Study design and setting
The present study is part of a population-based Dutch birth cohort study: The Prevention and Incidence of Asthma and Mite Allergy (PIAMA) Study. A detailed description of the study design has previously been published [11]. In brief, mothers were recruited from the general population during pregnancy and the children were born in 1996–1997. The baseline population consisted of 3963 children. The study protocol was approved by the medical ethics committees of the participating institutes and all parents gave written informed consent.
Study population
Of 3963 infants, we excluded 190 infants who were born < 37 weeks of gestation as well as 33 infants with missing information on gestational age. We additionally excluded infants with missing values on maternal smoking during pregnancy (n = 33), maternal pre-pregnancy overweight (n = 451), and birth weight (n = 15) from the analysis. Therefore, the population for analysis consisted of 3241 infants.
Data collection
In the PIAMA study, data were collected by self-administered questionnaires which were sent to the parents during pregnancy, at 3 months after birth, yearly from the child’s age of 1 to 8 years, and at 11, 14 and 17 years. The present study used data collected during pregnancy, at 3 months after birth and at the child’s age of 1 and 2 years. Data on maternal smoking were collected during pregnancy. Maternal age (years) and parity was assessed by questionnaire at 3 months after the child’s birth. Data on maternal educational level, weight and height before pregnancy were collected in the 1-year questionnaire and maternal ethnicity in the 2-year questionnaire.
Definition of variables
Exposures and outcomes
Maternal smoking during pregnancy was categorized into smoking and non-smoking and refers to those who reported smoking during at least the first 4 weeks of pregnancy. This variable was created based on information on current smoking during pregnancy and on information on timing of quitting in mothers who reported to be former smokers. Information on current smoking status, smoking intensity and months since quitting was assessed at the time of completing the ‘pregnancy questionnaire’: 95% of the ‘pregnancy’ questionnaires were completed in the 3rd trimester of pregnancy (mean 33 wks; SD 3). Smoking intensity was asked only if mothers were still smoking at the time of completion of the questionnaire, and represent only the number of cigarettes they were smoking at that time. Maternal pre-pregnancy body mass index (BMI) was calculated as body weight (kg) divided by height squared (m2). We dichotomized pre-pregnancy maternal BMI into overweight and obesity (BMI ≥ 25 kg/m2) and non-overweight (BMI < 25 kg/m2). With these two binary variables of maternal smoking and maternal overweight, infants were classified into four groups that will be referred to as: infants who were born to (i) non-smoking, non-overweight (the reference group), (ii) smoking, non-overweight, (iii) non-smoking overweight, and (iv) smoking, overweight mothers.
The primary outcome was term birth weight (grams). In the 3-months questionnaire, parents were asked to report their infant’s birth weight and gestational age from the delivery report. SGA was defined as infants with birth weight below the 10th percentile for gestational age, and LGA was defined as infants with birth weight above the 90th percentile for gestational age according to the standard national growth curve taking into account the infant’s sex and parity [12].
Covariates
Maternal age, ethnicity, educational level and parity (parity for birth weight only) were considered as potential confounders which were set at priori based on literature [13]. Maternal ethnicity was classified according to birth country as Dutch; western, non-Dutch (Europe (except Turkey), North-America, Oceania, Indonesia, Japan); and non-western (Africa, Latin-America, Asia (except Indonesia, Japan), Turkey). Educational level was categorized into three levels: low (primary schools, low vocational training, or lower secondary education), intermediate (intermediate vocational training, or intermediate/higher secondary education), and high (higher vocational training or university degree). The presence of older siblings was used as a proxy for parity and categorized into two groups: none and one or more. Gestational age was measured in weeks.
Statistical analysis
All statistical analyses were performed with Stata statistical software version 12.1 (Stata Corporation, TX). Statistical significance was defined by a two-sided alpha-level p < 0.05. Baseline characteristics of study participants were reported according to the four exposure categories. Differences in proportions for categorical variables were evaluated using Chi square or Fisher’s exact tests. Means with standard deviations were used to describe the distribution of continuous variables. One-way ANOVA with Bartlett’s test for equal variances was used to test the difference in means between exposure groups.
Multivariable linear regression analysis was performed to examine the association of maternal smoking during pregnancy, pre-pregnancy overweight, and the combination thereof on term birth weight. The beta-coefficients represent the difference in term birth weight between infants with the exposure of interest and the reference category. Multivariable logistic regression analysis was used to assess the associations of maternal risk factors with the risk of being SGA or LGA. Associations are reported as odds ratios (ORs) with 95% confidence intervals. All potential confounders were included in the full model, then backward elimination was performed to exclude non-significant confounders one at a time until all variables in the model had a p-value of < 0.20. This conservative level was chosen to be sure that no confounders were missed.
Stratified and joint analyses
After assessing the main effects of maternal smoking and maternal pre-pregnancy overweight, we performed stratified analyses to assess the presence of effect modification. We examined whether (1) the association between maternal smoking during pregnancy and infants’ term birth weight, and the risk of being SGA is modified by maternal pre-pregnancy overweight, (2) the association between maternal pre-pregnancy overweight and birth weight, and the risk of being LGA is modified by maternal smoking during pregnancy.
To assess the joint effect of maternal smoking during pregnancy and maternal pre-pregnancy overweight with 95% confidence intervals, we used a 4-category exposure variable (infants of non-smoking, non-overweight mothers, smoking, non-overweight mothers, non-smoking, overweight mothers and smoking, overweight mothers) in the regression models with adjustment for the same set of confounders using infants of non-smoking, non-overweight mothers as the reference category. The presence of interaction was not tested statistically due to a lack of power.
Sensitivity analyses
We performed three types of sensitivity analyses and we presented the results in the supplement. First, we performed a sensitivity analysis to assess whether the severity of maternal pre-pregnancy overweight changed the estimates and affected our interpretation. We therefore repeated our analyses with three categories of maternal weight: non-overweight, overweight (BMI 25–30 kg/m2), and obese (BMI ≥ 30 kg/m2). Second, we performed a sensitivity analysis taking smoking intensity into account. We expressed smoking intensity (number of cigarettes per day) as a categorical variable. Infants of non-smoking mothers formed the reference category, and infants of smoking mothers were categorized in quartiles of the numbers of cigarettes per day: > 0–2, 3–5, 6–10, and > 10 cig/d. Information on smoking intensity was available for mothers who were current smokers at the time of filling in the ‘pregnancy questionnaire’ which was in the 3rd trimester of pregnancy for 95% of the mothers. Since we defined our maternal smoking variable as smoking during ‘at least the first 4 weeks of pregnancy’, and since 22% of the smoking mothers had quit by the time of filling the ‘pregnancy questionnaire’, we have data on smoking intensity of 78% (429/550) of the infants with smoking mothers. To compare the effect estimates of maternal smoking intensity on birth weight with the effect estimates of the dichotomous smoking variable on birth weight, we repeated the main analysis (Table 2) using the study population excluding the 121 infants of smoking mothers without information on smoking intensity. The third sensitivity analysis was performed to explore the effect of adjustment for gestational weight gain (GWG) in the association under study. We expressed GWG as a continuous variable (kg) and as a categorical variable with cut-off points based on the guideline of the US Institute of Medicine [14]: inadequate, adequate and excessive weight gain. We repeated the main analysis additionally adjusted for GWG and added the results of the joint effect in the supplementary tables. Due to small numbers, we were not able to stratify on GWG and disentangle the independent effect of GWG and pre-pregnancy overweight.