Background

A total of 350 UK and 1100 US stillbirths each year are attributable to smoking during pregnancy1 as are 200 UK and 900 US infant deaths.2–4 Maternal smoking also increases a child’s chance of being born prematurely with low birth weight,5 and having asthma,6 attention deficit disorder7 and learning difficulties,8 which if causally related all add substantial costs to healthcare.9 Pregnancy is also an opportunity to help most women quit smoking permanently before their own health is irreversibly compromised. Cessation during pregnancy if permanent reduces the likelihood of offspring becoming smokers,10 stopping mothers by example passing on their smoking habit to their children.

A fall in reported smoking during pregnancy has been reported in Scotland11 (20.8% 2007, 15.5% 2015). In 2000, specific cessation services12 for pregnant women were established, supported by a National Institute for Health and Care Excellence guideline.13 However, self-reported smoking during and after pregnancy remains common. Twenty-five per cent of biochemically verified smokers self-report as non-smokers,14 and this proportion may be increasing as pressure to quit intensifies. Only 10% of pregnant smokers use cessation services in Scotland and as few as 3% quit during pregnancy.15 A recent report from the UK is more encouraging with on average 50% cessation during pregnancy.16

Objectives

We used a large linked Scottish dataset17 to produce contemporary estimates of the impact of maternal smoking measured in early pregnancy on pregnancy, infant and early child health outcomes. We focused on infant deaths and admissions to hospital for a range of conditions to assess the proportion of events attributable to smoking during pregnancy.

Study design, setting, participants, data sources

This was a retrospective cohort study of singleton births in Scotland between 1997 and 2009 using anonymised extracts of linked administrative data provided by the Information Services Division, National Health Service National Services Scotland. Approval for the project design and confidentiality of patient data was provided by the Privacy Advisory Committee of NHS National Services Scotland, a body set up to ensure the appropriate use of patient identifiable information.18 Further ethical permission was not required. All 731 595 live birth records in Scotland between the calendar years 1997 and 2009 inclusive were linked in two phases by probabilistic matching techniques including the use of the Community Health Index (CHI), a unique identifier developed for health records in Scotland.19 Phase 1 comprised linkage of births, deaths, migration, maternity, infant health and child health surveillance review records,20 which was extended in phase 2 to include all hospital admissions (ie, emergency admission and elective) from birth until death or up to March 2012. Thus, follow-up information on each child in the cohort was available for at least 2.25 years (for all children) and up to 15 years (depending on their birth year). The original cohort of 731 595 births was reduced to 697 003 as a result of excluding multiple births, babies of non-Scottish residents and duplicate linkage records.

Outcome variables and bias

The outcomes examined were neonatal and postneonatal mortality, pregnancy outcomes (infant small for gestational age and premature births) and hospital discharge up to the fifth birthday with a primary diagnosis from a list of conditions selected from a review of literature. Conditions of interest were asthma, bronchiolitis, bacterial meningitis, acute upper and lower respiratory infections, acute and chronic otitis media and mastoiditis and bronchopulmonary dysplasia. Emergency and planned admissions (with day cases) were included. Admissions for long bone fracture, not thought to be associated with maternal smoking, were used as a control outcome to assess if confounding had been satisfactorily controlled in the multivariate analysis.

Definition of smoking during pregnancy

Smoking behaviour in pregnancy is collected at a woman’s first antenatal booking appointment which usually takes place within the first 3 months of pregnancy. These booking appointments take place either at hospital or in the community and are recorded on the Scottish Woman-Held Maternity Record. The smoking at booking information is subsequently transcribed onto the woman’s Scottish Morbidity Record 02 which is submitted by the maternity units following delivery. The smoking behaviour is classified into either ‘never smoked’, ‘current smoker’, ‘former smoker’ or ‘unknown/missing’.

