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

5. Results

5.1 Results of the ANCOVA

ANCOVA analyses were carried out on a total of 79 child outcome measures collected as part of the BBBF longitudinal study when children were 33 months, 48 months, 6 years (Grade 1) and 8 years (Grade 3) of age. Each measure was independently analyzed using Analysis of Covariance (ANCOVA) procedures. For each measure, children exposed to tobacco during pregnancy were compared with those not exposed, yielding the Prenatal Exposure to Tobacco, or PET effect. Also, children whose mothers were considered higher risk for problem drinking during pregnancy based on a CAGE score of 1 or higher were compared with children whose mothers were considered lower risk, the Maternal Higher-Risk Drinker, or MHRD effect. Finally, for each measure, the PET and MHRD group was compared with the non-PET and non-MHRD group. This comparison was designed to determine the effects of the combination of prenatal exposure to tobacco plus maternal higher-risk drinking during pregnancy. For all analyses, the set of family socio-economic, cultural and demographic variables listed in Appendix 2 were employed as covariates to eliminate confounding effects of these variables associated with maternal prenatal tobacco use or higher-risk drinking. Appendix 1 presents a summary of the results of the three statistical comparisons of all 79 child outcome measures.

Since these three hypotheses are all directional in nature, one-tailed statistical tests of significance were used for all analyses. Also, the results in Appendix 1 are presented in terms of whether each analysis yielded a difference in means in the hypothesized direction (i.e. poorer performance represented by a negative sign (–), or a difference in means in the opposite direction from that hypothesized, represented by a plus sign (+)).

In this section, all statistically significant results are presented and described (the non-significant (NS) results are reported in Appendix 1). For each statistically significant effect (p <.01), the effect size (E.S.) is presented where the E.S. reflects how large the mean difference is in standard deviation units.

In social and health science research, an E.S. of .2 to .5 is considered small, .5 to .8 moderate, and >.8 is considered large (Cohen, 1977).

The results are presented separately according to the domain of child functioning reflected by the measures. Group means, standard errors and sample sizes for the analyses of all variables are presented in Appendix 3.

5.1.1 Child General Development

As summarized in Table 5, statistically significant results occurred for all three measures of children’s general development: the Developmental Inventory for Screening Children (DISC) Overall score at 33 and 48 months of age and the ABC School Readiness. Children exposed to tobacco and higher-risk alcohol use by their mothers during pregnancy showed significantly poorer outcomes on the three measures of general development compared with the group of children who were not exposed to either tobacco or high-risk maternal drinking during pregnancy (i.e. the PET and MHRD comparison).

The effect sizes for these group differences ranged from –.50 to –.65, differences that are considered to be moderate in size.

There were no statistically significant effects of exposure to tobacco alone, or exposure to higher-risk maternal alcohol use alone.

5.1.2 Children’s Cognitive Development and Academic Performance Measures

There were a total of 23 measures analyzed in the domain of children’s cognitive development and academic performance, and the results of these analyses are summarized in Table 6. Of the 23 measures, 7 showed statistically significant differences, with the children exposed to both tobacco and mother’s higher-risk drinking during pregnancy showing poorer performance when compared with the group that was exposed to neither. Six of these group differences yielded effect sizes in the moderate range (.52–.69), but the measure of teacher ratings of the child’s school preparedness in Grade 3 yielded a much larger difference, with an effect size of 1.03. (Note: E = directly from trained researcher, T = teacher, and P = parent indicate the source of each measure). Of the 7 significant effects, 6 were on measures collected when the children were in Grade 3, and 5 of these 6 were based on ratings by the child’s teacher.

There were also 4 significant negative effects of children exposed to higher-risk mother’s drinking during pregnancy. There were no differences on any of the 23 measures associated with exposure to tobacco during pregnancy.

5.1.3 Children’s Social/Emotional Functioning Measures

A total of 23 measures of various aspects of children’s social and emotional functioning were analyzed and the results are summarized in Table 7. Only 2 measures yielded statistically significant differences – teachers’ ratings of children’s emotional problems and their ability to manage conflict with peers at Grade 3. Children exposed to both tobacco and higher-risk maternal drinking during pregnancy showed higher levels of emotional problems and poorer conflict management, as rated by their teachers, than those children exposed to neither. Also, children exposed to higher-risk maternal drinking showed higher levels of emotional problems in Grade 3 than children not exposed to higher-risk maternal drinking. Again, there was no indication of compromised social or emotional functioning associated with children being exposed to tobacco during pregnancy.

