|Home | About | Journals | Submit | Contact Us | Français|
Smoking during pregnancy has been linked to an increased risk of several adverse birth outcomes. Associations with deficits in cognitive development have also been suggested. It is unclear if these associations are due to genetic and/or environmental confounding. In a population-based Swedish cohort study on 205 777 singleton males born to Nordic mothers between 1983 and 1988, we examined the association between maternal smoking during pregnancy and the risk of poor intellectual performance in young adult male offspring. In the cohort analyses, the risk of poor intellectual performance was increased in sons of smoking mothers compared to sons of non-smokers. Stratifying for maternal smoking habits across two pregnancies, there was an increased risk of poor intellectual performance for both sons if the mother was only smoking in the first pregnancy, but in neither son if the mother was only smoking in the second pregnancy. The effect of smoking during pregnancy on intellectual performance was not present when the association was evaluated within sibling pairs. Thus, the association between prenatal smoking exposure and offspring risk of low intellectual performance appears to be completely confounded by familial (genetic and early environmental) factors.
Maternal smoking during pregnancy has been causally linked to fetal growth restriction, and probably also with placental abruption, preterm birth, stillbirth and sudden infant death syndrome (SIDS).1–3 Several studies have found evidence of associations between prenatal exposure to smoking and subsequent deficits in childhood, including cognitive development, school achievement and behavioral adjustment.4–10 An association between smoking during pregnancy and offspring adult intelligence has also been suggested.11
Smoking during pregnancy is correlated with low parental education and socioeconomic status, poor parenting, early age at childbearing and use of other drugs during pregnancy, all potential risk factors for poor outcomes in children.6, 8, 12–15 Two U.S studies, using retrospectively collected information about smoking during pregnancy, found that the association between prenatal smoking exposure and offspring intellectual abilities was substantially or entirely confounded by maternal factors, foremost maternal IQ and education.14, 15
Sibling studies is an alternative approach to reduce confounding when investigating effects of prenatal smoking exposure in offspring. Since siblings share childhood environment and half of their segregating genes, sibling analyses adjust for unmeasured familial (common genetic and shared environmental) factors. Importantly, studies of successive pregnancies also allow us to investigate how change of smoking habits between pregnancies influence intellectual performance in offspring.
We investigated the association between exposure to maternal smoking during pregnancy and subsequent risk of poor intellectual performance in young adult Swedish males, taking both parental and birth characteristics into account. Further, to investigate if the association was confounded by unmeasured familial factors (shared environment and common genes), we studied maternal smoking habits across two pregnancies.
Five Swedish population-based registers were used to obtain data for the study: the Medical Birth Register, the Multigeneration Register, the Population and Housing Census of 1990, the Education Register of 2000 and the Swedish Conscript Register. The records were linked using the unique national registration number assigned to each Swedish citizen at birth and included in each register.
The Swedish Medical Birth Register was used to obtain data on maternal characteristics (age at delivery, parity, smoking, pre-pregnancy weight and height) and birth characteristics (gestational age, birthweight and head circumference). This register contains data on more than 99% of all births in Sweden since 1973.16 During pregnancy and delivery, information is prospectively recorded in standardized records. More than 95% of pregnant women in Sweden attend prenatal care before gestational week 15.17 A previous study on the Birth Register has found that the quality of the variables included in this study is high.16
At the first prenatal care visit, women report whether they are daily smokers (categorized as moderate [1–9 cigarettes per day] or heavy smokers [≥10 cigarettes per day]) or not. Maternal body mass index (BMI) before pregnancy was calculated as the ratio between weight in kilograms and the square of the height in meters (kg/m2), and was categorized according to the WHO recommendations.18 Family situation was categorized into parents living or not living together in early pregnancy. Birthweight and head circumference for gestational age were expressed in standard deviation scores for gestational age according to Swedish standard curves.19
The Multigeneration Register was used to identify the biological father of each son. Data on the parents’ socioeconomic status (SES) was collected from the Population and Housing Census of 1990 and classified according to the recommendations by Statistics Sweden.20 The Education Register of 2000 was used to establish the educational level of the parents. A validation study has shown that the highest level of formal education is correctly reported for 83% of the population.21
