Maternal Smoking in Pregnancy and Child’s Hospital Use up to 5 Years of Age in a Data Linkage Birth Cohort

The Tasmanian C2C study is a linked perinatal data set for all Tasmanian births, public hospital admissions, and ED data from July 2008 to June 2014, excluding children whose mothers resided outside of Tasmania. Tasmania is an island state of ∼500 000 people and has a relatively decentralized population, with 65% residing in inner regional areas, 33% residing in outer regional areas, and 2% residing in remote and very remote areas.¹⁴  Prevalence of maternal smoking during pregnancy was 14.5% (2017) in Tasmania.¹⁵  The data were linked by the Tasmanian Data Linkage Unit under best practice protocols involving the separation principle, which is the segregation of identifying information from clinical data, storage in a stand-alone server with no Internet connectivity, and encryption of stored data.¹⁶  This study was approved by the Tasmanian Human Research Ethics Committee (H0017161).

Maternal smoking during pregnancy data were self-reported as part of the National Perinatal Minimum Data Set during antenatal visits with midwives. Data are submitted within 30 days of the birth. From July 2008 to June 2010, mother’s smoking during pregnancy was recorded as “none,” “<10 cigarettes per day,” and “≥10 cigarettes per day.” From July 2010 onward, mother’s smoking during pregnancy was recorded as “none,” “yes, <20 weeks” or “yes, >20 weeks.” For our primary analysis, we categorized as “yes” if participant’s mother ever smoked during pregnancy and “no” if participant’s mother did not smoke during pregnancy.

Outcomes were presentation to any Tasmanian public hospital ED and hospital admission through the ED. We classified each child as having none or any ED presentations and none or any hospital admissions. We calculated age at first presentation and the total length of stay for hospital admissions. The following were not admissions: dead on arrival, died in the ED, departed in police care, did not wait to be seen by a medical officer, triage only with alternative medical assessment recommended, left at own risk, and transferred to another hospital.

Each ED presentation or hospital admission through the ED was grouped into 9 disease categories on the basis of regrouping of the 28 Urgency-Related Groups classification system (Supplemental Information): poisoning and injuries; respiratory system; digestive system; neuromuscular system; blood and immune system; eyes, ears, nose, and throat; reproductive and urologic system; systemic and parasitic infections; and psychosocial and other illness presentation.

We considered the following as potential covariates: maternal marital status at birth (single, married, or other [separated, divorced, or widowed]), maternal alcohol intake during pregnancy (yes or no), maternal age at birth, maternal ethnicity (Aboriginal and Torres Strait Islander or other), infant sex (male or female), gestational age (<37 weeks or ≥37 weeks), Apgar score at 5 minutes (<7 or ≥7), mother’s region of residence at birth (noting regional variation in ED presentation in Tasmania¹⁷ ), birth weight (<2500 g or ≥2500 g), year of birth, and Index of Relative Socioeconomic Disadvantage (IRSD) score based on the residential postcode of the mother (quartiles of IRSD scores).

Descriptive statistics were used to summarize the characteristics of the children, number of ED presentations, total admissions to the hospital, age at first presentation, and total length of stay for each hospital admission. χ² or t tests were used to examine characteristics of children by exposure to maternal smoking during pregnancy.

Maternal smoking observations were missing for 2382 (7.8%) children in the cohort who were managed up to ≤1 year of age and 33 (0.52%) children in the cohort who were managed up ≤5 years. These observations were missing at random because of a system issue (in the implementation of an electronic data set), which has since been rectified. Multiple imputation by chained equations was used to replace missing data on maternal smoking during pregnancy. The imputation models used the same variables as the analytic models, plus auxiliary variables associated with either missingness: year of birth, birth weight, IRSD score, maternal age at birth, gestational age, private or public patient status of the mother, electronic or paper hospital record, hospital name, and counts of ED presentations and hospital admissions for each diagnostic category plus total counts of ED presentations and hospital admissions. Ten imputations were performed, and the imputed data set was used for the main analysis.

