Study design and cohort

This study was embedded in the Generation R Study, a prospective cohort study that has been previously described in detail. The present study was restricted to prenatally enrolled Dutch women with a live-born singleton, without a medical history of chronic hypertension, diabetes mellitus, hypercholesterolemia, heart disorders, and systemic lupus erythematosus (n = 3187). The study was conducted following the World Medical Association Declaration of Helsinki. Approval was obtained from the Medical Ethics Committee of the Erasmus Medical Center. All participants provided written informed consent.

Nutritional intake

Participants’ responses to a self-administered semiquantitative food-frequency questionnaire (FFQ) assessed nutritional intake in the prior 3 months. The questionnaire was administered in early pregnancy (median, 13.5 weeks; interquartile range [IQR], 3.4), and represented a slightly adapted version of the validated FFQ of Klipstein-Grobusch et al. The FFQ consists of 293 items, structured according to meal pattern. Questions include frequency of consumption, portion size, preparation method, and additions. Portion sizes were estimated using Dutch household measures and photographs showing different portion sizes. We calculated average daily nutrient intake by multiplying the frequency of consumption by portion size and nutrient content per gram based on the 2006 Dutch food composition table.

Principal components analysis was used to identify dietary patterns. First the 293 food items were reduced to 20 predefined food groups. Subsequently, food groups were adjusted for total energy intake and principal components analysis was performed to construct overall dietary patterns by explaining the largest proportion of variation in food group intake. The 2 most prevalent factors–from here on referred to as dietary patterns–accounting for 21.5% of the total variation, were selected after rotating the solution using the varimax method. The decision to select these 2 factors was based on the criterion that they explained the highest percentage of nutritional intake variance in this study population. In addition, both dietary patterns showed resemblances with earlier reported dietary patterns. This empirical approach resulted in 2 statistically independent dietary patterns (correlation coefficient 0.00), each representing recognizable food consumption patterns in the observed world. The factor loadings, (ie, the associations between the respective dietary patterns and all measured food components) are presented by Spearman rank correlation coefficients ( Table 1 ). Foods with loadings ≥0.2 on a factor were used to describe the dietary patterns. The eigenvalue was used to quantify the percentage of variation explained by each dietary pattern. We assigned each participant a personalized score (ie, low adherence, medium adherence, high adherence) for the 2 dietary patterns, representing a quantification of the similarity of the individual’s diet with each of the 2 extracted dietary patterns. After computation of these personalized scores, all 3187 women were classified into equal tertiles according to their personal score for the respective dietary patterns.

Blood pressure

Primary outcome variables were maternal blood pressure (mm Hg), gestational hypertension, and preeclampsia. Maternal systolic blood pressure (SBP) and diastolic blood pressure (DBP) were measured in early, mid (median, 20.5 weeks; IQR, 1.3), and late (median, 30.4 weeks; IQR, 1.1) pregnancy using the validated Omron 907 automated digital oscillometric sphygmomanometer (Omron Healthcare Europe BV, Hoofddorp, The Netherlands). The presence of doctor-diagnosed gestational hypertension and preeclampsia was retrieved from medical records and was determined based on the criteria of the International Society for the Study of Hypertension in Pregnancy ( Supplement Table 1 ).

Covariates

Information regarding maternal age, prepregnancy weight, education, parity, smoking, folic acid use, and vomiting was available from questionnaires repeatedly applied during pregnancy. Additionally, at enrollment in early pregnancy, in mid pregnancy, and in late pregnancy, height and weight were measured and body mass index (BMI) (kg/m ² ) was calculated. Weight gain is represented as the difference in BMI from early pregnancy until late pregnancy. In early pregnancy venous blood serum and plasma samples were drawn to determine folate, vitamin B12, tHcy, and Hs-CRP concentrations. The between-run coefficients of variation for plasma folate were 8.9% at 5.6 nmol/L, 2.5% at 16.6 nmol/L, and 1.5% at 33.6 nmol/L; the coefficients for serum vitamin B12 were 3.6% at 148 pmol/L, 2.7% at 295 pmol/L, and 3.1% at 590 pmol/L; the coefficients for plasma tHcy were 3.1% at 7.6 μmol/L, 3.1% at 13.7 μmol/L, and 2.1% at 26.1 μmol/L; and the coefficients for plasma Hs-CRP were 0.9% at 12.8 mg/L and 1.3% at 39.3 mg/L.

Statistical analysis

To test differences in baseline characteristics between the dietary pattern categories the analysis of variance and χ ² test were used. Likewise, trend tests (linear regression) were used to relate the 3 dietary pattern groups to the biomarker concentrations and nutrient intake.

Linear regression was used to assess cross-sectional differences between the dietary pattern categories in DBP and SBP. In the multiple regression analyses the inclusion of confounding variables was based on earlier literature, and determined a priori. These were maternal age, BMI, parity, educational level, smoking, folic acid use, vomiting, and gestational age at time of measurement. In the multiple regression analyses, missing covariables were completed using multiple imputation (missing: BMI 0.4%, educational level 0.6%, parity 0.2%, smoking 7.5%, folic acid use 17.5%, and vomiting 8.3%). Data were imputed according to the Markov Chain Monte Carlo method assuming no monotone missing pattern. Five imputed datasets were created. Subsequently, multiple regression analyses were performed on each imputed dataset and thereafter combined to 1 pooled estimate.

To further explore blood pressure trajectories between the dietary pattern categories repeated measurement regression models were used with maternal blood pressure as repeated outcome measure. These models take the correlation between repeated measurements of the same subject into account. The best fitting models were constructed using fractional polynomials of gestational age ( Supplement Table 2 ).

Lastly, to analyze the associations of the dietary patterns with gestational hypertension and preeclampsia, simple and multiple logistic regression models were used.

We performed all statistical analyses using the Statistical Package of Social Sciences release 17.0 for Windows (SPSS Inc, Chicago, IL), the Statistical Analysis System version 9.2 (SAS Institute Inc, Cary NC), and R version 2.9.2 for Windows.