5.1.4 Children’s Health Measures

Of the 9 measures reflecting children’s health at various ages, only the measure of children being exposed to second-hand smoke yielded statistically significant effects in Grades 1 and 3. This is not, strictly speaking, a child outcome measure; instead, it reflects the fact that children whose mothers reported smoking during pregnancy as well as those who reported smoking and higher-risk drinking during pregnancy also indicated that their children were exposed to more second-hand smoke when in Grades 1 and again in Grade 3. No other measure of child health showed any indication of negative effects associated with either smoking or higher-risk drinking during pregnancy.

5.1.5 Children’s Behaviour Problems Measures

There were 22 measures of children’s behaviour problems. As summarized in Table 9, children who were exposed to both tobacco and higher-risk maternal drinking during pregnancy showed significantly higher levels of several types of behaviour problems than children exposed to neither. Of 22 measures analyzed, 10 were statistically significant, and 9 of the 10 significant effects were on ratings by the child’s teacher, primarily at 4 years of age and again in Grade 3. Further, most of these differences were quite large, yielding effect sizes near or above .80.

Six of the 22 behaviour problem measures were also significantly higher for children who were exposed to higher-risk maternal drinking during pregnancy compared with children not exposed. All 6 of these significant effects were on ratings by the children’s teacher, 3 when the children were 4 years old and 3 when they were in Grade 3. (Note: This is covered in some detail in the discussion section.) Again, there was no indication of an association between children’s prenatal exposure to tobacco and later ratings of behaviour problems.

5.1.6 Summary of Significant Child Outcomes

In this final section of results, we attempt to summarize the main findings of the analyses just described. The first summary is presented in Table 10. Here we show the number and percentages of outcome measures in each child domain that yielded statistically significant results.

The first column presents the results of comparisons between children who were exposed to some tobacco prenatally and those exposed to no tobacco, the PET effect. Out of the 79 measures analyzed, only 2 were statistically significant. Those were the measures of exposure to second-hand smoke at Grades 1 and 3, a finding that indicates the children of mothers who smoked during pregnancy were exposed to more second-hand smoke at ages 6 and 8 than children of mothers who did not smoke during pregnancy. There were no other significant outcomes in any of the other 79 measures associated with smoking versus no smoking during pregnancy. As discussed previously, although exposure to second-hand smoke is unhealthy, it is not truly a child health outcome such as asthma or physical illness, so this effect needs to be viewed with some caution.

Column 2 summarizes the results of comparisons between children whose mothers engaged in higher-risk drinking during pregnancy versus children of mothers who were not higher-risk drinkers, the MHRD effect. Eleven of the 79 analyses (14%) were statistically significant. Six of these 11 effects involved high levels of behaviour problems for children of higher-risk drinking mothers, and 4 involved poorer cognitive development for this group. Thus, the negative effects associated with children of mothers who engaged in higher-risk drinking during pregnancy are manifested primarily in poorer cognitive and academic functioning and also greater manifestation of behaviour problems such as aggression and hyperactivity.

Column 3 of Table 10 (PET and MHRD) summarizes the results of the comparison between the children who were exposed to both maternal smoking and higher-risk drinking during pregnancy versus children who were not exposed to either smoking or higher-risk drinking. Here 24 of 79 or 30% of the statistical comparisons were significant and, as for the MHRD comparisons, the differences were most pronounced in poorer cognitive and academic functioning as well as higher levels of behaviour problems.

In Table 11, the results of the analyses are reorganized according to the age of the child when the statistically significant effects occurred. The picture that emerges from the results in Table 11 is clear. The poorer performance of children exposed to higher-risk maternal drinking during pregnancy, either alone or in combination with prenatal tobacco exposure, occurred predominantly on measures collected when children were 4 years of age or 8 years of age, with very few effects noted at age 6 (Grade 1).

The final way in which we summarized the significant findings is in terms of the three data collection sources. Some measures were collected directly from the child by trained researchers in each neighbourhood. These measures included standardized cognitive and language lists such as the Peabody Picture Vocabulary Test (PPVT); the Wechsler Block Design Test; the Developmental Inventory for Screening Children (DISC), and height, weight, EQAO reading, math and writing scores from school records at Grade 3.

