The major determinants of children’s diet studied so far are embedded at different levels of the socio-ecological model( Reference Bronfenbrenner 1 ). They include environmental factors such as accessibility, availability and parental attitudes; socio-economic factors such as household income and parents’ occupational status; cultural factors such as parental education, beliefs, habits and tradition; psychosocial factors such as conviviality and family meals; and finally individual factors such as age, gender, taste, food preferences and specific physiological needs related to growth( Reference Scaglioni, Arrizza and Vecchi 2 , Reference Patrick and Nicklas 3 ).
It has been shown that children and adolescents from a low socio-economic position (SEP) are more likely to have suboptimal diets( Reference Darmon and Drewnowski 4 , Reference Fernandez-Alvira, Bammann and Pala 5 ), characterised by low intakes of fruit and vegetables (F&V), fish or dairy products and high intakes of snacks or sugar-sweetened beverages (SSB). It has been suggested that such diets contribute to low micronutrient intakes( Reference Cribb, Jones and Rogers 6 – Reference Pilgrim, Barker and Jackson 9 ) and high energy and fat intakes( Reference Pilgrim, Barker and Jackson 9 , Reference Dubois, Farmer and Girard 10 ). In France, only two national studies have analysed relationships between diet and SEP in school-aged children and adolescents: the Nutrition and Health Survey (ENNS)( Reference Castetbon, Deschamps and Malon 11 ) and the Individual and National Food Consumption Surveys (INCA1 and INCA2)( Reference Lioret, Dubuisson and Dufour 12 ). Data from the ENNS suggested that children from a low SEP (defined on the basis of both head of the household’s educational level and occupational status) had a lower intake of F&V. In children from the INCA1 survey, the head of the household’s low occupational status was positively associated with the ‘snacking and sedentary pattern’ partly characterised by high consumption of French fries and sweetened beverages and low consumption of dairy products; and inversely associated with the ‘varied food and physically active pattern’ partly characterised by high consumption of vegetables and dairy products( Reference Lioret, Touvier and Lafay 13 ). Using parental education as an indicator of SEP, similar trends were observed in 3- to 17-year-old participants of the INCA2 survey, with a positive social gradient observed for core foods such as dairy products and F&V and an inverse gradient observed for non-core foods such as savoury snacks and meat products( Reference Lioret, Dubuisson and Dufour 12 ).
It has been suggested that an energy-dense, nutrient-poor diet increases the risk of chronic diseases such as obesity, CVD and cancer( Reference Lim, Vos and Flaxman 14 ). Socio-economic disparities in diet starting from childhood could therefore partly explain socio-economic inequalities in health later in life( Reference Marmot 15 ). Public health policies therefore face the major challenge of improving the diet of the whole population. In France, nutritional and food policies have been implemented since 2001, including the National Nutrition and Health Programme (PNNS) and the National Programme for Food (PNA). Several of the initial objectives have been partially or fully achieved( Reference Hercberg, Chat-Yung and Chaulia 16 ), such as the stabilisation of the prevalence of overweight and obese children, a reduced consumption of sugar and salt, and the increase in adults’ fruit consumption. However, these improvements have not been spread evenly over all population groups, thus increasing social inequalities in health. The PNNS was updated in 2011, the main focus being the reduction of ‘nutrition-related health inequalities between social classes through specific actions’. Besides, the PNA launched in 2010 aims to ensure food security by providing access to suitable food in adequate quantities to everyone. The Food and Insertion Programme, renewed every year since 2003, also includes actions to promote nutritional balance and the prevention of nutritional deficiency in disadvantaged population groups.