The Swedish Conscript Register includes information about Swedish males conscripted for military service. Intellectual performance and age at conscription were obtained at conscription. The only information obtained from the Conscript Register regarding intellectual performance (IP) is that it is tested in 4 dimensions: logical/inductive, verbal, spatial and theoretical/technical. The time-limited test consists of 40 questions for each dimension. Results are presented as a combined standard nine (stanine) score with mean 5 and standard deviation 2. Poor intellectual performance in the current study was defined as a score of ≤2. These individuals can be expected to have difficulties coping with basic educational programs. While there was no formal evaluation of the data quality available, data from the Conscript Register has been successfully used in previous studies on intellectual performance.22, 23
Data on 232 378 Swedish live-born males born between 1983 and 1988 and conscripted between 2000 and 2006 was obtained from the Swedish Birth Register linked with the Swedish Conscript Registry. In order to obtain a more homogeneous study population, we excluded 13 240 males born to non-Scandinavian mothers, 4 260 males born in multiple births and 9 101 males with congenital malformations. Fewer non-Scandinavian mothers smoked during pregnancy, but the frequency of poor intellectual performance was higher in this group, indicating that these offspring may be exposed to other risk factors. For males born in multiple births and with congenital malformations, the frequency of maternal smoking and of low intellectual performance did not differ significantly from those in the cohort. Of the remaining 205 777 males, we obtained data on intellectual performance from 172 182 individuals. Information on maternal smoking habits during pregnancy was available for 94% of the study population.
The sibling subset included 14 722 full sibling pairs. The first two sons with data on intellectual performance were selected from each family with more than one child in the cohort. The small sample size of half siblings (N=386 pairs) limited our statistical power to investigate the association within these pairs. Complete information on maternal smoking during pregnancy was available for 88% of the sibling pairs. In order to get an estimate from the unrelated individuals, one son per mother was randomly selected from the total sample to represent the unrelated group (N=156 663), where 94% had information on maternal smoking during pregnancy.
We used generalized estimating equation models24 to estimate risks of poor intellectual performance in early adulthood in relation to maternal smoking during pregnancy, other parental characteristics and birth characteristics. A correlation matrix was specified in order to control for the dependence between siblings. Analyses were performed using Proc Genmod in SAS (SAS Institute, Inc, Cary, NC) and risks were presented as odds ratios (OR) with 95% confidence intervals [CI]. Associations between smoking during pregnancy and intellectual performance were estimated using both univariate and multivariable analyses. Risks of poor intellectual performance associated with smoking during pregnancy were investigated in three multivariable models. In the first model, we adjusted for parental characteristics (mother’s age, parity, height and BMI, family situation in early pregnancy, parents’ educational level and socioeconomic status) and age at conscription of the son. In the second model, we adjusted for birth characteristics (gestational age, birthweight and head circumference for gestational age), and in the third model, we adjusted for both parental and birth characteristics simultaneously. The final models only included variables that significantly influenced the risk of poor intellectual performance.
To examine whether the effect was influenced by unmeasured familial factors, we proceeded in two ways. First, we estimated the risks of poor intellectual performance in first- and second-born sons, stratified by maternal smoking habits across pregnancies. Risks of poor intellectual performance by changes in maternal smoking habits (non-smoker, continuous smoker, starting or quitting smoking between pregnancies) were estimated in first- and second born siblings separately using logistic regression models.
Second, to further examine the importance of familial confounding, we estimated the effects of smoking during pregnancy on intellectual performance in unrelated individuals and within sibling pairs. The effects of smoking during pregnancy on offspring intellectual performance within and between families were estimated using Proc Mixed in SAS. In these analyses, the exposure was decomposed into between- and within-family components.25 The between-family component was measured by the maternal mean of smoking across pregnancies, and the within-family component was centered on the family-mean for each son. The within-family component estimates the difference in intellectual performance between two siblings.25 For comparison, the effect of smoking on the continuous measure of intellectual performance was also estimated in an unrelated subset of the cohort. If familial confounding is present one should expect the estimate of the within component to decrease compared to the effect seen between unrelated males. Due to the complexity of these analyses we treated smoking during pregnancy as a dichotomous variable (smokers vs non-smokers). The stanine scale of intellectual performance was used as outcome in these models.
