Adriëtte Oostvogels

7 Data analysis Differences in maternal pBMI and offspring’s postnatal growth 1-3 months for different baseline characteristics were examined with χ 2-tests (categorical data) or one-way ANOVA (continuous data). Metabolic components were analysed as z-scores to make comparisons possible. Linear regression was used to examine the association between maternal pBMI and z-scores of the metabolic components, baseline models were adjusted for age at the time of the health check and sex (model 1). In model 2, maternal and offspring’s characteristics were added to control for possible confounders: maternal age, education level, ethnicity, parity, hypertension and smoking during pregnancy, birth weight, gestational age, duration of exclusive breastfeeding, screen time hours per day and energy intake per day (kcal) at 5 years. Subsequently, postnatal growth was added to determine the association between postnatal growth and metabolic components and, at the same time, to test whether postnatal growth mediates the association between maternal pBMI and metabolic components. Finally, the interaction between maternal pBMI and postnatal growth was added to test if postnatal growth was an effect modifier in the association between maternal pBMI and metabolic components. Since SBP and DBP are associated with height, all models including these outcome variables were additionally adjusted for height. Possible sex-specific effects were tested by adding the interaction between sex and maternal pBMI and/or postnatal growth in our model. To provide an adequate fit, all models were tested for possible non-linearity. Restricted cubic splines were used in the regression analysis to examine and characterize an association that is suspected to be non-linear by using higher order piecewise polynomials to accommodate potential changes in the direction of the associations across the exposure distribution. 26 Because the results for all given models indicated that a linear model provided an adequate fit (likelihood ratio test p-value>0.05), we report the results from the linear model. Finally, we undertook a sensitivity analysis in which we restricted our analyses only to those mother-child pairs with Dutch ethnicity to diminish residual confounding by ethnicity. Statistical analyses were conducted using SPSS 20 and R.15.1. A p-value <0.05 was regarded as statistically significant. 181 pBMI, postnatal growth and offspring’s CMP