Different indicators have been used to study the relationship between SEP and dietary intakes, but there is no consensus so far on the most relevant indicator to measure SEP in nutritional epidemiology. While educational level, income and, to a lesser extent, occupational status have most often been used, new integrative approaches have been developed to measure the household’s wealth by taking into account various characteristics, such as equipment, housing occupation status and how the financial situation is perceived( Reference Lioret, Touvier and Dubuisson 17 – Reference Camara, de Lauzon-Guillain and Heude 19 ). The SEP indicator chosen is important as the relationship of SEP with diet is likely to differ according to both the particular SEP dimension and the dietary item under study( Reference Turrell, Hewitt and Patterson 20 , Reference Galobardes, Morabia and Bernstein 21 ). A recent article by Zarnowiecki et al. ( Reference Zarnowiecki, Ball and Parletta 22 ) thus suggested using multiple indicators to analyse relationships between SEP and dietary intake in order to assess the influence of SEP on diet more comprehensively and to identify the most appropriate action levers for prevention.
The aim of the present study was to examine the associations between a variety of SEP indicators and both dietary and macronutrient intakes in a representative sample of children and adolescents aged 3 to 17 years living in France.
Materials and methods
Data from the cross-sectional French INCA2 national food consumption survey, carried out between December 2005 and May 2007 by the French Food Safety Agency (AFSSA), were used in the analysis. The survey was approved by the French Data Protection Authority (CNIL).
Population
The INCA2 survey was designed primarily to assess dietary intake in a nationally representative sample of French people. Two independent random samples of 3- to 17-year-old children and 18- to 79-year-old adults were defined on the basis of a multistage stratified sampling design. The sampling frame was extracted from the national census published by the French National Institute of Statistics and Economic Studies (INSEE). First, 181 primary geographical units, stratified by region of residence and size of urban area, were randomly selected with probability proportional to size. Next, households were randomly selected from each primary unit and two independent sampling frames were set up: the first, restricted to households including at least one child; and the second, including households with or without children. Lastly, within each household, either a child or an adult was randomly selected. The sampling design is described in more detail elsewhere( Reference Dufour, Lafay and Volatier 23 ). A participation rate of 69 % was obtained for children and adolescents, yielding a sample of 1455 participants aged 3–17 years.
Measurements
Dietary and macronutrient intakes
Dietary intake was assessed using an open-ended 7 d food record delivered by a trained and certified investigator who explained to the parents and their child how to complete the record. Children aged 10 years or younger were helped by their parents or caregivers to fill it out. Participants were asked to describe all food and beverage intakes over seven consecutive days. They estimated portion size using the SU.VI.MAX (Supplémentation en Vitamines et Minéraux AntioXydants) photographic booklet( Reference Hercberg, Deheeger and Preziosi 24 ) or expressed intakes by weight or household measures.
Dietary intake (in grams per day) was assessed based on the 7 d food record. Food groups were defined consistent with French recommendations (PNNS): vegetables and fruit (including fresh and dried fruit and fruit juices), starchy foods (bread, pasta, rice, potatoes, legumes and other cereals), meat/fish/eggs (meat includes offal and processed meat; fish includes crustaceans), dairy products (milk and dairy drinks, yoghurts and cheese) and SSB (nectars, soft fruit drinks and soda). Other groups considered separately included dairy desserts, cakes and pastries, confectionery, pizzas/sandwiches and stewed fruit/fruit in syrup. The latter were not included in the fruit subgroup because they often contain added sugar and only stewed fruit without added sugar should be considered as fruit according to the PNNS guidelines.
Macronutrient intakes (in kilocalories or grams per day) were evaluated using the French food composition tables( Reference Favier, Ireland-Ripert and Toque 25 , 26 ). Nusser et al. proposed a statistical method (semi-parametric transformation) to estimate usual long-term intakes from intakes observed over a short period( Reference Nusser, Carriquiry and Dodd 27 ). These usual intakes are a better estimation at population scale. They were estimated for the INCA2 study with MSM (Multiple Source Method) software developed by the German Institute of Human Nutrition( Reference Harttig, Haubrock and Knuppel 28 ). As this calculation needs the same number of days for each participant’s food record, we excluded seventy-three children who filled out the food record for fewer than 7 d.