Negative binomial regression and logistic regression were used to estimate the association between children’s rates of ED presentations and hospital admissions and exposure to maternal smoking during pregnancy (reference: not exposed). The incidence rate ratio (IRR) (with the 95% confidence interval [CI]) is reported for (1) the count of presentations to the ED for any condition, (2) the count of presentations to the ED within each of the 9 major disease categories, (3) the count of hospital admissions after ED presentation for any condition, and (4) the count of hospital admissions after ED presentation for each of the 9 disease categories. Logistic regression was used for some outcomes (reclassified as any presentation versus none) because of nonconvergence of the negative binomial model. Final models included the sex of the child, socioeconomic position (IRSD score), the mother’s region of residence at birth, and maternal age at birth (adjusted model 1). Early childhood exposure to tobacco smoke has been associated with group-level socioeconomic position and maternal age in a population-based birth cohort and so could account for tobacco smoke exposure after delivery.¹⁸ 

Limited availability of potential confounding factors may result in residual confounding in administrative data sets. Negative control outcomes may overcome this limitation.¹⁹  Hospital presentations and admissions for poisoning and injuries were considered negative control outcomes because association seen with these outcomes were not likely to be causally related to exposure. Children aged <5 years are susceptible to poisoning and injuries because of agility, an exploratory nature, a lack of understanding of safety, and a propensity to mouth objects.²⁰  These factors are developmental and are not causally associated with maternal smoking during pregnancy. Any association seen might therefore reflect other potentially confounding factors related to health service use.^19,21  Effect estimates (including 95% CIs) for other disease categories relative to estimates for poisoning and injuries allow us to evaluate less confounded associations between maternal smoking and the outcomes.

Follow-up data on health services use are also available from July 2008 to June 2014. As such, complete data for children up to the age of 5 years are only available for the subsample of children who were born in 2008–2009. The analyses described above were therefore completed for 2 overlapping groups of children: (1) those ≤1 year of age and (2) those ≤5 years of age.

Sensitivity analyses to determine a dose response of maternal smoking (options: none, <10 cigarettes per day, and ≥10 cigarettes per day) were done by using children born from July 2008 to June 2010 (results in the Supplemental Information). Maternal smoking during pregnancy is strongly associated with low birth weight,²²  which, in turn, is associated with poor childhood health.^8,23  Birth weight was therefore considered as a potential mediator in sensitivity analyses (Supplemental Information). Data were analyzed by using Stata version 15 (Stata Corp, College Station, TX).

The C2C study included 27 532 mothers and 36 630 infants (Fig 1, Table 1). There were 30 705 children in the data set including presentations or admissions at ≤1 year age and 6345 children in the data set including presentations or admissions at ≤5 years. Total numbers of presentations and admissions are presented (Supplemental Table 4).

A greater proportion of children exposed to maternal smoking during pregnancy presented to the ED overall and for specific illnesses (respiratory; digestive system; neuromuscular system; blood and immune system; eyes, ears, nose, and throat; reproductive and urologic system; systemic and parasitic infections; and psychosocial and other illnesses) compared with children not exposed (Supplemental Table 5). Children exposed to maternal smoking during pregnancy had a higher admission rate through the ED than those not exposed. Among those admitted to the hospital, length of stay in the hospital was similar between exposure groups.

There was a 52% and 45% higher incidence rate of presentations of any cause to the ED at ≤1 year and ≤5 years, respectively, in children exposed to maternal smoking during pregnancy (Table 2). This decreased to 26% at both ≤1 year (IRR_adjusted 1.26; 95% CI 1.24–1.28) and ≤5 years (IRR_adjusted 1.26; 95% CI 1.23–1.29) after adjustment.

Children exposed to maternal smoking during pregnancy had a higher incidence of presentations to the ED at ≤1 year and ≤5 years for respiratory system; eyes, ears, nose, and throat; systemic and parasitic infections; and psychosocial conditions. The incidence rates for these illnesses at 1- and 5 years were greater than the incidence rate for the negative control outcome of poisoning and injuries. For digestive system and neuromuscular illnesses, the rate of presenting was higher in exposed children at ≤5 years only (Table 2).

In sensitivity analyses, it was suggested that birth weight explained some of the association between maternal smoking during pregnancy and ED presentations and hospital admissions. When controlling for birth weight, the magnitude of the IRR at 1 and 5 years was reduced for respiratory illnesses, systemic and parasitic infections, and psychosocial and other illnesses (Supplemental Table 6). After adjustment, all associations remained greater than those for injuries and poisoning. There were clear dose-response relationships between the number of cigarettes smoked per day (Supplemental Tables 8 and 10) and all rates of presentations to the ED for respiratory illnesses; eyes, ears, nose, and throat illnesses; systemic and parasitic infections; and psychosocial and other illnesses. For each of these outcomes, the incidence rate was greater with more exposure (>10 cigarettes per day), including for digestive and neuromuscular conditions, which were shown to have the opposite association with any level of exposure. In comparison, there was no evidence of a dose-response association with poisoning and injuries.

