Effect of maternal smoking in pregnancy and childhood on child and adolescent sleep outcomes to 21 years: a birth cohort study

The Mater-University of Queensland Study of Pregnancy (MUSP) comprises a birth cohort of 7223 singleton infants born during 1981–83, with mothers being enrolled at the first antenatal visit (average 18 weeks gestation) [28]. Questionnaires were completed by mothers at enrolment, delivery, 6 months, and 5 years and by both mothers and offspring at 14 and 21 years (Additional files 1, 2, 3, 4, 5, 6, and 7). Numbers vary depending on the follow-up stage and sleep items.

The study was approved by the Human Research Ethics Committees of the Mater Hospital and The University of Queensland, with written informed consent being obtained from the mother at each stage of follow-up and from the youth at 21 years.

Maternal smoking pattern over the previous week was examined in maternal questionnaires at the first clinic visit (FCV) (average 18 weeks gestation), 3–5 days after delivery, 6 months, 5 years and 14 years. Mothers were asked to record whether they smoked (yes or no) and if yes, their frequency and quantity of smoking in the previous week. Mothers were also asked at the FCV about whether and how much they smoked before they became pregnant. At 3–5 days after delivery, mothers were asked to recall their smoking level during the last trimester.

Information on maternal smoking was gathered in circumstances designed to maximize the accuracy of the data. This involved interviews conducted within a clinical setting, assurances of confidentiality, detailed questions, and trained interviewers. Reports do vary as to the accuracy of self-reported smoking among pregnant women. Significant agreement between self-reported smoking and serum cotinine levels (the major metabolite of nicotine) has been found [29, 30]. However, varying levels of under-reporting of smoking by pregnant women have been noted [31, 32] and it is important to acknowledge the importance of the setting in which information is collected. This is highlighted by Carabello and colleagues [33] whose findings from a population-based survey of adults regarding their smoking status attest to the accuracy of self-reported smoking status if collected in a private medial setting. Pickett and colleagues [34] illustrate the complex nature of this issue, as their prospective research of pregnant women involved repeated measures of both self-reported smoking status and that based on cotinine levels. They concluded that in epidemiological studies where the intensity and timing of exposure is of particular interest, self-reported smoking status provides a valid measure of fetal exposure. The information was collected in the early 1980’s when the prevalence and extent of smoking were higher and the issue was less prominent as a public health concern. Any potential bias is likely to lead to underreporting of cigarette use, with reduction of effect size.

The category of prenatal smoking included mothers who reported smoking in early or late pregnancy and other times. Maternal smoking status was categorised as ‘never smoked at any stage of the study’ and ‘smoked during pregnancy’ (i.e. smoked in either early or late pregnancy and other times).

The final category consisted of women who smoked postnatally but not during or before pregnancy (women who responded “no” to smoking before pregnancy, at FCV and third trimester but “yes” at 6-months, 5 years or 14 years). The offspring whose mothers smoked during pregnancy and other times were exposed to the effects of maternal smoking whilst in utero as well as the effects of maternal smoking during childhood whereas the offspring of mothers who smoked postnatally but not during or before pregnancy were exposed to passive maternal smoking in infancy/childhood only. This enabled examination of the effects of in utero exposure to smoking over and above the effects of exposure to maternal smoking during early post-natal and childhood development. These categories are mutually exclusive. Women who smoked only before pregnancy were excluded.

At 5 years, mothers were asked how often their child had experienced trouble sleeping over the previous year, rated as often, sometimes, and never/rarely. At 14 years, mothers completed the Child Behaviour Checklist and the youth completed the Youth Self Report [35]. These assessments were chosen because they cover broad domains of mental health and child behavioural issues, they have established reliability and validity [35‐37] and they have been widely used across many countries. The YSR is a standardized self-report questionnaire for adolescents aged from 11 to 18 years. The CBCL is a maternal report questionnaire that assesses the same behavioural subscales as the YSR [38]. Both scales contain the same five common items examining different aspects of sleep initiation and maintenance, parasomnias, and daytime tiredness over the previous 6 months. These five items were “trouble sleeping”, “sleeps less than most kids”, “nightmares”, “sleeps more than most kids”, and “overtired”, with each item rated as often, sometimes, and rarely/never. “Talks or walks in sleep” was an additional question in the maternal CBCL. Mothers were also asked if their child had snored over the previous year. At 21 years many, though not all, of the items of the Pittsburgh sleep inventory [39] were administered to the young adult, as well as additional sleep questionnaire items. Four questions examined difficulties in initiating or maintaining sleep (trouble sleeping, sleep quality, restless sleep and night waking) and three questions examined daytime somnolence (trouble staying awake, day time drowsiness and overtired). Young adults also reported nightmares and snoring. Presence of problems was rated over the previous month.