Indicators of socio-economic position
A face-to-face questionnaire was also administered to the adult caregiver. It included various SEP characteristics such as occupational status and educational level of the adult caregiver (the mother in 80 % of cases) in addition to household income. The following household wealth characteristics were also reported: ‘having gone away on holiday for more than 4 consecutive days within the last 12 months’ (yes/no), ‘number of cars in the household’, ‘number of domestic electrical appliances’, ‘how the financial situation is perceived’ (positively/negatively), ‘financial access to desired food products’ (yes/no), ‘whether the idea of lacking food would be a concern’ (yes/no), ‘giving up health care for financial reasons’ (yes/no) and ‘housing occupation status’ (first-time buyer/owner/tenant/tenant in social housing and others).
Five different SEP indicators were used for the current study: the occupational status and educational level of the adult caregiver, the household’s income per consumption unit (ICU), a household wealth composite index and a global SEP composite index which synthesised the previous four indicators. Occupational status was divided into four categories: ‘low’ (unemployed and manual workers), ‘intermediate’ (middle professions and farmers), ‘high’ (executive, top management and professional categories) and ‘inactive’ (retired, homemakers and students). Educational level was divided into three categories: ‘low’ was assigned to those without any qualifications (mid-secondary or below), ‘intermediate’ to secondary school level and ‘high’ to higher education (undergraduate and postgraduate). The ICU was calculated as the income corrected for household size. As 20 % of values were missing, simple regression imputation was used to complete missing income data through the Kohonen algorithm with R software( Reference Cottrell, Smaïl and Letrémy 29 , Reference O’Prey 30 ). First, homogeneous clusters of individuals were identified using a range of variables (such as sex, educational level and the variables used to define household wealth); then individuals with a missing value for income were assigned the income of a randomly chosen individual within the same cluster. The ICU was calculated as recommended by INSEE( 31 ): the household income was divided by the weighted sum of the number of adult and child household members. The most widely used scale, the OECD (Organisation for Economic Co-operation and Development) scale, uses the following weighting: 1·0 for the first adult in the household, 0·5 for all other members aged 14 years or over and 0·3 for children under the age of 13 years. The ICU was then divided into tertiles corresponding to the following thresholds: <725, 725 to <1330 and ≥1330 €/month per consumption unit.
Composite SEP indices were the first components derived from two correspondence analyses performed on the matrix of indicator variables coding SEP characteristics, consistent with previous studies based on the INCA2 study( Reference Lioret, Touvier and Dubuisson 17 , Reference Fillol, Dubuisson and Lafay 32 ). Briefly, the household wealth composite index was constructed taking into account all the measures of wealth previously described. The global SEP composite index was derived from occupational status, educational level, ICU and all the variables also used to define household wealth. In each correspondence analysis, the score of each participant for the first principal component was used as the summary index since it explained about 80 % of the variability. The variables that contributed most to the construction of the two scores were the financial aspects (access to desired food products, fear of lacking food, giving up health care for financial reasons, perception of financial situation) as well as the ICU for the global SEP index. Each score was then divided into tertiles representing low-, intermediate- and high-score groups. Composite indices and tertiles were estimated for the entire population (3- to 17-year-olds) in order to ensure comparability in analyses on children and adolescents.
Statistical analysis
All analyses were conducted using the statistical software package SAS version 9.3. The INCA2 data were weighted for unequal sampling probabilities and for differential non-responses by region, agglomeration size, household size, age, gender, occupation of the household head and season. The complex design of the survey was taken into account using appropriate procedures (survey). The critical P value was set at P=0·05.
All the analyses were stratified by age, so performed separately for children (3–10 years old; n 574) and adolescents (11–17 years old; n 881). Associations between each of the five SEP indicators and both dietary and macronutrient intakes were tested using multivariable linear regression analyses adjusted for age, gender and energy intake.
Results
The characteristics of the studied sample are provided in Table 1. Children and adolescents were on average aged 7 and 14 years, respectively, with about half of each group made up of girls. Children’s caregivers tended to have a higher level of education than adolescents’ caregivers. Regarding occupational status, there was a proportionally higher number of inactive caregivers among adolescents than among children.
