Adriëtte Oostvogels

106 Chapter 4 Screen (television and computer) time in hours a day was reported in the 5-year questionnaire. Saturated fat and fibre intake were available from a validated self- administered food frequency questionnaire (FFQ) 18 completed two weeks before the health check. Missing data on covariates were imputed by a multiple imputation procedure using all covariates included in the present study: maternal education level (0.4%), pBMI (5.4%), duration of breastfeeding (0.8%), screen time hours (5.6%), saturated fat (7.8%) and dietary fibre intake (7.8%). No adjustment was made for birth weight, because this was considered as an intermediate between maternal lipid profile and offspring’s lipid profile and glycaemic control. 6-10 Data analysis Baseline characteristics of the study population are described for boys and girls separately. To investigate associations between the independent maternal lipids and the offspring’s lipids and the three glycaemic control characteristics, linear regression was used. Since the residuals in offspring’s TG, C-peptide and HOMA2-IR were not normally distributed, the Box-Cox method in R Package was used to find appropriate transformations. TG was natural log transformed, and C-peptide and HOMA2-IR were both transformed to the power of -1. Potential non-linearity was examined in R with a likelihood ratio test using restricted cubic splines. Non-linear associations were found between two maternal lipids (TG and ApoA1) and offspring’s HDL. We considered these associations as linear, since this is more comprehensible and the non-linear approach gave only slightly different results. Three models were developed to examine the associations. Model 1 was adjusted for gestational age at blood sampling, and offspring’s sex and age at the health check. Model 2 was additionally adjusted for maternal age, ethnicity, pBMI, smoking during pregnancy, parity, alcohol use during pregnancy, hypertension, education level, duration of breastfeeding, offspring’s saturated fat intake, fibre intake, and screen time. In model 3 offspring’s fat percentage was added to assess the influence of this potential mediator on the associations. Effect measure modification by sex was tested on an additive scale by adding an interaction term with sex to model 3. The mediating role of fat percentage was further tested with the mediation analysis by Preacher and Hayes. 19,20 A bootstrapping approach (resampling rate 10,000) was used to estimate a confidence interval for the regression coefficient of the mediated effect (Figure 1: path a*b) for all maternal lipids combined with an individual outcome variable. These analyses were adjusted for all variables in model 2, thereby also taking potential confounders in the association between offspring’s fat percentage and offspring’s lipid profile and glycaemic control into account. Mediation of offspring’s fat percentage was present if the confidence interval indicated statistical