Objective: To study differences in dietary intake between adults with different socioeconomic status (SES) and trends over time.
Design: Cross-sectional study based on data of three Dutch National Food Consumption Surveys (DNFCS-1 1987/88; DNFCS-2 1992; DNFCS-3 1997/98), obtained from a panel by a stratified probability sample of the non-institutionalized Dutch population.
Subjects: A total of 6008 men and 6957 women aged 19 y and over.
Method: Dietary intake was assessed with a 2 day dietary record. Background information was obtained by structured questionnaire. Sociodemographic variables were available from panel information. SES, based on educational level, occupation and occupational position was categorized into (very) low, middle and high. Analysis of variance with age as covariable was used to explore the effects of SES on dietary intake and anthropometry. Statistical tests for trend were carried out with models in which week-weekend-day effects and an interaction term of time with SES were also included.
Results: The prevalence of obesity and skipping of breakfast was higher among people with a low SES. In all three surveys, subjects in the (very) low SES group reported having a higher consumption of potatoes, meat and meat products, visible fats, coffee and soft drinks (men only). Subjects with a high SES reported consuming more vegetables, cheese and alcohol. As regards nutrients, in all surveys a higher SES was associated with higher intake of vegetable protein, dietary fibre and most micronutrients. A higher SES was also associated with a lower fat intake but the differences between social classes were rather small and not consistent when the contribution of alcohol to energy intake was taken into account.
Conclusion: In general, dietary intake among subjects in higher SES groups tended to be closer to the recommendations of the Netherlands Food and Nutrition Council and this phenomenon was quite stable over a period of 10 y.
Sponsorship: The present study was supported by the Dutch Ministry of Health, Welfare and Sports.
In most Western European countries standardized morbidity and mortality are inversely related to socio-economic status (SES; Fox, 1989; Illsey & Svensson, 1990; Townsend & Davidson, 1992; Blaxter, 1990; Feldman et al, 1989; Mackenbach et al, 1997). Among the Dutch population the greatest disparities in health inequalities are based on socio-economic status (SES). People with a low (primary) educational level die, on average, 3.5 y earlier and have almost 12 fewer years of good health than people with university education (Mackenbach & Verkley, 1997). Health inequalities based on SES are not declining, and in some respects they have actually increased (James et al, 1997; Mackenbach & Verkley, 1997). Countries in Western Europe were found to be generally similar in the size of socioeconomic inequalities in health (Mackenbach et al, 1997). Several studies showed that some risk factors, such as unfavourable lifestyle factors, have a higher prevalence in lower SES strata (Cavelaars et al, 1997; Adler et al, 1994; Hoeymans et al, 1996). In general, less educated and lower income groups appear to consume a less healthy diet (Prättälä et al, 1992; Smith & Baghurst, 1992; Shimakawa et al, 1994).
In the Netherlands, the Dutch National Food Consumption Survey (DNFCS) provides the opportunity to get an insight into the consumption of food among different population groups. In 1987–1988, a national representative survey (DNFCS-1) was conducted; similar surveys were carried out in 1992 (DNFCS-2) and in 1997–1998 (DNFCS-3). The results of a former study, based on the DNFCS-1 indicated that a lower SES was accompanied by a higher prevalence of several indicators of an unhealthy lifestyle (Hulshof et al, 1991). The three cross-sectional DNFCS allow an analysis of the dietary intake over a period of 10 y. The aim of the present study was to investigate whether differences in food consumption according to SES have changed during 1987–1988 to 1997–1998.