SEP, socio-economic position.
* Student’s t test for age; Rao–Scott χ 2 test for percentages.
Socio-economic disparities in dietary intakes
In children aged 3–10 years
As shown in Table 2, children from a lower SEP (lower educational level, ICU or global SEP index) had a lower intake of F&V (with a difference from the higher to lower SEP categories ranging from −39 to −75 g/d depending on the SEP indicator). Within the F&V group, vegetable intake was more strongly associated with SEP (with a difference ranging from −29 to −41 g/d depending on the SEP indicator). Similarly, children from a lower SEP had a lower intake of stewed fruit/fruit in syrup (significant for all the SEP indicators).
F&V, fruit and vegetables; SSB, sugar-sweetened beverages.
* Mean intake (sd) adjusted for age, gender and energy intake·
† Categories for occupational status: C1, low; C2, intermediate; C3, high; C4, inactive.
‡ Categories for educational level: C1, low; C2, intermediate; C3, high.
§ Tertiles for the household wealth index, the global SEP index and the income per consumption unit: T1, <€725/month; T2, €725–1330/month, T3, ≥€1330/month.
Children from a low SEP (lower ICU, household wealth index and global SEP index) also had a higher intake of starchy foods, with differences ranging from +12 to +15 g/d depending on the SEP indicator. Among starchy foods, pasta intake was more strongly associated with SEP (in particular educational level, ICU and global SEP index, attaining a difference of 22 g/d with educational level).
Children from a low SEP (lower occupational status, educational level or global SEP index) had a higher intake of items from the ‘meat/fish/egg’ food group. Differences ranged from +11 to +16 g/d depending on the SEP indicator and were mostly driven by meat intake. Fish intake was not significantly associated with any of the SEP indicators.
In contrast to the overall intake of dairy products, the type of dairy products consumed differed according to SEP. Children from lower SEP had a higher intake of milk (significant associations observed with household wealth and the global SEP index) and a lower intake of yoghurt (significant across all the SEP indicators except ICU).
Children from lower SEP (lower ICU, household wealth index or global SEP index) drank almost twice as much SSB than children from higher SEP (differences ranging from +20 to +45 g/d depending on the SEP indicator).
They also had a higher intake of pizzas/sandwiches (significant across all SEP indicators except ICU) and a lower intake of confectionery (significant across all SEP indicators except occupational status).
Educational level was the SEP indicator most often and strongly associated with dietary intakes.
In adolescents aged 11–17 years
As shown in Table 3, adolescents from lower SEP (lower educational level or household wealth index) had a lower intake of F&V (differences ranging from −45 to −72 g/d depending on the SEP indicator). In contrast to children, vegetable intake was not, however, associated with any of the SEP indicators. The contrary was observed for fruit intake (differences ranging from −24 to −55 g/d across all SEP indicators).
F&V, fruit and vegetables; SSB, sugar-sweetened beverages.
* Mean intake (sd) adjusted for age, gender and energy intake·
† Categories for occupational status: C1, low; C2, intermediate; C3, high; C4, inactive.
‡ Categories for educational level: C1, low; C2, intermediate; C3, high.
§ Tertiles for the household wealth index, the global SEP index and the income per consumption unit: T1, <€725/month; T2, €725–1330/month, T3, ≥€1330/month.
The consumption of starchy foods was not associated with SEP.
Adolescents from lower SEP (lower occupational status, educational level or global SEP index) had a higher intake of processed meat, but no such association was observed with any other item in the ‘meat/fish/egg’ group, nor with this group considered as a whole.
Total dairy product intake was not significantly associated with SEP. However, within this group, yoghurt intake was positively associated with SEP (all SEP indicators except occupational status).
