Smoking in pregnancy, cord blood cotinine and risk of celiac disease diagnosis in offspring

Ecological observations suggest an inverse relationship between smoking in pregnancy and celiac disease (CD) in offspring. While individual-level analyses have been inconsistent, they have mostly lacked statistical power or refined assessments of exposure. To examine the association between pregnancy-related smoking and CD in the offspring, as well as its consistency across data sets, we analyzed: (1) The Norwegian Mother and Child Cohort (MoBa) of 94,019 children, followed from birth (2000–2009) through 2016, with 1035 developing CD; (2) a subsample from MoBa (381 with CD and 529 controls) with biomarkers; and (3) a register-based cohort of 536,861 Norwegian children, followed from birth (2004–2012) through 2014, with 1919 developing CD. Smoking behaviors were obtained from pregnancy questionnaires and antenatal visits, or, in the MoBa-subsample, defined by measurement of cord blood cotinine. CD and potential confounders were identified through nationwide registers and comprehensive parental questionnaires. Sustained smoking during pregnancy, both self-reported and cotinine-determined, was inversely associated with CD in MoBa (multivariable-adjusted [a] OR = 0.61 [95%CI, 0.46–0.82] and aOR = 0.55 [95%CI, 0.31–0.98], respectively); an inverse association was also found with the intensity of smoking. These findings differed from those of our register-based cohort, which revealed no association with sustained smoking during pregnancy (aOR = 0.97 [95%CI, 0.80–1.18]). In MoBa, neither maternal smoking before or after pregnancy, nor maternal or paternal smoking in only early pregnancy predicted CD. In a carefully followed pregnancy cohort, a more-detailed smoking assessment than oft-used register-based data, revealed that sustained smoking during pregnancy, rather than any smoking exposure, predicts decreased likelihood of childhood-diagnosed CD. Electronic supplementary material The online version of this article (10.1007/s10654-019-00522-5) contains supplementary material, which is available to authorized users.


SUPPLEMENTARY TABLES
All analyses were adjusted for calendar year of birth and degree of hemolysis of cord blood samples (model I).
Model II also included maternal education level, while Model III in addition to previous covariates also accounted for parental type 1 diabetes, celiac disease, income, occupation, cohabitation and paternal education level. The number of children varies between analyses because of missing data (model I: n=910; model II: n=856; model III: n=733). A Analysis of trend over categories of increasing cotinine concentrations. B Stratified analyses by the child's human leukocyte antigen, HLA, genotype classified as conferring a moderate and high risk for celiac disease. Analyses adjusted for non-HLA genetic risk score. See Methods for details. The low-risk HLA group was excluded due to lack of events. 95% CI, 95% confidence interval; aOR, adjusted odds ratio; ETS, environmental tobacco smoke.

SUPPLEMENTARY FIGURE LEGENDS
Supplementary Figure 1. Directed acyclic graph depicting the hypothesized relationships between maternal smoking in pregnancy and offspring celiac disease.
Confounding variables are coloured in red and mediating variables in blue. Biasing paths and causal paths are coloured in red and green, respectively. Child's sex is shown in white indicating predictor for celiac disease that was included as an adjustment variable despite not being associated with maternal smoking. CD-associated HLA and non-HLA genetic risk markers are shown in grey indicating unmeasured predictor for celiac disease as they were only measured in the nested casecontrol study. To ease the interpretation of the diagram, arrows between predictors and mediators (e.g. the arrow between sex and birth weight) and arrows between confounders and mediators (e.g. arrow between parental celiac disease and birthweight) have been omitted because these do not have any consequence for the selection of adjustment variables. All analyses were adjusted for calendar year of birth (model I) and further adjusted for maternal education level (model II). In MoBa and our nested case-control sample we were also able to adjust for parental type 1 diabetes, CD, income, occupation, cohabitation and paternal education level (model III). A sensitivity analysis considered the effect of additional adjustment for parental age at delivery, maternal infections in pregnancy, the child's infection frequency by age six months, sex, breastfeeding duration, parity, delivery mode, gestational age and birth weight. In our nested case-control study we were also able to adjust for HLA and non-HLA genetic risk markers for CD.

Supplementary Figure 2. Complete-case analyses of the association of parental smoking with childhood celiac disease in the MoBa cohort.
All analyses were adjusted for calendar year of birth (model I). Model II also included maternal education level, while Model III in addition to previous covariates also accounted for parental type 1 diabetes, celiac disease, income, occupation, cohabitation and paternal education level The number of children varies between analyses because of missing data. 95% CI, 95% confidence interval; aOR, adjusted odds ratio.

Recording of maternal smoking in the Medical Birth Registry of Norway
The Medical Birth Registry of Norway (MBRN) is held by the Norwegian Institute of Health. 1 The registry contains information on essentially all births in Norway since 1967. Information on maternal smoking in pregnancy is available since 1999. Women may choose not to register smoking status in MBRN; in 16% of children born in 2004-2012 the mother declined to register smoking.
At the antenatal care clinics, the mother's current smoking status (or her declination for it to be registered) is documented in a medical record form ("Helsekort for gravide"); 2 data are recorded at the time of the first antenatal visit (around 10 weeks of pregnancy, range 8-12 weeks) and at the last antenatal visit at 36 weeks of pregnancy. On both occasions, the mother's current smoking status is indicated by check boxes: "non smoker", "occasional smoker, not daily" or "daily smoker", and, if smoking daily, the number of cigarettes smoked per day. 3 At the delivery units, maternal smoking status as recorded in the medical record form is transferred to a summarizing form sent to the MBRN, originally as paper forms that were computerized at the Norwegian Institute of Health. Since 2005, paper forms have been gradually replaced by an electronic transmission of information to the MBRN (by 2014, summarized paper forms were completely abandoned). 3 Minor differences exist between the recordings of maternal smoking in the paper-based and computer-based MBRN record systems; 3 while the original paper form recorded maternal smoking in "early" and "late" pregnancy, the electronic version requests maternal smoking in "first" and "last" trimester. Critically, however, both procedures defines "early"/"first trimester" as the mother's current smoking status as recorded at the time of the first antenatal visit (around 10 weeks of pregnancy) and "late"/"last trimester" as her current smoking status at the last antenatal visit at 36 weeks of pregnancy.
Since 1999, MBRN also provides information on maternal smoking before pregnancy (nonsmoker, occasional and daily smoker). However, data on smoking during the pre-conception period suffer from a high degree of missingness (missing for 85% of children born in 2004-2012); for that reason, the current study did not examine smoking before pregnancy.