The three DNFCSs (1987–1988; 1992 and 1997–1998) were carried out within the framework of the Dutch Nutrition Surveillance system (Löwik & Hermus, 1988; Hulshof & Van Staveren; 1991; Löwik et al, 1994, 1998; Anonymous, 1998). Data collection was performed by a marketing research institute experienced in nationwide surveys. The surveys were financially supported by and carried out under the authority of the Ministry of Agriculture, Nature Management and Fisheries, the Ministry of Health, Welfare and Sports and the Netherlands Nutrition Centre (DNFCS-3). Households were selected from an existing panel. Institutionalized individuals, households with a housekeeper aged 75 or over, children younger than 1 y, and households with a person with inadequate fluency in the Dutch language were not eligible. In 1997–1998 an additional sample of persons living in households, where the housewife is older than 75 y was included as well. Of the 2793 households contacted in 1987–1988, 2203 agreed to participate, comprising 5898 subjects aged 1–85 y. The response rate was 79% on a household level and 80% on an individual level. The sample in DNFCS-2 comprised 6218 subjects, aged 1–92 y, living in 2475 households (response 72% on a household level as well as on an individual level). In DNFCS-3 5958 subjects excluding and 6250 subjects including the additional sample, aged 1–97 y participated (response 70.5 and 68.5%, respectively). In the present study analyses are based on individuals aged 19 y and over. To study the trends properly, elderly belonging to the additional sample in DNFCS-3 (n=292) were excluded. Younger people were excluded because lifestyle characteristics and food consumption patterns for children, adolescents and adults differ and therefore should be studied separately. Because pregnancy may affect dietary habits, pregnant women were excluded too. In total, 12 965 subjects (6008 men and 6957 women) remained for statistical analyses.
The food consumption data were collected through a 2 day dietary record method. The foods consumed at home were recorded in a household diary for all individual members of the household by the person usually engaged in preparation of the meals. Consumption away from home was recorded by every participant in a personal diary (children less than 13 were assisted by a parent or both parents). In each survey food consumption data were collected during 40 weeks per year and evenly distributed over the seasons and the 7 days of the week. No field work was conducted during public holidays because low response levels were expected on these days and in these periods. Specially trained dietitians were responsible for the field work, including contacting participants, instructions regarding completion of the diaries, checks on completeness of data and estimates of capacity of household utensils (common household measures and foods regularly used were weighed), and coding of the data.
In each survey food consumption data on the individual level were converted into energy and nutrients using the most up-to-date version of the Dutch food composition table. To enable valid comparisons of nutrient intake over time specific food composition tables were created. The food composition table used for DNFCS-3 (last update in 1997) was the reference table. To avoid artificial changes, differences as a consequence of improvement of the database were corrected in retrospect; real changes in the food composition remained intact (Beemster et al, 2000). Intake of vitamin and mineral supplements was not included in the calculations of nutrient intake since in DNFCS-1 brand levels of these supplements had not been recorded.
In addition to the food consumption data, information was collected on the respondent's body weight, height (both self-reported figures), use of nutritional supplements (in DNFCS-1 only yes/no), special dietary practices (on own initiative and on prescription) and smoking habits by means of a structured questionnaire included in the personal diary. The method of data collection has been extensively reported elsewhere (Hulshof et al, 1991; Löwik et al, 1994, 1998). Other personal data, such as household composition, education, occupation and residence, were already known to the marketing research institute responsible for the field work.
SES, based on education, occupation and occupational position (Hulshof et al, 1991), was categorized into high (for instance university graduated liberal professions, executive managers supervising at least 10 persons), medium, low and very low (unskilled work force). Associations among discrete variables were analysed using age-adjusted and gender-specific logistic regression models. Analyses of variance with age as covariable was used to explore the effects of SES on dietary intake and anthropometry among men and women separately. Statistical tests for time effects (trends) were carried out with models in which week–weekend effects and an interaction term of time with SES were also included. The assumptions of normality and equality of variance were checked by analysis of residuals. When a distribution was skewed values were transformed logarithmically. Only differences with a P-value <0.01 were taken into account. All analyses were performed using the SAS package V6.12 (SAS Language, 1998).
Distributions of age and SES are shown in Table 1. In particular, as a consequence of a better education over time, a shift was observed from a low SES to a medium and high SES. The overall mean age of women was 43.0 y and for men 42.9 y.
