r statistic. We looked for meta-analytic studies not included in Meyer et al.’s (2001) article that would expand the range of comparators for our results.

Bhutta, Cleves, Casey, Cradock and Anand (2002) reported a meta-analysis of the cognitive and behavioural outcomes of children who were born preterm. When we interpolate from their measure of weighted mean difference (WMD) by dividing by the theoretical standard deviation of the ability test scores, they demonstrate a mean d of about 0.72 for cognitive measures. Thus, the preterm children in the studies they found were about 11 IQ points or .7 standard deviations below the comparator children born at term. This effect is proportional to gestational age (r = .71), showing an increase in the WMD of roughly .67 points or in the d statistic of about .044 per week of prematurity. Our observed mean d statistics for measures of general child development (see Appendix 1) were –.29 for prenatal maternal tobacco use, –.28 for prenatal maternal alcohol use and –.57 for the difference between children of mothers who used both substances and mothers who used neither. Thus our observed d statistics correspond to those observed in Bhutta, Cleves, Casey, Cradock and Anand’s (2002) study at roughly 6,6 and 13 weeks of prematurity for tobacco, alcohol and joint exposure, respectively. In rough terms, smoking and drinking during pregnancy even at the relatively low levels found in our sample seem to be the equivalent of 6 weeks of prematurity – per substance used.

Schachter, Pham, King, Langford and Moher (2001) conducted a meta-analysis of the effects of Ritalin on children and adolescents with ADHD. In general, they found that the medication was effective based on an array of behavioural measures, but noted that the effects as reported by teachers were stronger than those reported by parents. These researchers reported a mean effect size of .78 for teacher reports and .54 for parent reports. Our results for behavioural problem reports showed effect sizes for teachers of .29, .40 and .70 and for parents of .21, .12 and .25 for prenatal tobacco exposure, alcohol exposure and the combined exposure, respectively.

Paolucci and Violato (2004) reported a meta-analysis on the effects of child sexual abuse. These authors reported a variety of weighted mean effect sizes for diagnoses ranging from .16 to .44. The effect size for academic performance was .19. This compares with our effect sizes for all measures of academic and cognitive performance of –.09, –.26 and –.37 for prenatal tobacco exposure, alcohol use and the combined use, respectively. (See Appendix 1.)

Kitzmann, Gaylord, Holt and Kenny (2003) reported a meta-analysis of the effects on children of witnessing domestic violence. On a variety of problems (internalizing, externalizing, social, academic and other), they reported a mean effect size of .40 when comparing children who witnessed violence with those who did not. Our mean d statistics for measures of general development (see Appendix 1) were –.29 for prenatal maternal tobacco use, –.28 for prenatal maternal alcohol use and –.57 for the difference between children of mothers who used both substances and mothers who used neither.

These findings help to put the measured effects of smoking and drinking into perspective. When we compare the effects we found with those observed with three predictors of worse performance (prematurity, witnessing domestic violence and child sexual abuse), there was a strong overlap. When we compare with intervention for ADHD with Ritalin, the effects of combined prenatal substance use are roughly comparable (but reversed in sign) to the impact of Ritalin on behavioural measures. In broad terms, the impact of the substances appears to be comparable to moderate prematurity (6 weeks for each substance, 12 for both), child sexual abuse (either substance, or twice the magnitude if both were used), or witnessing domestic violence (less impact for one substance, greater impact for combined). The impact of ADHD that is reversible by using Ritalin appears to be roughly comparable to use of both substances.

While these comparisons to meta-analytic findings must be considered approximate, they do give a sense of scale. In our sample, the impact of smoking and drinking during pregnancy is of a magnitude that compares with prematurity, sexual abuse, witnessing domestic violence and the Ritalin-reversible effects of ADHD. It is also important to note that these are aggregate findings, and it is hard to imagine that every child would show the same magnitude of effects – there will be lots of variation.

In conclusion, the results of the present study suggest that children whose mothers report higher-risk alcohol consumption during pregnancy show long-term negative outcomes in measures of school performance and behaviour problems compared with mothers who report lower-risk drinking. These problems are accentuated in children whose mothers also report smoking during the pregnancy. Further, the negative effects are most apparent when children are 4 years of age, and faced with the challenges of formal school entry (i.e. poor school readiness), and again at age 8, when individual differences in conceptual thinking may be particularly salient to teachers. The percentage of measures reflecting the disadvantage of children exposed to prenatal alcohol and tobacco increased from 37% at age 4 to 47% at age 8. If this pattern continues, the negative effects on children’s academic and social behaviour may continue to be compromised as they enter early adolescence; that is, prenatal exposure to maternal high-risk drinking and smoking may be linked to disrupted cognitive and social development at critical periods in children’s development, with lifelong consequences.