Adolescents from a low SEP (lower educational level, household wealth index and global SEP index) had a higher intake of SSB, with differences ranging from +47 to +92 g/d depending on the SEP indicator. They also had a higher intake of dairy desserts (significant association with occupational status and educational level) and a lower intake of cakes and pastries (significant association with occupational status and global SEP index).
Educational level was the SEP indicator most often and strongly associated with dietary intake in adolescents.
Socio-economic disparities in macronutrient intakes
In children aged 3–10 years
As shown in Table 4, there was no association between SEP and the intake of total energy, protein, carbohydrate or lipid. There were, however, differences in sugar (simple carbohydrate) and starch (complex carbohydrate) intakes. On the one hand, children from lower SEP (all indicators) had a significantly lower intake of sugar. On the other hand, they had a significantly higher intake of starch (significant associations with educational level and ICU). The largest differences were observed with educational level.
* Mean intake (sd) adjusted for age, gender and energy intake (except for energy intake)·
† Categories for occupational status: C1, low; C2, intermediate; C3, high; C4, inactive.
‡ Categories for educational level: C1, low; C2, intermediate; C3, high.
§ Tertiles for the household wealth index, the global SEP index and the income per consumption unit: T1, <€725/month; T2, €725–1330/month, T3, ≥€1330/month.
In adolescents aged 11–17 years
As shown in Table 5, neither total energy intake nor protein intake was associated with SEP. Adolescents from lower SEP (lower occupational status or educational level) did, however, have a higher carbohydrate intake, which was essentially driven by higher intake of starch (significant across all the SEP indicators). Adolescents from lower SEP (lower occupational status, educational level or global SEP index) had a lower lipid intake, which was attributable to differential intakes of both MUFA and SFA.
* Mean intake (sd) adjusted for age, gender and energy intake (except for energy intake)·
† Categories for occupational status: C1, low; C2, intermediate; C3, high; C4, inactive.
‡ Categories for educational level: C1, low; C2, intermediate; C3, high.
§ Tertiles for the household wealth index, the global SEP index and the income per consumption unit: T1, <€725/month; T2, €725–1330/month, T3, ≥€1330/month.
Discussion
The present study is the first in France to study the multidimensionality of the association between SEP and both dietary and macronutrient intakes in school-aged children and adolescents using comprehensive SEP indicators, thus providing important insights into the most appropriate action levers to use within nutritional policies.
Consistent with other studies, our findings confirm both the lower intake of F&V( Reference Cribb, Jones and Rogers 6 , Reference Xie, Gilliland and Li 8 – Reference Castetbon, Deschamps and Malon 11 , Reference Cameron, Ball and Pearson 33 – Reference De Coen, Vansteelandt and Maes 35 ) (in particular vegetables in school-aged children and fruit in adolescents) and the higher intake of SSB( Reference Cribb, Jones and Rogers 6 , Reference Currie, Gabhainn and Godeau 18 , Reference Niven, Scully and Morley 34 – Reference Wijtzes, Jansen and Jansen 37 ) in lower SEP backgrounds. The maximum differences observed between lower and higher SEP categories were equivalent to one 80 g portion of F&V per day and 90 ml of SSB per day, which are equivalent to five fruits (e.g. apples) and two-and-a-half glasses of SSB per week, respectively. In the French national survey on nutrition and health( 38 ), 80 % of the children and adolescents did not consume the recommended five servings of F&V daily and one-third consumed more than the PNNS recommendation of half a glass of SSB (125 ml) daily. The present study thus indicates that the gap between reality and the guidelines is even larger in children and adolescents from more disadvantaged backgrounds. These observations are worth considering for the development of future preventive actions involving diet, especially if they target lower-SEP population groups. In contrast with certain other studies( Reference Fernandez-Alvira, Bammann and Pala 5 , Reference Cribb, Jones and Rogers 6 , Reference Eloranta, Lindi and Schwab 39 ), but in keeping with others( Reference Fernandez-Alvira, Bammann and Pala 5 , Reference Holmes, Roberts and Wells 40 ), our findings did not confirm a positive social gradient for fish intake in children and adolescents. The type of fish consumed is, however, likely to vary according to SEP because of cost differences (for instance, fried or canned fish was shown to be consumed more often in disadvantaged populations than fresh fish, which is more expensive( Reference Darmon and