The results of this large population-based study support that smoking during pregnancy is a modifiable dose-dependent risk factor of adverse fetal growth that also has a strong relationship with other risk behaviour and low SES indicators. Compared to all lower levels of smoking, heavy smokers (≥ 10 cigarettes/day) had substantially worse birth outcomes and were also at increased risk to be identified for alcohol use and drug use, be a single parent, attended fewer prenatal care visits and have pre-pregnancy weight greater than 74 kg. Although the addition of a major SES variable, level of education, was limited to only 10% of our study population, the main effects and general trends were corroborated. Heavy smokers were 3.8 times more likely to have not graduated high school compared to moderate, light and non-smokers combined supporting the possibility that reports of smoking greater than ten cigarettes per day might be an early marker for the need for comprehensive supports to reduce adverse outcomes.
The adjusted ORs for the impaired fetal growth outcomes (SGA, IUGR and term-LBW) were nearly twice the magnitude between heavy and light smokers. The addition of the education variable into the logistic models attenuated the effect of light smokers to the degree of no significant difference between light, former and never smokers while the effect of moderate and heavy smoking remained relatively stable with roughly double the risk. The effect of smoking on PTB was completely removed after adjusting for maternal education. This suggests that while behavioural and SES indicator variables, particularly maternal education, explain some or all of the risk attributed to light smoking, heavy smoking remains a robust marker of increased risk for the impaired fetal growth outcomes. Whether this observed effect is strictly biological or is partially a marker for some latent unmeasured risk factor, heavy smoking readily identifies approximately 5% of the BC population who could benefit from additional support services. These results were consistent with findings from a population-based study from Nova Scotia [21
] as well as a prospective cohort study that used anthropometric ultrasound measurements to compare fetal growth in smoking and non-smoking expectant mothers [22
The mechanisms to which cigarette smoke exposure effects fetal growth is not completely understood; however, IUGR correlates with defects in placental transport and metabolism functions which seems to restrict nutrient supply [23
]. Zdravkovic et al. report that constituents in cigarette smoke directly affect placental cytotrophoblast proliferation and differentiation which reduces blood flow and creates a hypoxic environment [24
]. Using a mouse model, Detmar et al. found that polycyclic aromatic hydrocarbons (PAHs), a main component in cigarette smoke, caused IUGR in the fetuses of exposed dams and yielded alterations in placental vascularisation with significantly reduced arterial surface area and volume [25
]. PAHs are also a main constituent of vehicular exhaust, particularly diesel, and there is mounting evidence of an association between said pollutant and growth restricted birth outcomes [26
The results from the pp-odds model show that most of the covariate risk factors primarily predict maternal smoking in general versus non-smokers. Variables such as single parent, drug and alcohol indication and young maternal age were significant across all levels comparison, but had the strongest effects in comparing non-smokers to all other levels of smoking. Conversely, parity exhibited its strongest effects in the third comparison (heavy smokers versus never, light, and moderate smokers combined). This suggests that while being multiparous is a marker for maternal smoking in general, it predicts heavy smoking versus moderate or light smoking habits. A similar observation was found in a study of UK women regarding gravida and smoking behaviour in subsequent pregnancies, commenting on the double exposure of the previous children to cigarette smoke both pre- and postnatally [28
]. While older maternal age was associated with having reported 'never smoked', older mothers who did smoke were more likely to be heavy smokers. This trend of older mothers being heavy smokers was also observed in the Nunavut chart review study [12
The results for the pp-odds model including maternal education generally hold true to the first model. The effects for age, parity, single parent, drug flag, and alcohol flag are slightly attenuated but remain significant with the same trend. Education (no grade 12) had a strong constant effect on maternal smoking across all levels of comparison, suggesting an important role in health literacy. Maternal level of education has been shown to be a powerful determinant of perinatal health, independent of, and stronger than that of neighbourhood income [29
]. Having low maternal educational attainment, being young and a single parent are indicators of low socio-economic status that may exert additional stress on the pregnancy. The biochemical response to stress via elevated basal cortisol levels has been associated with low birth weight [30
]. Three major systems are thought to be involved in the biological pathway linking maternal mental health and stress with adverse birth outcomes which include the neuroendocrine, the immune/inflammatory, and the cardiovascular systems with placental corticotrophin releasing hormone playing a central coordinating role [31
]. Indicators of women's mental health during pregnancy such as psychosocial stress, level of social and financial support and depression may be one possible pathway to which low SES is associated with adverse birth outcomes.
