Study design and population

Data were used from the Amsterdam Born Children and their Development (ABCD) study, a large ongoing community-based birth cohort (http://www.abcd-study.nl/). The cohort study design has been described previously [18]. Between January 2003 and March 2004, all pregnant women living in Amsterdam were invited to participate in the ABCD study. Of the 12373 women approached, 8266 women filled out a pregnancy questionnaire that was available in Dutch, English, Turkish and Arabic. A 5-year questionnaire was sent to the woman’s home address and 4488 questionnaires were filled out by the mothers. These woman’s received a self-administered Food Frequency Questionnaire (FFQ) and 2851 mothers returned the FFQ. At age 10, fathers received a 10-year questionnaire and 2268 questionnaires were returned. A number of 2782 children were included in the present analysis (Fig 1). This study was conducted according to the guidelines laid down in the Declaration of Helsinki. Approval of the study was obtained from the Central Committee on Research Involving Human Subjects in the Netherlands, the medical ethics review committees of the Academic Medical Center, Amsterdam, the VU University Medical Center Amsterdam and the Registration Committee of the Municipality of Amsterdam. All women provided written informed consent. This informed consent procedure was approved by the committees of the Academic Medical Center, Amsterdam, the VU University Medical Center Amsterdam and the Registration Committee of the Municipality of Amsterdam.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flow-chart of the sampling procedure of the ABCD cohort (n = 2782).

https://doi.org/10.1371/journal.pone.0240423.g001

Assessment of socio-economic factors

Data on maternal education, household finance and neighbourhood SES were collected as part of the 5-year questionnaire. Paternal education was collected as part of the 10-year questionnaire. Level of maternal and paternal education were defined as the highest education completed and were defined as: low: (a few years of) primary education/lower secondary education/lower vocational education, middle: general higher secondary education/intermediate vocational education or high: higher professional programmes or university programmes leading to a bachelor or master degree [19]. Household finance was based on the question ‘what is your current financial situation?’ and defined as low (insufficient or exactly enough) or high (ability to save money). Neighbourhood SES was determined using 2010 status scores obtained from the Social and Cultural Plan Bureau (SCP), a division of the Netherlands Institute for Social Research [20]. The children’s neighbourhoods were defined by the four digit postal code at the time of their residence at age 5. Postal code areas in Amsterdam are on average 2.5 km2 and they had a mean population of approximately 9593 residents in 2010 (interquartile range (IQR): 4530–13345) [21]. Status scores were calculated by the SCP for each postal code area using data from registries on the mean educational level, household income and employment status of area residents, analysed by Principal Component Analysis. Mean (SD) status score in the study population was 0.414 (1.18) and mean status score in Amsterdam was -0.282 (1.458) in 2010. Higher status scores represent higher SES of the neighbourhood.

Assessment of potential covariates

Potential covariates that might influence the association between dietary patterns and SES factors were children’s age (y), sex (boys/girls) and ethnicity. Age and sex were collected via the FFQ and ethnicity via the pregnancy questionnaire. Ethnicity was based on the country of birth of the pregnant woman and her mother including both first-generation women (born outside the Netherlands) and second generation women (born in the Netherlands but whose mother was born in another country). Five ethnic categories were formed: Dutch, African Surinamese, Turkish, Moroccan and other ethnicities (mainly non-western origin).

Assessment of diet

At age 5, mothers filled in a validated 71-item FFQ developed by TNO Food (Zeist, The Netherlands) [22]. In brief, per food item, consumption frequency, portion size and the type of product consumed over the last 4 weeks was reported by the mother of the child. Frequency options were “never”, “less than once a week”, “once a week”, “2–3 times a week”, “4–5 times a week”, and “6–7 times a week”. Food items were assessed in units (e.g. a piece of fruit and a slice of bread) and in household units (e.g. a glass and a tablespoon). Intake of items such as breakfast cereals, vegetables or pasta were asked in standard tablespoons which could give a reliable idea of the actual eaten portion size for children. Dietary patterns were derived in a previous study using Principal Component Analyses (PCA) [17]. In that study four dietary patterns were identified: snacking; full-fat; meat; and healthy, explaining 21% of the variation in dietary intake. In the current study we focused on children’s intake of the snacking pattern.

