Adriëtte Oostvogels

2 39 pBMI, lipids and BP during pregnancy imputation by chained equation (MICE) was used after descriptive analyses. Missings for maternal education (0.61%), ethnicity (0.16%), alcohol use (0.03%) and smoking (0.06%) during pregnancy, foetal sex (0.03%) and pBMI (6.13%) were imputed. 23 BP course during pregnancy was investigated using a model described by Macdonald-Wallis 7 : as a sequence of individual BP data are likely to be dependent, two-level mixed models were used to account for the possible correlation between the observations within the same woman. Moreover, the model accounted for gestational age at birth, as women with higher BP tend to give birth earlier than women with lower BP. 7,9 Piecewise linear splines with BP as a function of time with three internal knots, based on the best fitting model, and also allowing for individual departure from the mean slopes [and intercept] were fitted on the study sample of 3100 women. Based on the Akaike’s Information Criterion (AIC) the best fitting model with three internal knots for SBP course yielded internal knots at 16.5, 31.5 and 35.5 weeks of gestation, while the best fitting model for DBP was with internal knots at 19.5, 31.5 and 35.5 weeks of gestation. Associations between weight status, early pregnancy lipid profile and BP course in the four time periods were tested in three subsequent models. First unadjusted associations were tested in a crude model, then adjustments were made for age, education, parity, ethnicity, smoking and alcohol use during pregnancy and foetal sex. To test for differences between women with normal weight and overweight in BP course we also fitted a model with interaction between weight status and time period, and compared the model with and without the interaction term through a likelihood ratio test. Finally, we added the early pregnancy lipids, divided into tertiles, separately. Tertiles were chosen as it is difficult to make comparisons in these analyses using the continuous measurements. Moreover, because preliminary analyses revealed little differences in results between quintiles and tertiles, to increase the statistical strength we chose tertiles for further analyses. To test for differences in BP course for women in the different early pregnancy lipid tertiles, we also fitted a model with interaction between tertiles of early pregnancy and time period, and compared the model with and without the interaction term through a likelihood ratio test. Furthermore, to investigate the possible modification by weight status of the above-mentioned association, three-way interaction terms between weight status, lipids and time periods were incorporated into the model. The overall significance of these possible alterations was again assessed using likelihood ratio tests. Sensitivity analyses were performed by repeating all analyses in two subsamples. These subsamples were created by dividing our study population into a subsample with no complications (gestational age at birth ≥37 weeks and no PIH/PE) and a subsample with complications (gestational age at birth <37 weeks and/or PIH/PE).