Maternal lipid levels in early pregnancy as a predictor of childhood lipid levels: a prospective cohort study

This study was embedded in the Generation R Study, an ongoing population-based prospective cohort study from early pregnancy onward in Rotterdam, the Netherlands [14]. Women enrolled between April 2002 and January 2006. All methods were performed in accordance with the relevant guidelines and regulations. For this study, we included women with a live born singleton and available information on lipid measurements in early pregnancy. We excluded women with a twin pregnancy, (gestational) diabetes mellitus, and women on glucose or lipid regulating medication at enrolment. Children without available lipid measurements at the different time points (6 and 10 years after pregnancy) were also excluded (Fig. 1). Additional file 2 contains a Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement for the current study [15].

The study has been approved by the Medical Ethical Committee of the Erasmus Medical Centre in Rotterdam (MEC-2007-413). For minors, written informed consent from a parent and/or legal guardian was obtained. From the rest of the participants written informed consent was obtained.

Non-fasting blood samples were obtained at enrolment in early pregnancy (median 13.2 weeks [90% range 10.6; 17.1]) by trained research nurses to determine the levels of total cholesterol, triglycerides, and high-density lipoprotein cholesterol (HDL-c) [16]. Total cholesterol, triglycerides and HDL-c were analysed using standard laboratory methods. The mean interassay coefficients of variation for total cholesterol, triglycerides, and HDL-c were, respectively, 4.2, 3.4, and 2.8%. Levels of LDL-c were calculated using the Friedewald equation [17]. Non-HDL-c was calculated by subtracting HDL-c from the total cholesterol and remnant cholesterol as ([total cholesterol – LDL-c] – HDL-c). Details of processing procedures and lipid calculations have been described previously [2, 7, 18].

Placental syndrome was defined as having pre-eclampsia, a child born small-for-gestational age (SGA) or a spontaneous preterm birth (sPTB) in the index pregnancy. We obtained information on clinically diagnosed pre-eclampsia from medical records that were cross-checked with the original hospital charts [19]. Pre-eclampsia was defined, using the ISSHP criteria that were in effect at the time of the study, as the development of systolic blood pressure (SBP) ≥ 140 mmHg and/or a diastolic blood pressure (DBP) ≥ 90 mmHg with new-onset proteinuria in a random urine sample and no evidence of a urinary tract infection [20]. Midwife and hospital registries provided information on gestational age at birth, birth weight, and child’s sex. SGA was defined as a child with a birth weight below the 10th percentile, adjusted for gestational age and sex of the child. We defined sPTB as the spontaneous onset of labour before 37 weeks of gestation.

Non-fasting venous blood samples were collected to measure total cholesterol, triglycerides and HDL-c in the children at the age of 6 (median 6.0 years [90% range 5.7; 7.5]), and 10 years (median 9.7 years [90% range 9.5; 10.3]). Blood samples were obtained, transported, and stored as described in detail previously [16]. Serum concentrations of total cholesterol, triglycerides and HDL-c were measured at the Erasmus Medical Centre with enzymatic methods (Cobas 8000, Roche, Almere, the Netherlands) [16, 21, 22]. Levels of LDL-c, remnant cholesterol and non-HDL-c were calculated [17].

Information on maternal characteristics during pregnancy including age, ethnicity, educational level, parity, smoking, and the use of cholesterol or glucose regulating medication, and information on smoking and gravidity 6 and 10 years after pregnancy was obtained through questionnaires. We obtained information on pre-pregnancy weight through questionnaires at study enrolment. Maternal weight (kilograms) and height (centimetres) were measured at enrolment in early pregnancy and 6 and 10 years after pregnancy without shoes and heavy clothing, after which body mass index (BMI) was calculated (kilograms per square meter). Pre-pregnancy weight and measured weight at enrolment were highly correlated (Pearson’s correlation coefficient 0.97 [value of P < .001]), and therefore, pre-pregnancy weight was used to calculate BMI in the analyses. At the age of 6 and 10 years, weight and height were measured without shoes and heavy clothing. We calculated BMI, and categorized children into age- and sex specific groups; normal-weight, overweight and obese, using the definition of Cole et al. [23]. Women’s dietary intake in early pregnancy was assessed using a semi-quantitative 293-item food frequency questionnaire (FFQ) at enrolment. The FFQ included foods that were frequently consumed in the Dutch population and was modified for use during pregnancy. National dietary guidelines were used to develop a predefined diet quality score for pregnant women. The diet score included 15 components in which the scores for the individual components were summed, resulting in an overall score ranging from 0 to 15, with a higher score representing a healthier diet [24].