Selected characteristics of the population are presented in Figure 1. The overall prevalence of obesity body mass index; (BMI≥30) increased over a period of 10 y from 4.2 to 7% and 7.6 to 13.2% for men and women, respectively. The proportion of obese subjects was highest in the lower SES classes, particularly among those with a very low SES (Figure 1). More women than men reported being on a dietary regimen. The proportion of dieting men was quite stable. For women an interaction between time and SES was observed; compared with the first two national surveys, in DNFCS-3 the prevalence of dieters was higher in the highest SES group and lower in the lowest SES group (Figure 2). In all groups most of the diets were prescribed by a physician or dietitian. Breakfast skipping was highest in the low SES group (results not shown). Overall, the use of nutritional supplements increased with time and was more common in the higher classes; a slight but significant interaction between time and SES was observed (Figure 2). Most common kinds of nutritional supplements were multi-supplements and supplements with only vitamin C. In all surveys and in all SES groups these supplements were the most important.
In all surveys, a low(er) SES was associated with a higher proportion of users for potatoes and soft drinks. A higher SES was associated with a higher proportion of users for fruit, cheese, savoury snacks, fish, ready-to-eat meals (men only), fruit juice, wine and tea. Compared with the lower class, among women in the higher class more users of beer, cereals and sugar and sweets were also observed, whereas in the lower SES groups the proportion of coffee drinkers was higher. In contrast to men, among female soft drink users no difference between high and very low social class was found.
In Table 2 the mean daily consumption of selected food groups for users only is presented according to SES for each survey. Men in the low(er) SES group(s) were found to consume more potatoes, bread, sugar and sweets, spreads and cooking fats, meat, beer and coffee then men in the higher SES groups. Compared with the higher SES groups female users in the low(er) group(s) consumed more potatoes, fats and oils, meat, soft drinks and coffee. On the other hand, men and women in higher SES groups consumed more vegetables, cheese, wine and women also ate more cereals and drunk more tea than those in the low(er) social class. Interactions between time and SES were observed for cereals and soft drinks among men and for savoury snacks among women (Figure 3).
Over a period of 10 y in all SES groups the consumption of vegetables, fruit and sugar and sweets decreased significantly, whereas the consumption of ready-to-eat meals increased. Among men the consumption of potatoes also declined; among women the consumption of cheese decreased and the consumption of tea increased in all SES classes.
Intake of energy and nutrients
The average intake of energy and selected nutrients according to SES for each survey is summarized in Table 3. Median values are presented when a distribution was skewed. Men in the low SES groups had a slightly higher energy intake (ca 0.4 MJ) compared with the higher SES groups. Among men and women a higher SES was associated with a higher (vegetable) protein intake, a lower fat intake (total fat and unsaturated fatty acids) and a higher alcohol intake. However, the differences were rather small and not always consistent when the contribution of alcohol to energy intake was taken into account. For most minerals and vitamins the absolute intake as well as the intake per MJ (results not shown) was positively related to SES. Interactions between time and SES were hardly observed. Only among men was an interaction observed for dietary fibre in g/day. However, expressed per MJ this interaction disappeared and the lowest intake was noted for both sexes in the low SES classes.
Over the years, among men in all SES groups total energy intake decreased. For both genders, in all SES classes the percentage of energy from protein, particularly vegetable protein, and carbohydrates rose and the percentage of energy from fat and the median intake of vitamin A (RE) declined.
The main findings of the present study are that, over a period of 10 y, significant changes were observed in food and nutrient intake, that food consumption patterns varied according to SES and that the differences between SES classes were rather stable over time.
The drop in the percentage of energy from fat between 1987/1988 and 1997/1998 is an encouraging indication that the Dutch are modifying the fats in their diet in line with the recommendations (Netherlands Nutrition Council, 1986). However, at the same time some other trends are of concern. The most negative finding with regards to health consequences is the decrease in the consumption of fruit and vegetables. These trends were present in all social classes.