There was a 57% and 65% higher incidence of admission to the hospital through the ED for any disease category for children exposed to maternal smoking during pregnancy (Table 3). This reduced to 33% and 45% with account for covariates. In terms of specific illnesses, children exposed to maternal smoking during pregnancy had a higher rate of being admitted for respiratory illnesses at ≤1 year (62%) or ≤5 years (59%), for systemic and parasitic infections at ≤5 years, for reproductive and urologic system illnesses at ≤5 years, and for psychosocial and other illnesses for both age ranges, than for poisoning and injuries.

In sensitivity analyses, there was some effect on the association between maternal smoking and admission to the hospital with the inclusion of birth weight (Supplemental Table 7). The associations were reduced for respiratory illnesses (38.7% at ≤1 year; 30.5% at ≤5 years), reproductive and urologic system (40.7% at ≤5 years) illnesses, systemic and parasitic infections (17.3% at ≤5 years), and psychosocial and other illnesses (47.9% at ≤1 year; 50.7% at ≤5 years). For the subsample with information on the dose of maternal smoking during pregnancy from 2008 to 2010, there was evidence that the associations were dose dependent (Supplemental Tables 9 and 11) for all admissions through the ED; admissions for respiratory illnesses, systemic and parasitic infections, psychosocial and other illnesses, blood and immune system disorders, and eyes, ears, nose, and throat conditions were shown to have opposite associations in the analyses with any exposure. Importantly, there was no evidence of a dose-response relationship between maternal smoking during pregnancy and poisoning and injuries.

Children exposed to maternal smoking during pregnancy presented more frequently to the ED and had a higher rate of hospital admission through the ED. Respiratory illnesses; eyes, ears, nose, and throat illnesses; systemic and parasitic infections; and psychosocial and other presentations were the major categories associated with hospital use in children exposed to smoking during pregnancy.

We have uncovered a previously hidden burden of ED presentations associated with exposure to smoking during pregnancy in children involving nonrespiratory illnesses. Our findings support earlier studies that found an increased risk of ED presentation and hospital admission for respiratory and any illness at <2 years of age due to exposure to tobacco smoke in utero.^24,25  We observed an ongoing importance of exposure to smoking during pregnancy to an increased rate of hospital use at up to 5 years of age. We also identified higher hospital admission rates for respiratory illnesses, psychosocial presentations, and infections among those exposed to maternal smoking during pregnancy. By implication, these higher admission rates could portend that exposed children experience more severe illness.²⁶  Because of the limitations of the administrative data sets used for data linkage, residual confounding by social and family factors is possible. We used a nonbiologically plausible association with poisoning and injuries as a negative control outcome, but we acknowledge the limitations of this approach.¹⁹  Our results are in line with findings on hospitalization for respiratory illnesses in children exposed to tobacco smoke either in childhood or during the pregnancy.^5,6,27–29  On the basis of our findings, children exposed to maternal smoking during pregnancy will require use of more health care resources because of more frequent ED presentations and hospital admissions, adding significant costs to the health care system.

Few studies have quantified the health care costs. In Hong Kong, illnesses at 1 year in children exposed to smoking in utero were estimated to result in an additional 1581 hospital visits, costing >$2.1 million US dollars to the health care system.²⁴  In the United Kingdom, the total average health care (primary care consultation, diagnostic testing, prescriptions, and secondary care) cost difference between children exposed to maternal smoking during pregnancy and unexposed children was £91.18 at 1 year of age and £221.80 at 5 years of age.³⁰  Applied to our statewide numbers, in utero exposure to smoking would equate to >£1.5 million in preventable direct health care costs at up to 5 years of age. More detailed modeling that includes both direct (eg, health care) and indirect (eg, missed school attendance) costs is warranted.