Statistical methods and quantitative variables

Descriptive and univariate analyses were used to identify variables associated with smoking during pregnancy and perinatal, mortality and hospitalisation outcomes. Cox regression was used to quantify the independent contribution of maternal smoking to hospitalisation outcomes, based on survival from birth to the earliest of first hospitalisation, death, migration, the fifth birthday or the end of the observation period (March 2012). Logistic regression was used for perinatal and mortality outcomes. The models included variables identified from directed acyclic graphs as potentially biassing (confounding) the association between smoking and the relevant outcomes. Adjustment was made for maternal age, maternal and paternal socioeconomic status, infant feeding at 6–8 weeks, country of birth, infant sex, parity, mode of delivery and quintile of the Scottish Index of Multiple Deprivation (SIMD). Each Cox regression model was tested for proportionality using inspection of log–log plots. Additional analysis was conducted to assess the risk of hospitalisation for long bone fractures (a condition not causally associated with maternal smoking), to test the adequacy of adjustment for socioeconomic and demographic confounders included in the linked dataset. Statistical modelling was carried out using Stata V.11; StataCorp LP.

Population attributable fractions (PAFs) were used to quantify the number of outcome events that would have been avoided if no women had smoked, using the formula:

where RR is the relative risk and P_c is the proportion of cases exposed to the risk factor.

Patient or public involvement

No patient or public involvement.

Participants

Of the 697 003 singletons born between 1997 and 2009 included in the analysis (table 1), 63% were born by spontaneous vaginal delivery, 8% had teenage mothers, 16% were born to single parents or parents living apart, 44% were born to first-time mothers and 25% were residents in the most socioeconomically deprived quintile. Twenty-three per cent of infants had mothers who smoked (table 1). A total of 332 386 (48%) had at least one hospital admission by 31 March 2012 (figure 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Admissions to hospital for children born in Scotland between 1997 and 2009.

Descriptive data

As previously reported,17 there was a greater relative risk of hospital admission among infants resident in more deprived areas, of fathers and mothers with a semiroutine/routine occupation, of babies bottle fed or combination fed (bottle and breast), among infants with siblings, and those born by Caesarean delivery. Conversely, risk of infant admission decreased with increasing maternal age, among female infants and if a mother’s country of birth was non-British.17 During the study period, 87 877 (12.6%) infants were hospitalised at least once for one of the selected conditions in the first 5 years (figure 1). At the first recorded hospital event for any of the selected conditions, 43% were under 1 year of age, and 57% were between first and fifth birthday. Most of these first hospital events were ‘emergency admissions’ (91%), especially among infants less than 1 year of age at admission (99%).

Younger unmarried pregnant women and women from deprived postcode areas and socioeconomic circumstances were more likely to be current smokers, as were British born compared with non-British born pregnant women (table 1). Smoking varied little by mode of delivery or number of previous children, however, babies who were bottle fed, or whose mother’s religion was Christian or Jewish were more likely to have mothers who were current smokers at maternity booking (table 1).

Outcome data

The crude rate of hospitalisation (any cause) per 1000 person-years was 141 among infants born to mothers who smoked, 108 for those who did not smoke, 119 for former smokers and 109 for those with missing smoking status (data not shown in tables or figure).

Main results

Other analyses

Hospital admissions related to long bone fractures were analysed as a control group. There were 5757 children admitted for long bone fractures from the cohort during the study period. After adjustment for other factors (maternal age, maternal and paternal socioeconomic status, infant feeding at 6–8 weeks, country of birth, infant sex, parity, mode of delivery and quintile of the SIMD), a mother being a smoker at maternity booking was not a significant predictor of hospital admission for long bone fractures.

Key results

Strengths and limitations

Other similar studies generally only looked at one or two outcomes,31 for example, birth outcomes (small for gestational age and stillbirth) usually derived from a single dataset.32 It is the linkage of routinely collected datasets that sets this study apart, allowing, in this case, hospitalisation outcomes to be gathered over time. This is possible in Scotland because the excellent routine data collection systems allow data linkage between separate datasets on all infants and their mothers.19 Data linkage will only be improved by widespread use of the CHI, a unique identifier (https://www.ndc.scot.nhs.uk/Dictionary-A-Z/Definitions/index.asp?ID=128&Title=CHI), now given to every baby in Scotland which can be linked to the mothers’ CHI number. Future linkage studies will provide a powerful tool with many research and surveillance applications to support improved healthcare and surveillance for the population of Scotland.

