Links between Fiber in Infant Diet and Heart and Metabolism Health at School Age

In this study we examined data from The Generation R Study, a population-based prospective cohort study from fetal life onward in Rotterdam, The Netherlands [25]. Pregnant women were enrolled between April 2002 and January 2006, and 7893 live-born children were available for postnatal follow-up, of whom 4215 had a Dutch ethnic background [25].

Because the food-frequency questionnaire (FFQ) was designed for dietary assessment of a Dutch population and was validated in Dutch children, we restricted our analyses to Dutch children. Children without information on diet (n = 1778) or any of the cardiometabolic health measurements (n = 405) were excluded, resulting in a study population of 2032 children (Figure 1). Because not all of these children had information available on all outcomes, the population for analysis ranged from 1314 to 1995 per cardiometabolic outcome. The study was approved by the local Medical Ethics Committee of Erasmus Medical Center, Rotterdam (MEC 198.782/2001/31, 2001), and written consent was given by parents.

Dietary data were collected using a 211-item semiquantitative FFQ, as described in detail elsewhere [26,27]. This FFQ was validated against three 24 h-recalls and the intraclass correlation coefficient was 0.7 for DF intake [27]. The median age of the children was 12.9 (interquartile range (IQR) 12.6–13.9) months. Total DF intake was calculated using the Dutch Food Composition Table (NEVO), where DF is defined as plant cell wall components that are not digestible by human digestive enzymes, including for example, lignin, cellulose, hemicellulose, and pectin [28]. Thereafter, we divided DF intake in DF from four different food groups, based on the different types of DF that are mainly present in these food groups [3,22]: (1) DF from cereals, as proxy for insoluble DF, defined as fiber from bread, cereals, pasta, rice, cookies, cakes, pastries, crackers, and ready-to-eat meals; (2) DF from potatoes and potato products, containing resistant starch; (3) DF from fruits and vegetables; and (4) DF from legumes, containing different types of soluble DF, insoluble DF, and resistant starch [29].

Children’s height and weight up to the age of 4 years were repeatedly measured during routine visits to Child Health Centers [30]. At a median age of 5.9 (IQR 5.8–6.1) years, children visited our dedicated research center in the Sophia Children’s Hospital in Rotterdam where they were examined in detail. As a primary outcome we used a cardiometabolic risk factor score including the components body fat percentage (BF%), high-density lipoprotein cholesterol (HDL-C), insulin, and triglyceride concentrations, and diastolic blood pressure (DBP) and systolic blood pressure (SBP). Body weight, measured with a mechanical personal scale (SECA), and body height, without shoes and heavy clothes, were measured to the nearest 0.1 kg or 0.1 cm, respectively, and BMI was calculated (kg/m²). Body fat was determined using a dual-energy X-ray absorptiometry scanner (iDXA; General Electrics-Lunar, 2008, Madison, WI, USA) and enCORE software v.13.6 (GE Healthcare, Little Chalfont, UK) [31]. Total body fat was expressed as percentage of total body weight to define BF%. Additionally, we calculated fat mass index (FMI) as fat mass (kg)/m².

Non-fasting blood samples were drawn for measurements of HDL-C, insulin, and triglyceride concentrations, using enzymatic methods (using a Cobas 8000 analyzer, Roche, Almere, The Netherlands) [32]. Quality control samples demonstrated intra-assay and inter-assay coefficients of variation ranging from 0.77% to 1.69%. DBP and SBP were measured, while the children were lying down, at the right brachial artery, using the validated automatic sphygmomanometer Datascope Accutorr Plus™ (Paramus, NJ, USA) [33]. Measurements were repeated four times with 1 min intervals and we used the mean of DBP and the mean of SBP of the last three measurements.

For all outcomes we calculated age- and sex-specific standard deviation scores (SDS) on the basis of our study population. Outliers, defined as the SDS ≥ |3.29| [34], were excluded (n = 0–7 per outcome). Individual cardiometabolic outcomes were combined into a continuous cardiometabolic risk factor score as described elsewhere [17,35]. This score included the sum of age- and sex-specific SD scores of BF%, the inverse of HDL-C, insulin, triglycerides, SBP, and DBP and was transformed into an SD score. A higher score is indicative of a less favorable cardiometabolic profile.