There are causal pathways linking exposure to maternal smoking during pregnancy and the conditions resulting in ED presentations or admissions to the hospital. Immune system functioning could be a unifying explanation, with evidence of direct effects of maternal smoking during pregnancy on various epigenetic pathways in children leading to suppression of immune system functioning.^31,32  Low birth weight and preterm birth are also causally related to exposure to smoking during pregnancy and associated with an increased rate of lower respiratory infections and viral gastroenteritis,^8,22,23  thereby making low birth weight a potential mediator of the associations. Inclusion of birth weight in analyses did not completely remove the effects of the exposure, suggesting that direct effects of exposure are also likely.

Continued exposure post partum to maternal smoking may partly account for the higher ED presentations and admissions to the hospital.^26,33  In few studies have researchers examined the contribution of exposure to smoke during pregnancy versus postnatally. In one study, increased hospitalizations in children exposed in utero remained, even after adjusting for exposure during childhood, suggesting that exposure during pregnancy has independent effects.⁹  Other investigators have shown that pregnant women who smoke during pregnancy often continue post partum,³⁴  and many women who give up smoking during pregnancy relapse post partum.³⁴  Mothers who smoke are also more likely to have partners who smoke.³⁵  Smoking inside the home has decreased significantly over the past few decades,^36,37  but there is increasing recognition of the adverse effects of thirdhand smoke.³⁸  This may have an impact on children’s health despite no direct exposure to sidestream smoke, even after parents stop smoking, given the increased cancer risk in nonsmokers exposed to nitrosamines in thirdhand smoke.³⁹  But in utero exposure likely has a direct effect on offspring independent of exposure after birth (leading to illness^40–42  associated with ED presentations and hospital admissions) as well as an indirect effect through continued exposure to passive smoking after birth.

Reducing in utero exposure can be achieved by preventing smoking during reproductive years, encouraging quitting during pregnancy, and avoiding relapse to smoking during the postpartum period. Guidelines suggest that pregnant women should be aided to quit smoking through counseling, advice, and support, with oral forms of nicotine replacement therapy used if these earlier measures fail but the pregnant woman is still motivated to quit.⁴³  During the period of this study, the state of Tasmania implemented tobacco control policies that may impact smoking during pregnancy, including a ban on smoking in vehicles with children present, increased funding for smoke-free messages, training for health professionals in brief interventions, and the establishment of specialized smoking cessation clinics in public hospitals.⁴⁴  In addition, pregnant women were designated as a priority population group, with special focus on interventions at the population and individual levels.⁴⁵  Our findings indicate the importance of continued advocacy for funding population-level interventions to reduce smoking in men and women of childbearing age and to enhance quitting within households during pregnancy.

We could not discriminate between children exposed to maternal smoking during pregnancy and those who continue to be exposed during the postnatal period. We acknowledge that part of the increase in ED presentations and hospitalizations could have been due to sustained exposure in childhood. We did not have access to individual-level measures of socioeconomic position. However, area-level measures of socioeconomic position are highly correlated with individual-level measures of socioeconomic position, with correlation coefficients ranging from 0.85 to 0.96.⁴⁶  For some conditions, including neuromuscular and blood and immune system disorders, the numbers of presentations and admissions was low, meaning that there may be less certainty in these estimates.

The strengths of this study include statewide record linkage for public hospital admissions and ED presentations in Tasmania, making it widely generalizable, particularly in Australia or places with similar sociodemographic characteristics. We considered all possible diseases rather than only respiratory illnesses, which have been the focus of other studies, and we used a negative control to account for residual confounding.

Maternal smoking during pregnancy is associated with more frequent visits to the ED and admissions to hospital through the ED in exposed children. Health service use is particularly due to respiratory and infectious diseases, which are linked pathologically to passive smoke exposure during fetal development. Further efforts should be directed toward decreasing smoking among people of reproductive age and increasing smoking cessation among pregnant women and their partners.

The members of the original C2C study group included A/Prof Neil and Prof Judd, Prof K. Sanderson, and Dr Wagg.

For this research, we used data owned by the Tasmanian Department of Health. We thank B. Stokes and N. Wiggins from the Tasmanian Data Linkage Unit in progressing establishment of the C2C Database. A/Prof Neil is supported by a Select Foundation Research Fellowship. A/Prof Magnussen is supported by a National Heart Foundation of Australia Future Leader Fellowship (100849). A/Prof Gall is supported by the National Heart Foundation of Australia Future Leader Fellowship (102061 and 100446).