Parental and birth characteristics in relation to offspring risk of poor intellectual performance are presented in Table 1. Risks of poor intellectual performance increased with declining maternal age and height. Sons of overweight (BMI 25–29) and obese (BMI ≥ 30) mothers also showed increased risks of low test scores. The risks were also highly influenced by socially related factors, including lower education and socioeconomic status of mothers and fathers, and single motherhood.
Males with a low birthweight or a small head circumference for gestational age were at higher risks of poor intellectual performance than males born with a normal birthweight or head circumference, respectively. Compared with males born at term (37–41 weeks), the risk of a low score on the general intellectual performance test was slightly higher for those born moderately preterm (32–36 weeks) and substantially higher for very preterm (≤31 weeks) born individuals.
Maternal smoking habits during pregnancy were closely associated with parental socioeconomic and educational status. More than half of the teenage mothers were smokers in contrast to less than 25% of mothers aged 30 years or more. Mothers of low education level (49%) and low socioeconomic status (39%) were more likely to smoke than mothers with the highest level of education and high-level white collar workers (14% in both groups). Maternal smoking habits during pregnancy were similarly associated to lower paternal educational level and socioeconomic status. More than half of mothers not living with the father of the child were smokers. Males with a low birthweight for gestational age and males born preterm were far more likely to have smoking mothers (51 % and 44%, respectively) than males born with a normal birthweight or born at term (29% in both groups).
Table 2 displays the association of smoking during pregnancy and intellectual performance as crude odds ratios and odds ratios adjusted for parental and birth characteristics. In the crude analysis, sons of moderate and heavy smokers were at increased risks of having a low score on the intellectual performance test (OR = 1.70 [95% CI 1.63, 1.78] and 1.91 [95% CI 1.81, 2.00], respectively). After adjusting for parental characteristics, corresponding odds ratios were decreased, but remained significant (OR = 1.27 [95% CI 1.19, 1.34] and 1.22 [95% CI 1.14, 1.31], respectively). While all parental characteristics were significantly associated with the outcome, the reduction in the smoking-related risk was primarily attributed to adjustment for mother’s and father’s educational level. Next, to examine whether the effects on intellectual performance were mediated by fetal growth restriction, adjustments were made for birth characteristics. The decrease in risk was negligible, although birthweight and head circumference were found to be significantly and independently associated with poor intellectual performance. When we in the final model adjusted for both birth and parental characteristics, smoking during pregnancy was still significantly associated with increased risk of poor intellectual performance (OR = 1.25 [95% CI 1.18, 1.33] for sons of moderate smokers and 1.21 [95% CI 1.12, 1.30] for sons of heavy smokers).
Analyzing intellectual performance as a continuous variable showed the same increase in risk of low intellectual performance with increasing level of smoking (data not shown). Similar effects on intellectual performance could also be seen in the original dataset (including males born to non-Scandinavian mothers, in multiple births and with congenital malformations) and in complete case analyses (data not shown).
We also investigated possible interactions of smoking with maternal age, education and socioeconomic index. None of the tested interactions were significant in the adjusted analysis (data not shown).
The results from comparing the risk of poor intellectual performance between full brothers in relation to the mothers’ smoking across pregnancies are shown in Table 3. In the crude analysis, the risk of a low score on the intellectual performance test was nearly doubled among sons of mothers who smoked in both pregnancies, compared with sons of mothers who did not smoke in any pregnancy (OR = 1.85 [95% CI 1.57, 2.18] for the first son and 1.89 [95% CI 1.65, 2.17] for the second son). If the mother smoked in first but not in second pregnancy, the risk of poor intellectual performance was similarly increased in both sons. If the mother had started smoking after the first pregnancy, both sons also seemed to have an increased risk of a low score. In the adjusted analyses, all smoking-related risks were decreased. However, not only smoking in both pregnancies but also smoking in first but not in second pregnancy was still associated with increased risks of poor intellectual performance in both sons (OR = 1.40 [95% CI 1.03, 1.91] for the first son and 1.39 [95% CI 1.08, 1.79] for the second son). Compared to mothers not smoking in either pregnancy, maternal smoking only in the second pregnancy was not associated with an increased risk of poor intellectual performance in either son.