Maternal age, household income, educational levels of both parents, and proxies for mother’s pre-pregnancy cardiometabolic health (i.e., hypercholesterolemia, diabetes mellitus, or hypertension) were obtained with a questionnaire at enrolment in the study. Educational level of both parents was assessed with the same questionnaire and highest finished education was categorized into: no higher education; one parent with higher education; or both parents finished higher education, with higher education defined as higher vocational training, a Bachelor’s degree, or university degree [36].

Maternal height and weight were measured at the research center at enrollment in the study, and BMI was calculated. Information about smoking, alcohol intake, and folic acid supplementation during pregnancy, as markers of health-conscious behavior of the mother, was obtained with questionnaires during pregnancy [27]. Information about pregnancy complications (i.e., gestational hypertension, preeclampsia, or gestational diabetes mellitus) was retrieved from medical records.

Medical records and hospital registries were used to collect information about child’s sex, birth weight and gestational age. Sex- and gestational age-specific z-scores for birth weight were calculated using reference data [37]. Information about receiving breastfeeding at 4 months and timing of introduction of fruit and vegetables was collected with postnatal questionnaires [27].

A food-based diet score for preschool children was used to assess overall diet quality [27] and macronutrient intakes and glycemic load of the diet were calculated [38] from data obtained with the FFQ. Information about receiving dietary supplements (e.g., vitamins and mineral supplements) as a proxy of health-conscious behavior, was retrieved using the FFQ. Information about average screen time as proxy for sedentary behavior and about time spent walking or bicycling to school and playing outside as proxies for physical activity, and smoking in the household (categorized into never; less than once per week; or once per week or more) was obtained with a questionnaire at the child’s age of 6 years.

To explore whether potential associations of DF intake with cardiometabolic health were explained by energy intake, we examined both absolute DF (i.e., not adjusted for energy intake) and energy-adjusted DF intake. DF was adjusted for energy intake using the residual method [39]. Insulin was root-transformed, to obtain a normal distribution. Potential nonlinearity of the associations was assessed using natural cubic spline models with two to four degrees of freedom [40]. Because there were no indications for nonlinear associations for any of the outcomes (all p > 0.05), we assessed all associations using linear regression analyses only. In these models, we analyzed total DF intake and DF intake from different sources (per 1 g/day) and the cardiometabolic risk factor score and individual cardiometabolic health components at the age of 6 years.

In the crude model, we included child’s age at FFQ and sex. To identify potential dietary confounders, we first examined correlations of DF intake with intake of total protein, vegetable protein, animal protein, fat, polyunsaturated fatty acids, monounsaturated fatty acids, saturated fatty acids, carbohydrates, glycemic load, and the diet quality score. We observed a strong correlation for energy intake, glycemic load, and the diet score with DF intakes, and therefore included these variables in the analyses. To avoid overadjustment, the diet quality score and glycemic load were adjusted for DF intake using the residual method. All other covariates described in the Covariates section were included as potential confounders using the manual forward stepwise method starting with the crude model. Variables were retained in the covariate-adjusted model when they resulted in a ≥5% change of the effect estimates of absolute or energy-adjusted DF intake on at least one of the outcomes. Following this procedure, all variables were retained, so covariate-adjusted models were adjusted for: maternal age, maternal BMI, household income, educational level of the parents, smoking, alcohol intake and folic acid supplementation during pregnancy, pregnancy complications, child’s birth weight, sex, receiving breastfeeding at 4 months, timing of introduction of fruit and vegetables, age at FFQ, receiving dietary supplements, glycemic load, diet quality score, physical activity, screen time, and smoking in the household.

We tested for potential effect modification by child’s sex, age at FFQ, and overweight status at the age of 6 years in the crude and covariate-adjusted models. We additionally explored whether DF intake was associated with repeatedly measured height, weight, and BMI using multivariable linear mixed models including the same covariates as in the main models. For the cardiometabolic risk factor score, we performed sensitivity analyses in which we excluded individual components from the score one at a time. Furthermore, we repeated the analyses in a selection of the children with complete data on all cardiometabolic outcomes (n = 1314).