Drewnowski 4 )). This could explain inconsistent results in the literature when fish is considered as a whole, with no distinction. Similarly, while we did not find a positive social gradient for dairy product intake as described in other studies( Reference Xie, Gilliland and Li 8 , Reference Dubois, Farmer and Girard 10 , Reference Rasmussen, Krolner and Klepp 41 ), we found that the type of dairy products consumed was socially differentiated (positive links with SEP for yoghurt and cheese, negative links for milk). Other studies reported that children from a low SEP had higher intakes of high-fat dairy products( Reference Cribb, Jones and Rogers 6 , Reference Holmes, Roberts and Wells 40 , Reference Sanchez-Villegas, Martinez and Prattala 42 ) and lower intakes of low-fat dairy products( Reference Shi, Lien and Kumar 43 ). As for fish, cost barriers could explain social differences in the type of dairy products consumed( Reference Drewnowski, Tang and Brazeilles 44 ). These SEP level-based differences are worth considering since food experiences during childhood influence the development of taste and food preferences, which in turn affect subsequent eating habits( Reference Cooke 45 ). Furthermore, there is some evidence that dietary intakes from childhood carry on into adolescence and even adulthood( Reference Mikkila, Rasanen and Raitakari 46 , Reference Northstone and Emmett 47 ). While social differences in intakes may appear rather small when food groups are considered separately, their accumulation across food groups is worth considering and may have an impact on the opportunity to fulfil nutritional requirements with regard to vitamins and minerals (data not shown). These differences over the life course are also worth considering. Indeed, the early establishment of suboptimal and socially differentiated dietary intakes, even low, which are maintained once habits are formed, is thought to lead to a cumulative increase in positive energy balance over the life course, favouring the development of body fat among the most disadvantaged groups and contributing to social inequalities in health.
With regard to macronutrients, there was no evidence of a SEP gradient for protein intake as described in other studies( Reference Cribb, Jones and Rogers 6 , Reference Dubois, Farmer and Girard 10 , Reference Kronsberg, Obarzanek and Affenito 48 ). This is coherent with the absence of major social differences for protein-rich foods when addressed as a whole (i.e. meat/fish/egg and dairy products). Findings are, however, inconsistent in the literature, which suggests a SEP-driven substitution of animal products within a given food group (fat meat v. lean meat, processed meat v. meat, milk v. yoghurt, canned fish v. fresh fish, etc.), resulting in a similar protein intake across SEP groups( Reference Darmon and Drewnowski 4 , Reference Xie, Gilliland and Li 8 ). In children, carbohydrate intake was not associated with SEP, whereas adolescents from lower SEP had a higher carbohydrate intake. While Cribb et al. observed similar results in their study( Reference Cribb, Jones and Rogers 6 ), findings are not consistent in the literature( Reference Darmon and Drewnowski 4 , Reference Dubois, Farmer and Girard 10 ). This could be partially explained by the distinction made or not made between starch and sugars. We found that children from lower SEP had a higher starch intake and a lower sugar intake (despite higher intake of SSB), which is coherent with their higher consumption of starchy foods (in particular pasta and pizza/sandwiches) and their lower consumption of fruit-based products, yoghurts (lactose) and confectionery. Despite the absence of a relationship between sugar intake and SEP in adolescents, we can however assume that naturally occurring and added sugars are likely to be socially differentiated since adolescents from lower SEP had a lower intake of both fruit and foods containing lactose (e.g. yoghurts and cheese) but a higher intake of SSB. Total lipid intake was not significantly associated with SEP in children, whereas a positive association was found in adolescents (especially for SFA and MUFA). It should be noted that the literature has tended to report an inverse association( Reference Cribb, Jones and Rogers 6 , Reference Xie, Gilliland and Li 8 , Reference Dubois, Farmer and Girard 10 , Reference Kronsberg, Obarzanek and Affenito 48 ). These findings are nevertheless coherent with the observed associations in children between SEP and foods that substantially contribute to lipid intake: SEP was inversely associated with pizzas/sandwiches, meat and processed meat, but positively associated with yoghurts. The interpretation of the findings for adolescents is less straightforward, as SEP was inversely associated with dairy desserts, meat and processed meat but positively associated with yoghurts, cheese and cakes/pastries. Finally, we