First, we describe pregnancy and follow-up characteristics for all women and children. The mean ± standard deviation is presented for data with a normal distribution and the median with 90% range for data with a skewed distribution. Second, we examined the distribution of early pregnancy lipid levels in women, and in their children at the age of 6 and 10 years. Third, we present the lipid distributions in children stratified on child’s sex. Paired sample t test and Wilcoxon signed rank test were used for comparison of paired samples. Independent samples were tested through Students t test and Mann-Whitney U test. Fourth, triglyceride and remnant cholesterol levels were log transformed to achieve a normal distribution. To enable comparison of effect estimates, we constructed SD-scores (SDS) for all lipid levels [7]. We tested whether the association of maternal lipid levels with corresponding childhood lipid levels was nonlinear. Fifth, since we found a linear relation, we further examined the association between maternal lipid levels in early pregnancy and the corresponding childhood lipid measurements at the age of 6 and 10 years using multiple linear regression models. These models were adjusted for maternal age at enrolment, gestational age at blood sampling, ethnicity, parity, educational level and pre-pregnancy BMI. Sixth, missing data of the covariates were imputed. We used the Markov Chain Monte Carlo multiple imputation procedures to reduce potential bias attributable to missing data [25]. In this study, 1.3% had missing information on ethnicity, 4.8% on educational level, 0.5% on parity, 9.6% on smoking in early pregnancy, 16.6% on pre-pregnancy BMI. For the regression analysis, we used the SDS for all lipid levels as exposure. Regression models adjusted for confounders were: basic model (maternal age at enrolment, gestational age at blood sampling, educational level, ethnicity, parity and smoking), BMI model (basic model additionally adjusted for pre-pregnancy BMI) and maternal diet model (BMI model additionally adjusted for diet score). These confounders were selected based on previous studies, and on their associations with the exposures and outcomes of interest. The effect estimates in Fig. 2 are regression coefficients from the BMI model and represent an increase in the outcome measure for each unit increase in the exposure. Seventh, we tested for unmeasured confounding through the E-value method (Additional file 1, Table S1) [26]. Eight, we tested whether there was effect modification in all associations by sex of the child, through inclusion of the interaction term (exposure * sex of child) in each regression model. Since we found no effect modification for the associations, we did not stratify the results on child’s sex. In addition, child’s sex did not contribute to our models, therefore we chose not to include child’s sex as a confounder. Ninth, in attempt to exclude the effect of placental syndromes, we performed the regression analyses in a subgroup of women without placental syndromes in their index pregnancy. Lastly, for clinical purposes we defined the 90th percentile for maternal gestational lipid levels (total cholesterol, triglycerides, LDL-c, remnant cholesterol and non-HDL-c), whereas for HDL-c we defined the 10th percentile. Thereafter we tested the association of these cut-off values with the corresponding lipid measurements in their children. Statistical analyses were performed with SPSS version 21.0 for Windows (SPSS INC, Chicago, IL, USA).

This study included 2692 mothers with lipids measured in early pregnancy and children with lipids measured at 6 years. Ten years after pregnancy, lipid measurements were available in 1673 children (Fig. 1). Women were on average 30.3 (SD 4.9) years at the start of pregnancy (Table 1). They were mostly European, highly educated and nulliparous. Their pre-pregnancy BMI was 23.7 (90% range 19.5; 32.9) kg/m² and 26.6% of the women smoked during their pregnancy. At the age of 6 years (median 6.0 years, 90% range 5.7; 7.5) children’s BMI was 15.9 (90% range 14.0; 19.4) and at the age of 10 years (median 9.7 years, 90% range 9.5; 10.3) children’s BMI was 16.9 (90% range 14.4; 22.9).

Table 2 shows the lipid profile of women in early pregnancy and the lipid profile of children at the age of 6 and 10 years. BMI, total cholesterol, triglycerides, HDL-c, and remnant cholesterol levels are significantly higher, while LDL-c and non-HDL-c levels are significantly lower at the age of 10 years than at the age of 6 years. Additional file 1, Table S2 shows the lipid distribution in children stratified by child’s sex. At 6 and 10 years, girls have a more atherogenic lipid profile with higher levels of total cholesterol, triglycerides, LDL-c, remnant cholesterol, non-HDL-c, and lower levels of HDL-c than boys.

Maternal total cholesterol, triglyceride, LDL-c, HDL-c, remnant cholesterol and non-HDL-c levels are all positively associated with the corresponding measurements in children at the age of 6 and 10 years (Fig. 2). These results are independent of maternal pre-pregnancy BMI and diet (Fig. 2 and Table 3). In addition, the results of our regression analyses were not influenced by placental syndrome since the results were similar when the analyses were repeated in a subset of women without placental syndrome in their index pregnancy (n = 2143) (Additional file 1, Table S3). In Additional file 1, Table S4 we show the association of cut-off values of maternal lipid levels in early pregnancy with the corresponding childhood lipid measurements. We found that all levels were positively associated, independent of confounders.

We had a prospective data collection from early pregnancy onwards and a large sample of 2695 women and children with lipid measurements 6 years after pregnancy, and additional lipid measurements of 1673 children at the age of 10 years. Having two time points of lipid measurements during childhood is unique. In addition to the traditional lipid levels, we also assessed non-HDL-c and remnant cholesterol measures in children, since remnant cholesterol and non-HDL-c have already been proven important risk factors of cardiovascular disease in adults [58, 59]. Blood samples were collected with a minimum fasting time of 30 minutes and are therefore non-fasting. Changes in lipid and lipoprotein levels are considered minimal in response to normal food intake [60, 61]. In addition, lipid levels will also differ between fasting measurements in the same individual [60]. Non-fasting lipid measurements may also be more applicable in clinics since it may be more difficult to measure pregnant women in a fasting state. Regarding dietary assessment, data from the FFQs are self-reported and memory based, and the diet quality score ranging from 0 to 15 is rather rough. However, FFQs are still widely used as the primary dietary assessment tool in epidemiological studies. We tested unmeasured confounding for the regression models, and found that our models are considered rather robust. However, physical activity prior to or in early pregnancy may be considered as a potential confounder. Unfortunately, we did not have information on this variable and therefore did not include this potential confounder. Generalizability of this study may be comprised due to non-response at follow-up and since the women in this study had a relatively low BMI, which indicates selection towards a healthy population. Although we included many lifestyle factors, residual confounding may still be present. Future studies are warranted to investigate the genetic contribution on maternal lipid levels in pregnancy and lipid levels of their offspring years later.