Finally, to investigate the importance of familial factors, we studied the effect of smoking during pregnancy on intellectual performance between unrelated sons and within siblings (Figure 1). Along the lines of preceding results, sons of mothers who had smoked during pregnancy scored approximately 0.65 points less than sons of non-smoking mothers in the unrelated subset [95% CI −0.67, −0.63]. Adjusting for parental characteristics decreased this difference to 0.24 points [95% CI −0.27, −0.22]. In the sibling subset, there was a small but significant increase in intellectual performance for the sons exposed to smoking during pregnancy (regression coefficient β = 0.22 [95% CI 0.10, 0.34]). However, after adjustment for birth order and maternal age, this effect was no longer significant (β = 0.06 [95% CI −0.06, 0.18]).
Analyzing the complete dataset without exclusions showed virtually the same results (data not shown).
We found that the association between maternal smoking during pregnancy and poor intellectual performance in young adult male offspring can be explained partly by parental education but also by unmeasured familial factors, such as common genes and shared environment.
In the cohort analyses, more than half of the smoking-related increased risk was accounted for by parental educational level, indicating that the association largely depends on familial factors. Two recent studies reported that the association between maternal smoking and offspring intellectual performance was diminished after adjusting for maternal education and IQ.14, 15 Our results in relation to these studies indicate that other familial effects besides education are of importance for the association. In the current study, smoking-related birth outcomes did not mediate the association between smoking during pregnancy and poor intellectual performance in the offspring. This indicates that any effect of prenatal smoking exposure on intellectual performance in offspring is independent of fetal growth or gestational age.
In the sibling subset of the studied cohort, we found an increased risk of poor intellectual performance for both sons if the mother was smoking in the first pregnancy, regardless of her smoking habits during the second pregnancy. In contrast, if the mother was only smoking in the second pregnancy there was no increase in risk for either son. The finding that change of exposure to maternal smoking across pregnancies seems to be unrelated to offspring risk of poor intellectual performance does not support the hypothesis that prenatal smoking exposure influence intellectual performance in offspring.
The effect of maternal smoking on offspring intellectual performance could not be seen within full sibling pairs, suggesting that the association seen in the cohort is confounded by unmeasured familial (genetic or environmental) factors. Similarly, a recent study on siblings from the Collaborative Perinatal Project found no effect of maternal smoking during pregnancy on twelve different outcomes of cognitive and physical development within 2064 sibling sets where the mother’s had changed smoking habits between pregnancies.26 In the current study, the apparent positive effect of prenatal smoking exposure on intellectual performance within siblings was lost after adjusting for birth order. Though debated, intellectual performance has been negatively associated with increasing birth order within families in previous studies.27, 28
The data on maternal smoking during pregnancy was collected at the first prenatal care visit. The proportion of women who stop smoking during pregnancy has been reported to be around 30% in previous Swedish studies. In these studies, women with a higher level of education and a higher age at the initiation of smoking were more likely to stop smoking when pregnant, while women not living with the father of the child, heavy smokers and women exposed to daily passive smoking at home were more likely to continue smoking throughout the pregnancy.29, 30 Thus, misclassification of smoking exposure during pregnancy may have influenced the obtained risk estimates. Finally, most pregnant smokers continue to smoke after pregnancy, and we cannot distinguish between the effects of prenatal and postnatal smoking exposure.
Some strengths of this study are its size, the prospective design and the range of covariates gathered from population-based registers.16 An advantage of measuring intellectual performance in early adulthood is that it is less affected by postnatal environment than other measures of cognitive ability in early ages. Intelligence measured in children and even more so scholastic achievement are likely to be influenced by motivation and ability to concentrate, and might consequently reflect attention deficits or behavioral problems rather than life-long cognitive ability. A limitation of this study was residual confounding that we could not control for by parental education and SES, such as parental intellectual abilities and behavioral problems, parenting and environmental stimulation, passive smoking and maternal alcohol consumption during pregnancy. We were also unable to control for the duration of breast feeding. However, while breast feeding has been positively linked to intellectual development,31 other studies have found no significant advantage of breast feeding on offspring intelligence when adjusting for maternal characteristics and home environment.32, 33
In conclusion, our results support the importance of familial factors to the association between mothers smoking during pregnancy and poor intellectual performance in young adult males. Further research in this field is warranted to clarify the nature of familial risk factors and their potential role in the association between smoking during pregnancy and intellectual performance.
This project was supported by grants from the Swedish Research Council (Project number K2007-70P-20518-01-4).