Population characteristics

Population characteristics based on maternal education level are presented in Table 1. Children with lower educated mothers had significantly more often lower educated fathers, lower household finance, lower neighbourhood SES and higher snacking pattern scores than children from middle-high and higher educated mothers, reflecting higher intakes of snack items.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Socio-economic characteristics of the study population by maternal education (n = 2766).

https://doi.org/10.1371/journal.pone.0240423.t001

SES indicators, including maternal education, in relation to the snacking pattern

The associations between maternal education, paternal education, household finance, neighbourhood SES and the snacking pattern score are presented in Table 2. Maternal education was strongly related to the child’s snacking pattern score. Model 1 shows a 0.946 SD higher snacking pattern score (95% CI 0.829; 1.064) for children from low educated mothers and a 0.438 SD (95% CI 0.352; 0.524) higher snacking pattern score for children from middle-high educated mothers compared to children from high educated mothers. After adding paternal education to the model (Model 2), the association between maternal education and snacking score remained strong. Adding household finance and neighbourhood SES (resp. Model 3 and 4) also had a minimal influence on the association between maternal education and snacking pattern scores. Although paternal education, household finance and neighbourhood SES were all significantly associated with the snacking pattern score individually (Table 2), these factors did not overrule the strong association between maternal education and snacking pattern score. Adding all SES factors to the model simultaneously (Model 5) showed similar results.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Associations between maternal education, different SES indicators and the snacking pattern score (n = 2782).

https://doi.org/10.1371/journal.pone.0240423.t002

Effect modification between maternal education and household finance

We observed that the role of middle-high maternal education on the snacking pattern score was somewhat moderated by low household finance (p = 0.089), indicating a difference in the association between maternal education and finance on the snacking pattern score. Therefore analyses were stratified by household finance and all models were run within the low and high household finance group separately (Table 3). Higher maternal education was used as the reference. In Model 1, the association between maternal education and the snacking pattern remained strong in both groups. Within the low household finance group, children from lower educated mothers had a 0.821 SD (95% CI 0.641; 1.000) higher snacking pattern score and children from middle-high educated mothers had a 0.4.92 SD (95% CI 0.334; 0.651) higher snacking pattern score compared to children from high educated mothers. We also examined the potential effect modification between household finance and the other SES factors (S1 Table). Results of paternal education were similar to those for maternal education, but the association with neighbourhood SES was small and not statistically significant.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Effect modification between maternal education and other SES factors, stratified by household finance (n = 2736).

https://doi.org/10.1371/journal.pone.0240423.t003

Interpretations of findings

The associations between different SES indicators and unhealthy dietary patterns indicate distinct pathways of influence. First, higher educated parents may have higher nutritional knowledge and may perceive long term consequences of diet on the health of their children [33], which is associated with healthier food purchasing leading to healthier dietary patterns of their children [34]. Potential explanations are maternal nutrition knowledge; positive concern and attitudes of the mother towards healthy eating [33,35], self-efficacy, feeding practices and modelling, as well as the quality of foods made available and accessible at home [36]. Knowledge and skills attained through education may affect a person's cognitive functioning, making them more receptive to health education messages, or more able to communicate with and access appropriate health services [3]. Parents, and in particular the mother is considered to take the role of gate-keeper that decides which food items are present in the house and on the table [37,38]. In the HELENA study, results from the Healthy Eating Questionnaire showed that mother-child communication was more effective than father-child communication. Mother was the family member most likely to promote healthy dietary habits [26,39].

Second, household finance reflects the availability of economic and material resources and reflect the potential to purchase a healthy diet [3,4]. Possible determinants are the lower cost per mega joule of energy dense and nutrient poor diets and the easy physical access to low-cost energy-dense foods [40]. Other suggested mechanisms are access to health services and indirect to education, promoting self- esteem and social standing [3]. In the group of middle-high educated women, economic recourses are perhaps a stronger limiting factor than knowledge and attitudes.