In our study the average reported consumption of vegetables and fruit was higher in the high SES group. For fruit this was due to a higher proportion of users; the consumption among the users did not differ or was even greater among men in the low SES group. A systematic review of socio-economic differences in food habits in Europe also showed that, particularly in the North and West of Europe, a higher SES is associated with a greater consumption of both fruit and vegetables (De Irala-Estévez et al, 2000). In agreement with our observations, those with higher education also tend to consume less fats and oils but more cheese (Roos et al, 1999). Subjects with a higher educational level tend to be more aware of the characteristics of a healthy diet (Margetts et al, 1997) and have more knowledge about food items which are healthier (Martinez-Gonzáles et al, 1998; Hjartaker & Lund, 1998; Margetts et al, 1997). This might partly explain the differences in food consumption between SES classes. On the other hand, poverty and low income also may restrict the ability to buy food on the basis of health and limit access to healthy food (Dowler et al, 1997; James et al, 1997).
In our study socio-economic differences were more evident on the food level than on the nutrient level. A higher SES was associated with a lower fat intake but the differences were rather small and not consistent when the contribution of alcohol to energy intake was taken into account. This is in agreement with other studies. If differences are found in the intake of energy-yielding nutrients it tends to be fat intake; those with high education tend to have a smaller intake. The finding that cheese is used more by people belonging to higher educational groups may level the differences in the intake of saturated fatty acids (Martinéz-Gonzaléz et al, 1998; Prätällä et al, 2000; Roos et al, 1999). There is not much evidence of lack of energy and macronutrients among lower SES groups. If differences in nutrient intake do exist then the lower SES groups seem to have lower levels of certain micronutrients. British data point to micronutrient and antioxidant intakes as the most likely nutritional influences on health inequalities (Davey Smith & Brunner, 1997). In our study we also found that a low SES was associated with a lower intake of most minerals and vitamins both in absolute amounts and expressed per MJ.
A potential limitation of our study and most nationwide population surveys is that the poor are usually not well presented, because homeless, unemployed or migrants not speaking the dominant language are difficult to reach and often specially tailored methods are needed to obtain accurate dietary data from these groups. In the Netherlands the nutrition (and health) situation among migrants is studied in separate surveys (Brussaard et al, 2001). In our study the number of subjects with a very low SES was relatively small and in contrast to the low SES group the intake often did not differ significantly from any of the other social classes. This hampers the interpretation of the results for the very low social class.
The higher prevalence of lifestyle factors such as obesity and breakfast skipping in lower SES groups is in line with other studies. In affluent societies obesity is associated with low SES, especially among women (Sobal & Stunkard, 1989). Results of two successive Whitehall cohorts in the UK illustrated that both overweight and obesity appeared to have become more strongly interlinked with SES in the past 30 y (Davey Smith & Brunner, 1997). A recent analysis based on Monica data in 26 population groups showed that lower education was associated with higher BMI in about half of the male and in almost all female population groups. Over the last 10 y (from 1979–1989 to 1989–1996) there has been a small shift towards stronger inverse associations too (Molarius, 1999).
Among men we did observe an overall decrease in energy intake over time which is surprising in view of increasing obesity. However the relationship between obesity and caloric intake is not simple and there is evidence that physical activity may also be decreasing. On the other hand, although a 2 day dietary record can be considered as valid dietary assessment instrument, subjects might become more aware of their eating pattern and could alter their behaviour. Moreover, the method is, like all survey methods that rely on self-reported behaviour, subject to problems of reporting error and bias. Under-reporting of dietary intake has been observed in many dietary surveys. Pryer et al (1997) found different rates of dietary under-reporting according to social class, with highest rates in the lower SES groups. Other studies found an association with high SES or high educational level (Lafay et al, 1997; Hirvonen et al, 1997). Several studies indicate that obesity is a major determinant of under-reporting. In our study the lowest mean ratio of energy intake to estimated basal metabolic rate was observed among women in the very low SES group (EI/BMR 1.34 for women with very low SES vs 1.40–1.42 for women other SES groups). Together with the high prevalence of obesity our findings indicate that the influence of under-reporting might be over-represented among women in the very low SES group. Among men, however, no differences were observed.
