Adriëtte Oostvogels

International Society for the Study of Hypertension in Pregnancy (www.isshp.org) . 21 Duration of exclusive breastfeeding (<1 month, 1–3 months, >3 months) was available from the infancy questionnaire completed with YHC and the 5-year questionnaire. Because of the longitudinal design of the study we did not correct for postnatal characteristics, as these cannot influence growth at the beginning of life. Hence, duration of exclusive breastfeeding was only used as descriptive characteristic and was not used to correct for in the analyses. Data Analysis Differences in baseline characteristics between the three groups of maternal pBMI and differences between the response and non-response group were examined with χ 2 tests for categorical data and one-way ANOVA for continuous data. The impact of maternal pBMI on the growth of their child was analyzed with a two- stage procedure. In the first stage, because of the skewness in the ABCD growth measures 22 , we transformed the growth data into standard deviation scores (SDS) using a Generalized Additive Model for Location Scale and Shape (GAMLSS) model based on our study population. 23 Population reference curves were preferred over WHO growth standards as we aimed to compare children’s weight, height and BMI SDS curves between different maternal prepregnancy weight status groups. Therefore, the best fitting curves should be used. More detail about the first step can be found in the appendix. In the second step of the analysis, we quantified the associations between maternal pBMI and growth patterns of the children by fitting a mixed effect model to the SD scores. More specifically, the SD score of the i th child at age t , denoted as Z i  (t) , was described by the following model Z i  (t) = X i β + ƒ (t | pbmi i  , sex i  ) + ϕ i + ε where X i are the characteristics of the mother and sex i is the gender of the child. In addition, ƒ (t | pbmi i  , sex i  ) was a smooth function of time conditional on the pBMI of the mother and the gender of the child. To capture the dependency between observations from each child, a child-specific random intercept term ϕ i ~ N ( 0, σ ) was added to the model. The random intercept followed a normal distribution centered at zero with standard deviation σ . The residual error ε ~ N ( 0, τ ) was normally distributed with mean zero normal and standard deviation τ . For the function of age, we used a penalized spline function based on a cubic spline basis defined by a modest sized set of knots (i.e. 9 knots) spread evenly through age. The function was penalized based using the square second derivative, where the optimum amount of penalization was determined by maximum likelihood. 24 152 Chapter 6