The majority of results from this British Columbia based study were consistent with recent findings from Norway [32
], Germany [33
], and a national Canadian survey that analyzed the associated risk factors of smoking during pregnancy [8
]. The Canadian study found that non-immigrant, single parent, low household income, no/little prenatal classes, less education, passive (i.e. partner) smoking, older maternal age and a higher number of stressful events were significantly associated with maternal smoking in general but did not assess quantity of cigarettes smoked [8
]. An Australian study of similar design to our research also used registry data and found young maternal age, lack of antenatal care and low SES were associated with maternal smoking [34
]. Both papers highlighted the importance of antenatal care as a critical access point to educate expectant mothers regarding a healthy pregnancy. Importantly, the study from Australia found that first-time mothers and those who accessed prenatal care early in their pregnancies had an increased likelihood of smoking cessation [34
The province of BC has a relatively healthy birthing population compared to the rest of Canada and has amongst the lowest rates of maternal smoking and exposure to 2nd
hand smoke in Canada [35
]. Further, BC has high grade 12 completion rates among pregnant women which likely influence the relatively low rates of risk behaviours such as maternal smoking. BC had lower rates of preterm birth and is around the Canadian average for rates of SGA. Despite these positive outcomes, the ability to recognize those at particular risk early in pregnancy and provide preventative programs could help achieve better outcomes for all expectant mothers. Specifically, our findings suggest that heavy maternal smoking will identify approximately 5% of women in BC at particular increased risk of adverse outcomes that may benefit from additional services to promote a healthy pregnancy. With respect to epidemiological analysis of population-based perinatal datasets, there is potential to use heavy maternal smoking as a proxy for unreliable or unmeasured individual-level behavioural and/or socio-economic data. Maternal self-reported smoking tends to be routinely collected for most birth registries making it an accessible variable compared to many other risk factor variables or when linkage to external data in not available.
There were several limitations to this analysis. First, there were no data on passive smoking rates (i.e. exposure to environmental tobacco smoke or having a partner who smokes), psychosocial stress, ethnicity, whether the pregnancy was planned, birth intervals for multiparous women, potential occupational exposures, or household income. Further, some of the covariates examined had high missing data. As described earlier, pre-pregnancy weight was missing approximately 25% of its values, and maternal education data were only available for 28,210 records (12%). The greatest concern when an important variable is poorly populated, is that the absence/presence of values is biased (i.e. are not missing at random). For instance, care providers may only be asking those individuals about their education status where literacy is a concern, and as a result the distribution would be biased and shifted to the left. Therefore missing data not only reduces the statistical power due to list-wise deletion (i.e. records with missing data are not used in that particular test), but also reduces the overall reliability of that variable and potentially the appropriateness of the model.
To address this potential bias, sensitivity analyses were carried out for level of education and pre-pregnancy weight. These tests demonstrated that records with missing education data tended to have small increased risks for all adverse birth outcomes and some maternal risk characteristics; while those missing pre-pregnancy weight data had mixed birth outcomes results but increased risk for most maternal characteristics. However, the overall age structure and cigarette consumption between those with and without missing data were nearly identical with similar medians and inter-quartile ranges. Taken as a whole, these results suggest that the missing education and pre-weight data may result in underestimating the risk of some adverse birth outcomes but the degree of missing data is relatively consistent across the levels of smoking and therefore it would be predicted not to affect the observed general trends of our analyses. Further, a review of the mean number of years of education for the Canadian female population (age 25-36) fall within the mean and standard deviation of our maternal years of education variable, 14.2 versus 13.9 ± 2.6 [36
]. However, given the strong association between maternal education, smoking and risk of adverse outcomes, these results reinforce the importance for the standardized and complete collection of SES variables for all
patients by all prenatal health care providers.
Another potential limitation is the self-reporting bias of cigarette consumption. Self-reported smoking status among pregnant women is susceptible to bias, and may lead to attenuation of the true effect of smoking on birth outcomes [37
]. Rates of misclassification in the United States using data collected from the National Health and Nutrition Examination Surveys (NHANES) estimated non-disclosure to be around 20% [38
]. The rate of misclassification was consistent with other studies [39
], as was the demographic predictors of non-disclosure. Many studies used serum, salivary, or urinary cotinine as a biomarker to assess the degree of non-disclosure in smoking status, and have found a range of between 13% and 25% depending on the cut-off values used to classify one as an active smoker. Non-disclosure was higher among those who reported they were former smokers, and younger maternal age (20-24). The stigma around maternal smoking may lead some respondents to under report their actual consumption habits. Possible recall bias can also be assumed given the peaks in the histogram (Figure ) corresponding at multiples of 5. This could be due to responses given in terms of some fraction in packs of cigarettes per day, such that half a pack is equal to ten cigarettes. None the less, our results suggest that reported
number of cigarettes smoked correlates with adverse birth outcomes and associated socio-economic risk factors suggesting the information as provided will help identify those at highest risk.
Future analyses may include running a hierarchical multilevel model with the inclusion of neighbourhood-level deprivation scores to determine if clustering of observations by neighbourhoods regarding birth outcomes, smoking rates or prenatal care attendance. This type of analysis would be useful as a baseline to further study the effect of local air pollution exposure measured at the neighbourhood-level on birth outcomes and the potential interactions with SES and other risk variables.