In studying differences in food consumption over time we tried to be aware of several pitfalls that might hamper interpretations, such as methodological issues, changes in demographic characteristics of the population, changes in the composition of food products and changes in food choices (Löwik et al, 1997). Because the food consumption methods used in the three surveys and the sampling frame were similar there was a fair level of standardization of research methods and procedures. However, the surveys differed with respect to non-response, survey starting date (for DNFCS-2), distribution of survey days over the week and the season. Analyses of variables regarding characteristics of respondents and non-respondents at recruitment from the existing panel and seasonal variation did not indicate that these factors have a strong influence on the results (Löwik et al, 1997, 1998). Since analyses of the first DNFCS revealed significant differences between weekend days (Saturday and Sunday) and mid-week days in the intake of energy and several nutrients, and since the distribution of survey days differed in the surveys according to SES, we included week–weekend effects in the statistical models.
In the DNFCS the household was used as a sampling frame. The data refer to the consumption inside and outside home. Due to clustering effects for the inside consumption the variance might be underestimated. This phenomenon is however present in all DNFCSs and in all SES groups. Therefore we assumed that this effect did not interfere with the conclusions in the present study.
The development of the Dutch population is characterized, among other things, by aging, a declining family size and an increase of average educational level. In the analyses age has been included as covariable in the statistical test. It was decided not to standardize for education since the interest of the present study was to compare the consumption of food and the intake of energy and nutrients according to SES at the time of the survey. Possible changes in the consumption of the total Dutch population, standardized for sociodemographic variables including education are being studied at this moment.
In all DNFCSs food composition data were converted into energy and nutrients using the most up-to-date version of the Dutch food composition table. However, over time new products were introduced by food industry, the composition of several existing products changed and results of new analyses of existing products became available. Therefore, differences in the composition of foods can be caused by real changes but also by artefactual changes due to improvements in the data quality, without any real changes in the product. Based on experiences with analyses in the second DNFCS (Hulshof et al, 1996) and other studies (Popkin et al, 1992, 1996; Guenther et al, 1994) specific food composition tables were created to enable valid comparisons over time (Beemster et al, 2000).
Food intake might also be affected by national campaigns aiming at changes in food habits and food choice. In 1991, in the Netherlands the so-called ‘Fat Watch’ campaign was launched by the Dutch Steering Committee for a Healthy Nutrition. This campaign has been executed in five consecutive years with an extra year in 1995. The campaign could have an effect on both the supply and demand of food products and is one of the possible explanations for the observed differences in fat intake (Löwik et al, 1997). It is also possible that respondents reported consuming fewer fatty foods because health messages had emphasized the desirability of lowering fat intake. Since the better-educated class is more likely to be conscious of desirable eating habits this might have affected particularly the reporting behaviour of the higher SES group. We cannot exclude such a response bias. On the other hand, however, we would expect to have also seen an increase in the reported vegetable and fruit intake, since messages to encourage the consumption of these foods were also prominent during that period. The opposite was observed. Therefore we think that, despite several limitations we were able to get insight into the dietary pattern of subjects in different socio-economic groups over time. A further exploration of possible causes, eg prices, marketing polices etc, of the observed changes in intake require a broader approach than is possible in the context of this paper. Furthermore, the implications for nutrition policy of the results discussed here will be evaluated by the Netherlands Health Council.
In conclusion, the findings suggest that the dietary intake in Dutch adults changed over a period of 10 y and these changes were found in all socio-economic classes. In general, dietary intake among subjects in higher SES groups tended to be closer to the recommendations of the Netherlands Food and Nutrition council and this phenomenon was quite stable over the last decade.
About this article
High protein intake along with paternal part-time employment is associated with higher body fat mass among girls from South China
European Journal of Nutrition (2018)