did not find any evidence of a social gradient in energy intake as described in certain studies( Reference Cribb, Jones and Rogers 6 , Reference Xie, Gilliland and Li 8 ), but in keeping with others( Reference Darmon and Drewnowski 4 , Reference Pilgrim, Barker and Jackson 9 , Reference Dubois, Farmer and Girard 10 ). While this could be due to differential under-reporting bias based on SEP( Reference Lioret, Touvier and Balin 49 ), these findings shed light on social differences in terms of diet quality persisting from childhood to adolescence, with nutrient-dense foods less represented in the diet of children and adolescents from more disadvantaged backgrounds, conversely to energy-dense foods and SSB. Such a diet is supposed to partially mediate the negative association between SEP and overweight in children, as suggested from previous findings based on the INCA1 data set( Reference Lioret, Touvier and Lafay 13 ). In addition, a recent review showed that SSB consumption was positively associated with BMI in children, which could be explained by a high added-sugar content and low satiety effect of liquids( Reference Malik, Pan and Willett 50 ). These results suggest the need to continue and even step up policies to reduce added-sugar intake, particularly from SSB, and promote the consumption of more nutrient-dense foods.
We found that while the associations observed for a given food group were globally robust across the various SEP indicators, the social gradient for dietary intake was steeper for the ‘caregiver’s educational level’ indicator. Zarnowiecki et al. have also shown that the mother’s education appeared to be more frequently associated with children’s dietary intake than other SEP indicators( Reference Zarnowiecki, Ball and Parletta 22 ). The strong association found with the caregiver’s educational level could come from a poorer understanding and appropriation of general nutritional messages, as well as poorer knowledge of the best parental practices which impact food choices for children( Reference Zarnowiecki, Dollman and Parletta 51 – Reference Van Ansem, Schrijvers and Rodenburg 55 ). Adults with a higher education are more health-conscious and thus likely to adopt healthier dietary habits for their children, for example by increasing the availability of healthy foods at home( Reference De Coen, Vansteelandt and Maes 35 , Reference Rasmussen, Krolner and Klepp 41 , Reference Zarnowiecki, Dollman and Parletta 51 , Reference Cook, O’Reilly and Derosa 56 ). It should be noted that when income and educational level were both included in the same model, relationships with educational level remained significant, whereas relationships with income did not (data not shown). The parents’ occupational status, income and household wealth are more susceptible to change over time than educational level, which may also explain the stronger relationship observed between diet and educational level. Besides, in the population studied, income would appear less discriminant than educational level with regard to food choices. Given that the associations under examination were generally stronger with regard to education than with regard to occupation, income or household wealth, it was not surprising to find weaker associations with the global SEP index, the latter being based mainly on income and household wealth characteristics.
Strengths and limitations
Although there is no consensus on the best indicator for measuring SEP in the context of nutritional epidemiology, one strength of our study was to have addressed the multidimensionality of the association between SEP and diet while accounting for comprehensive SEP indicators. We also derived composite indices (household wealth and overall SES), which is a complementary approach rarely undertaken before( Reference Turrell, Hewitt and Patterson 20 ). Our results confirmed a social gradient for a wide range of socio-economic factors, providing robustness and original insights into the issue of social inequalities in diet during childhood, based on the most recent representative data available in France using gold standard measures for diet. However, even if we could investigate the relative associations between different aspects of SEP and both dietary and nutrient intakes at the national level, the measurements of social status addressed here do not encompass all aspects of SEP (for instance, the parents’ cultural origin and area deprivation were not included in the analysis because they were not available in the study). We also acknowledge that household income was imputed to a substantial proportion of the sample (20 %). Sensitivity analyses were therefore conducted excluding participants with missing data for income (data not shown). Despite a trend towards higher P values (probably due to a lower sample size), findings were generally consistent in terms of effect size.