Many of the measures were collected from parents through a lengthy in-home interview.

Of the 79 measures analyzed for this report, 11 were collected by the local site researchers, directly or indirectly from the child.

Of the 79 child outcome measures, 31 were based on parents’ reports, while 37 were provided by the child’s teachers via a teacher report form that they completed on each child in the longitudinal research sample when the children were in junior kindergarten (age 4), Grade 1 (age 6) and Grade 3 (age 8) (See Discussion section.)

The significant outcome results for each of these three data sources are presented in Table 12 separately for each of the five child domains. As in previous summaries, Table  12 highlights the fact that most significant outcomes occurred on measures in the two domains of Cognitive/Academic Performance and Behaviour Problems associated with mother’s higher-risk drinking during pregnancy either alone (MHRD) or also including smoking during pregnancy (MHRD and PET).

Of the 11 statistically significant effects in the domains of cognitive/academic performance, 8 resulted from measures provided by teachers. In the domain for behaviour problems, 15 of the 16 significant effects were based on ratings provided by teachers. Overall, in all five domains of measures, there were 37 significant outcomes and 27 of these were based on teacher-provided data.

5.1.7 Summary of Major Findings of ANCOVA Analysis

5.2 Results of the Structural Equation Modelling

5.2.1 Drinking and Smoking

Although the modelling of the smoking and drinking measures was an exploratory analysis, we were able to use both order effects and structural relationships to simplify the model (we assumed that measures taken at 33 months did not have a causal effect on measures taken at 3 months). We placed the measures of smoking and drinking behaviour prior to the product of those two variables. All variables in this model were manifest variables (i.e. they were measured directly and included the six drinking and smoking variables listed in Figure 1). We chose to limit paths to those between measures of the same behaviour at different times and to measures of different behaviour at the same time. Figure 1 (Drinking and Smoking Structural Equation Model) shows the results of the modelling. With a Root Mean Square Error of Approximation (RMSEA) of .062 and Comparative Fit Index (CFI) and Tucker-Lewis Index (TLI) of .988 and .965, respectively, the model is a good enough fit to the data.

Figure 1

Drinking and Smoking Structural Equation Model

5.2.2 Parent and Teacher Ratings of Internalizing and Externalizing Behaviour

This analysis is a confirmatory factor analysis of the six measures rated by teachers and the six measures rated by parents. The internalizing/ externalizing split is a well-established relationship closely associated with, but not limited to, the Achenbach measures (e.g. Achenbach & Rescoria, 2001). Our preliminary attempts at fitting models that incorporated both teacher and parent data produced either trivial or ill-fit models, so we split the models into a teacher model and a parent model, as presented below in Figure 2 (Parent Confirmatory Factor Analysis) and Figure 3 (Teacher Confirmatory Factor Analysis).

Both models have adequate, but not great goodness of fit (Teacher RMSEA = .091, CFI = .944, TLI = .852; Parent RMSEA = .083, CFI = .963, TLI = .904) with the parent fit appearing to be somewhat better than the teacher fit.

Figure 2

Parent Confirmatory Factor Analysis

Figure 3

Teacher Confirmatory Factor Analysis

Figure 4 

Parent-Reduced Model

Figure 5

Teacher-Reduced Model

Next, we linked the smoking/drinking model with the teacher model and then with the parent model to estimate paths from smoking and drinking measures to latent traits of externalizing and internalizing behaviours (the latent traits are the factors – teacher internalizing, teacher externalizing, parent internalizing and parent externalizing. They are latent in that they are not measured directly; rather, they are inferred from the behaviour of other variables). The unreduced model had 12 paths from the six predictor manifest variables to the two latent trait outcome variables. Using a reverse stepwise technique, we deleted the smallest path with p value greater than .2 until every remaining path on the diagram from the tobacco/alcohol variables to the internalizing/externalizing variables had a p value of .2 or smaller. These results are presented in Figure 4 (Parent-Reduced Model) and Figure 5 (Teacher-Reduced Model).

Both models have good fit (Teacher RMSEA = .048, CFI = .970, TLI = .950, Parent RMSEA = .054, CFI = .964, TLI = .942), with the parent fit appearing to be somewhat better than the teacher fit. Because this is an exploratory analysis, the parent model retains three non-significant paths with p < .2, to facilitate replication. All of the retained paths (path coefficients) in the teacher model are significantly greater than 0.