Missing values of covariates were multiply imputed (n = 10 imputations) according to the Fully Conditional Specification method (predictive mean matching), with the assumption of no monotone missing pattern [41]. Effect estimates and population characteristics (Supplementary Materials Table S1) were similar before and after imputation and we report the pooled results after the multiple imputation procedure. Statistical analyses were performed using SPSS 21.0 (IBM Corp., version 21.0, Armonk, NY, USA) and R (The R Foundation for Statistical Computing, version 3.2.0, Aalborg, Denmark) and results were considered statistically significant at a p-value < 0.05.

The population characteristics are shown in Table 1. At the age of 1 year, mean ± SD intake of total DF was 15.0 ± 4.3 g/day, with a range of 3.0–38.6 g/day, with 46.4% of the children having a DF intake above the Dutch recommended intake of 15 g/day for 1–3 year old children [42]. Most of the DF in the diet of our study population came from cereal products (median (IQR) 8.0 (6.2–10.0) g/day) and from fruits and vegetables (4.7 (3.2–6.2) g/day). Overall, parents of most of the children had a high educational level and a high income.

Table 2 presents the crude and covariate-adjusted associations between DF intake and energy-adjusted DF intake and age- and sex-adjusted outcomes of cardiometabolic health. In covariate-adjusted models, a 1 g/day higher energy-adjusted DF intake was associated with a 0.022 (95% CI −0.038, −0.006) SD lower cardiometabolic risk factor score. The effect estimate for DF without adjustment for energy was similar (−0.022 SDS; 95% CI −0.037, −0.006).

Thereafter, we analyzed the components of the cardiometabolic risk factor score separately (Table 2). A 1 g/day higher energy-adjusted DF intake was associated with 0.026 SDS (95% CI 0.089, 0.042) higher HDL-C concentrations. In addition, a higher energy-adjusted DF intake were associated with 0.020 SDS lower triglyceride concentrations (95% CI −0.037, −0.003). DF intake was not associated with BF%, insulin, DBP, or SBP. Intake of DF was also not associated with repeatedly measured height (−0.004 (95% CI −0.012, 0.005)), weight (−0.007 (95% CI −0.017, 0.003)), or BMI (−0.001 (95% CI −0.010, 0.008)), or with FMI at 6 years (−0.003 (95% CI −0.009, 0.009)). Also for the individual cardiometabolic outcomes, effect estimates with and without energy adjustment were similar (Table 2).

Table 3 presents the covariate-adjusted associations between intake of DF from different sources and cardiometabolic health. A 1 g/day higher energy-adjusted DF intake from potatoes was associated with a 0.051 SDS (95% CI −0.094, −0.009) lower cardiometabolic risk factor score, and a 0.076 SDS lower triglyceride concentration (95% CI −0.121, −0.031). Effect estimates without energy adjustment were similar and in addition, there was a positive association between DF intake from potatoes and HDL-C (0.028 SDS; 95% CI 0.016, 0.072).

We observed that a 1 g/day higher energy-adjusted DF intake from fruit and vegetables was associated with a 0.028 SDS (95% CI 0.001, 0.054) higher HDL-C (similar without energy adjustment). Effect estimates for DF from legumes were similar to those obtained for total DF, however, the confidence intervals were wide and none of the associations for DF from legumes was statistically significant. For DF intake from cereals, we observed no associations with any of the outcomes.

We observed no significant interaction of DF intake with child’s sex or age at FFQ on any of the outcomes. Crude associations were comparable with those after adjustment for confounders (). When we restricted our analyses to children with complete data on all cardiometabolic factors, we observed similar results as obtained for the full sample (). When we analyzed the cardiometabolic risk factor score without triglycerides (TG) or without HDL-C in sensitivity analyses, effect estimates only slightly attenuated (). Finally, we observed no significant differences in DF intake, BF%, or blood pressure between children with or without blood samples (). Supplementary Materials Table S2 Supplementary Materials Table S3 Supplementary Materials Table S4 Supplementary Materials Table S5