We are aware that a potential bias towards under-reporting of food intake, a weakness inherent to all methods used to assess food intake, cannot be ruled out. This might have affected the estimation of dietary and nutritional intakes, especially in adolescents( Reference Lioret, Touvier and Balin 49 ). While this could possibly explain why fewer SEP differences were identified in adolescents, we cannot exclude that adolescents from all socio-economic groups simply have poorer diets than their younger counterparts. Several reasons led us to include the probable under-reporters in the analyses. It has been argued that low reporting of energy may concern not only those identified as under-reporters, but also those with plausible energy intakes( Reference Livingstone and Black 57 ). Therefore, selectively excluding those with implausible energy intakes could also bias the results. It is also likely that a significant proportion of the children identified as under-reporters are true under-eaters, in cases of weight-loss dieting. This issue is particularly sensitive during adolescence. In the context of a study aiming to describe overall diet on the national level, it makes sense to include individuals who are on restrictive diets.
Finally, the relationships found in the current study are consistent with those observed in adults( Reference Darmon and Drewnowski 4 ). However, our study in the general population did not include the most disadvantaged people. In France, the ABENA study (2004–2005) focused on food-aid beneficiaries and showed that these adults had an unhealthy diet characterised by low intakes of F&V, dairy products and foods from the meat/fish/egg group and ran a high risk of nutritional deficiencies( Reference Castetbon, Mejean and Deschamps 58 , Reference Mejean, Deschamps and Bellin-Lestienne 59 ). There are few data on children’s intake but, depending on the study area, between 25 and 78 % of food-aid beneficiaries had at least one child to care for. A recent article indicated a high prevalence of anaemia and overweight in the children of French homeless families( Reference Vandentorren, Le Mener and Oppenchaim 60 ). Given that nutrient-dense foods are more expensive than energy-dense foods( Reference Maillot, Darmon and Darmon 61 – Reference Andrieu, Darmon and Drewnowski 63 ), we cannot exclude that, in more disadvantaged population groups, income plays a more important role than educational level with regard to the introduction of non-core foods to the children’s diet. Remarkably, the current findings confirm social gradients in the types of food consumed even in this fairly low-risk sample, thus strengthening public health arguments.
Conclusion
By using several SEP indicators, the present study allowed us to explore the multidimensionality of the association between SEP and diet. Social inequalities across SEP indicators were more clearly expressed in terms of food choices than in terms of energy or macronutrient intakes. The gap with regard to national guidelines, wider in lower SEP, suggests the need to focus prevention on specific food groups, especially SSB and F&V. Moreover, the strong associations observed in our study with the caregiver’s educational level confirm that this indicator plays a major role in making healthy dietary choices. Our findings thus support the need to tailor messages for families with a low educational level to help them make healthier choices, pass on healthy eating habits to their children, give them the opportunity of meeting recommendations, and ultimately to give them the keys to better health.
Acknowledgements
Acknowledgements: The authors would like to thank the ISL team (Institut de Sondage Lavialle) for collecting data, and all the families involved for their cooperation. Financial support: This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors. The INCA2 survey was supported by a grant from the French Food Safety Agency (AFSSA). Conflict of interest: None. Authorship: P.D.-P. designed the study, analysed data and wrote the paper. C.D., S.L. and J.-L.V. contributed to INCA2 survey data collection. All the co-authors participated in study design and reviewed the manuscript. S.L. and J.-L.V. contributed equally and are co-last authors. Ethics of human subject participation: The INCA2 survey was approved by the French Data Protection Authority (CNIL).