In addition, in our study, we modelled hospitalisations for long bone fractures—not causally associated with maternal smoking—to act as a control outcome. This analysis suggests that the adjustment for socioeconomic confounders was sufficient.

Smoking behaviour in pregnancy was collected as self-reported data at a woman’s first antenatal booking appointment which usually takes place within the first 3 months of pregnancy. We know that 25% of current smokers at maternity booking do not admit to their smoking habit.14 We also know that: ‘Although many women quit smoking during their pregnancy, relapse rates are high and most start smoking again within 6 months of giving birth’.13 Both these issues would suggest that true smokers were in the non-smoker category in our analysis. This will have reduced our overall estimates of PAF. Therefore, the figures we have calculated are likely to be an underestimate of real effects of maternal smoking on outcomes. This study only looked at babies who were born alive and therefore does not include pregnancy outcomes leading to fetal death, that is, early miscarriage or stillbirth, both of which are devastating and are associated with maternal smoking.1 3 33 34 This study focussed on hospitalisation of infants and children to 5 years of age, and therefore does not include older child illnesses or those only requiring support from primary care. This may explain the low PAF for otitis media, as children are treated in primary care and only rarely admitted to hospital. It also excluded short-term and long-term consequences of smoking for the mother herself where continued smoking will shorten life by on average 10 years.35 Maternal smoking information was recorded as unknown or missing for 12% of individuals in the cohort. Pregnancy and mortality outcomes were significantly worse within this group, suggesting that perhaps unrecorded smoking or another unknown risk factor was important.

Although the coverage and completeness of variables using the routine datasets was relatively high, the study was limited by the availability of confounders on the linked dataset and uncertainty over the overall duration of maternal smoking. We do, however, know that few women who self-report as smokers at maternity booking quit during or probably after pregnancy.15 In addition, an important limitation of this study is that there was a high level of missing data for some of the variables used, particularly infant feeding method and maternal occupation. This is likely to relate to the phased introduction of the Child Health Surveillance Programme across Scotland during the study period which has been recognised for infant feeding data. The impact of missing data on maternal occupation may not have been significant because of the inclusion of the two other measures of socioeconomic level in the regression models: area deprivation and paternal occupation.

Interpretation

Overall, it is likely that there is an underestimation of the association between maternal smoking and poor outcomes including hospitalisation as 25% of pregnant smokers deny their habit when asked at maternity booking.14 This level of under-reporting may increase the attributable fraction for each condition by a third, making, for example, the proportion of small for gestational age babies attributable to maternal smoking 36% rather than 27.5%. Alternatively, if those with missing smoking data were all smokers, the estimates we present would be reduced. However, our analysis of ‘small for gestational age’ (table 2)—the most well-known effect of maternal smoking during pregnancy—does not suggest that all those with missing smoking data are in fact current smokers. The OR for ‘unknown/missing’ smoking status is 1.31 (1.26, 1.35) and for ‘current smoker’ is 2.67 (2.62, 2.73).

This study adds to the evidence identifying key adverse childhood outcomes of maternal smoking.

Maternal smoking is still making a substantial contribution to poor pregnancy outcomes and morbidity in early childhood.

This study confirms previous findings that maternal smoking is associated with large increases in the risks of being small for gestational age, of postneonatal mortality and of premature delivery, but less evidence of impact on neonatal mortality.

Meningitis, bronchopulmonary dysplasia, asthma, bronchiolitis and acute upper and lower respiratory infections (in the first year of life) make large contributions to the increased risk of hospital admission; ear infections are less prominent, probably because nearly all are treated in primary care, though still a significant risk.

The strength of the association between maternal smoking and increased infant and child morbidity and mortality, consistent with other studies, provides convincing evidence of the benefits of maternal smoking abstinence on child health.

Generalisability

This study presents results from a ‘whole population’ cohort of infants born in Scotland over a 13-year period from 1997 to 2009. Children have been followed through to March 2012. These results are contemporary and are likely to generalise to most other UK countries and other countries with high levels of smoking during pregnancy.