Finally, studies in the US and Australia, have shown that the foods available in low SES neighbourhoods are of lower quality, cost more, and have less variety, than foods available in higher SES neighbourhoods [10,11]. At the age of 5, familial factors probably play a more important role in influencing dietary patterns than neighbourhood SES [41], but it’s possible that a poor food environment might influence children through its effect on their parents. Indeed we found an independent association between neighbourhood SES and unhealthy snacking pattern. However, we also found that the association between maternal education and the unhealthy snacking pattern was not moderated by neighbourhood SES levels, implying that maternal education (and the underlying mechanisms) may provide some resistance to challenges presented by a low SES neighbourhood.

Strength and limitations

A major strength of this study is the ability to examine whether the role of maternal education in young children’s diet is moderated by other SES indicators. Additionally, this study is one of the few studies taking into account the potential role of neighbourhood SES on unfavourable dietary patterns in young children, in combination with more commonly used SES indicators as parental education and household finance. Studying the role of these SES indicators individually, simultaneously and the possible moderating effect of these indicators allowed specific relationships to be highlighted according to the differences in SES. Neighbourhood SES was calculated based on postal codes where children were living at age 5, when most SES factors and dietary intake were measured.

Several limitations should be taken into account. A possible limitation is that we did not have exact data on how long children had been living in their current neighbourhood. It could have been possible that a family moved to a neighbourhood only recently and was living in a neighbourhood with a different neighbourhood SES before. Our study population (mean neighbourhood status score 0.414 (1.18)), may not be representative of children in Amsterdam (-0.282 (1.458)), or other main cities in the Netherlands (-0.969 (1.573) in Rotterdam and 0.0069 (1.705) in the Hague), indicating higher SES of neighbourhoods in our study population. Neighbourhood SES was calculated by the SCP based on 4-digit postal codes. Using 6-digit postal codes would provide more specific information about actual neighbourhood SES, especially for children living at transitional areas of neighbourhoods [42]. However to our knowledge, no appropriate calculation of status score based on 6-digit postal codes was available for our population.

A multilevel approach for data analysis could be considered, given the hierarchical nature of studying how neighbourhood's SES may affect snacking pattern scores in children. By using multiple linear regression analysis instead, it might be possible that the assumption of independence were violated and the generalization of the research findings are limited. 388 different 4-digit postal codes were included in our analysis and the mean number of participants per postal code was 7.1 (13.6). To ensure sufficient power for statistical analysis, postal code areas with less than 10 children would have been excluded to ensure sufficient power for statistical analysis [32].

Paternal education was collected during the 10-year questionnaire and the number of missing values was substantial (n = 979). Compared to children with complete data, children with missing data on paternal education had more often lower educated mothers (resp. 6.4% and 17.5%), lower household finance (resp. 27.4% and 42.3%), lower neighbourhood SES (resp. 0.414 (1.179) and -0.038 (1.344)) and higher snacking pattern scores (-0.146 (0.902) and 0.268 (1.109)). However, sensitivity analyses within children with complete data on dietary patterns and all SES factors and potential confounders (n = 1765) showed similar results.

Attrition of study population was largely due to untraceable changes in address or migration. A nonresponse analysis determining the degree of selective response and selection bias between pregnancy and birth outcomes, indicated that selective non-response was present in the ABCD-study, but selection bias was acceptably low and did not influence the studied birth outcomes [43]. No further nonresponse analysis was done. It is possible that some biases may have been introduced into the analyses, particularly as the non-responders tended to come disproportionately from lower SES and ethnic minority groups.

Implications

We found that all SES indicators were independently related to snacking pattern scores of children age 5 and we observed a cumulative association of these different SES indicators that add to the risk of an unhealthy dietary pattern. However, the influence of maternal education remained strong, despite addition of other SES indicators. Lower snacking pattern scores were found in children of middle-high educated woman with sufficient household income. This suggests that primary focus for intervention strategies should be given to lower educated mothers, with particularly targeting of mothers were more unfavourable SES indicators are prevalent and middle-high educated mothers with insufficient levels of household finance.