We created composite indicators of sleep problems at 14 and 21 years, separately for maternal reported sleep items, youth reported sleep items and young adult reported sleep items. For sleep items (e.g. snoring) with a response option of ‘yes’, a score of ‘1’ was assigned, and ‘0’ for a response of ‘no’. Other items, with response options in three categories, were assigned ‘0’ for ‘rarely/never’ or equivalent response, ‘1’ for ‘sometimes’ or equivalent response, and ‘2’ otherwise. All the items were summed after this scoring. The overall score was grouped into three categories as follows: ‘lower 20%’, ‘middle 20–80%’, and upper 20%, with upper 20% having the highest number of sleep symptoms.

Two groups of factors were examined: These were (i) other pregnancy lifestyle exposures and (ii) social and maternal factors. (i) Other pregnancy exposures examined were alcohol, tea and coffee. At the first pregnancy visit and within 1–3 days of birth, questionnaire items examined amount and frequency of alcohol intake (classified as nil, <one glass a day and > one glass a day). Few mothers drank heavily. Coffee and tea each had four categories of none, 1, 2–3 or ≥ 4 cups per day. Analyses were undertaken combining tea and coffee consumption and also examining their effect on offspring sleep separately. Only 2% of mothers reported using marijuana in pregnancy and analysing the data with and without this group produced no differences in the results. (ii) Social and demographic factors were maternal age and level of education, and family income at the FCV. The cut-off for low income levels was the lowest approximate third, depending upon the distribution. At the first antenatal visit mothers were asked four questions concerning whether the pregnancy was planned or wanted and classified as planned/wanted, unplanned/unwanted. Duration of breastfeeding was reported at the six-month follow-up and classified as nil, < 4 months or > 4 months.

The relationship between the maternal smoking variable (3 categories) and the 22 sleep measures at different follow-up phases (refer Table 2) was initially examined, with the chi squared test being administered for statistical significance. As each test reflected the initial study hypothesis, a two tailed P-value of < 0.05 was taken to indicate initial statistical significance. However, due to multiple statistical comparisons, the effect of applying the Bonferroni correction was also examined. The relationship between maternal smoking and the eight sleep questions significant in this initial analysis is described in Table 2. To increase the statistical precision, for some outcomes we combined ‘sometimes’ and ‘often’ responses as one category (see Table 3) for the multivariable analyses. A sensitivity analysis was conducted to examine these results further. Of women who smoked cigarettes in pregnancy, more than half of them reported smoking ≥10 cigarettes daily. No dose-response relationship was found. Additionally, as some of the cell frequencies were small, especially for the postnatal only smoking group, in the adjusted analyses we considered coffee, tea, alcohol in pregnancy, maternal age and education. Combining tea and coffee or considering them separately did not have any impact on the results. In the sensitivity analysis, we added other confounding factors and found the effect size remained consistent. An adjusted analysis was performed using multinomial logistic regression examining sequentially the effects of other pregnancy lifestyle exposures, social factors and maternal factors, with all factors included in a final fully adjusted model. In these multinomial logistic models, the groups of prenatal exposure, and postnatal exposure, were contrasted to never smoking mothers as the reference category.

Table 1 describes the total birth cohort and the study group at each stage of follow-up. Mothers of children lost to follow-up were more likely to be younger, less well educated, more financially disadvantaged and the children more likely to be of birthweight < 2500 g or gestation < 37 weeks. Of 3954 women who prospectively reported their smoking status from pregnancy to 14 years follow-up, 55% never smoked, 41% smoked at some stage of pregnancy and other time and 4% smoked after pregnancy but not during or before pregnancy.

Table 2 provides an overview of the 22 sleep questions ordered according to their follow-up phase, their distribution, and the level of statistical significance for their overall association with maternal smoking status during and after pregnancy. Eight of the sleep measures were initially significant at P < 0.05, five of these at 14 years and three at 21 years. At 14 years the significant associations were: mother reports of talking or walking in sleep (P < 0.001), sleep more than other kids (P = 0.044) and snoring over the last year (P = 0.001). Youth-reported nightmares at 14 years was associated with maternal smoking in utero (P = 0.01). At 21 years, young adult-reported nightmares (P = 0.01) and being restless and trouble staying awake p (P = 0.017) were significant. There was no association between troubles sleeping at five years and maternal smoking. When the Bonferroni test was applied to the initial 24 comparisons, walking and talking in sleep retained significance. Each of the eight overall sleep measures was further examined for an interaction between smoking and gender and no interaction terms were significant.

In Table 3, the strength of relationship between three mutually exclusive categories of maternal smoking status (no smoking as the reference category) and the eight sleeping outcomes are shown adjusted for other pregnancy exposures, and maternal and social factors. For maternal smoking during pregnancy and other times, mother-reported talks/walks in sleep and youth-reported nightmares were more likely (OR and 95% CI: 1.23, 1.04–1.46 and 1.23, 1.03–1.46 respectively) whereas problems staying awake at 21 years was less likely (OR and 95% CI: 0.80, 0.65–0.98). For mothers who smoked postnatally but not before or during pregnancy, maternal reported offspring snoring at 14 years was more likely (OR and 95% CI: 1.53, 1.06–2.23). The composite sleep variables in Table 4 show that the top 20% and middle 20–80% of maternal reports of sleep problems at 14 years were more likely in the offspring of both maternal smoking groups in the adjusted analysis (OR and 95% CI: 1.26, 1.05–1.50 and 1.36, 1.05, 1.76, respectively). For youth-reported composite sleep problems at 14 years, the only significant association in the adjusted analysis is with smoking in pregnancy and at other times (OR and 95% CI: 1.29, 1.02–1.64). At 21 years, no associations are significant in the adjusted analysis.

Several features of the current study make the results noteworthy. This is the first longitudinal study to examine sleep problems associated with prenatal maternal smoking and postnatal smoking across several developmental periods in a large community sample. This study is best able to examine consequences of prenatal smoking on sleep in adolescence and young adulthood and control for exposure to parental smoking during childhood. The only previous longitudinal research examined outcomes up to 12 years in a very different group – a high risk cohort with pregnancy exposures [27]. The analyses adjusted for a number of psychosocial characteristics that could potentially confound any relationship between maternal smoking and offspring sleep. The strategy for data analysis used in our study also has advantages for exposures such as cigarette use that are strongly correlated over time. Although we could examine the effect of maternal smoking during pregnancy on offspring sleep separately from smoking at other times, a limitation of this study is that level of smoking could not be evaluated using this methodology so a threshold effect with heavy prenatal smoking cannot be excluded. A further limitation is the lack of information on exposure to other sources of passive smoking including smoking by fathers. The exposure to second hand smoking by other people in the household may impact on the respiratory systems of the offspring resulting in increased risk of parasomnias and snoring. Smoking by others in the home, however, is likely to be correlated with maternal smoking and disentangling what is a direct effect of maternal smoking and a result of passive smoking was not possible in this study. While maternal smoking was measured only with self-reports, these involved detailed questions asked by trained interviewers in a confidential, clinical setting. There is therefore a possibility of a social desirability bias leading to under-reporting of smoking status.

Other limitations include the use of questionnaire measures to assess sleep and the level of attrition. We have previously assessed the impact of attrition using multiple imputation and sensitivity analyses, with little impact on the findings [28]. Comparisons to other cohort studies have also yielded similar results [46]. Attrition was more likely in those offspring whose mothers had adverse social circumstances and adjustment for these in our analysis made little difference to the findings. Moreover, the present results would only be biased if the associations were in the opposite direction in participants not assessed at each phase of the research which would be unlikely.