Background
An unhealthy diet is a leading risk factor for non-communicable disease and premature mortality [
1]. Diets, especially in countries with a market-based economy, typically include a high percentage ultra-processed foods [
2,
3]. Processed foods obtained from traditional methods of food processing such as fermentation to produce bread and cheese, the tinning of vegetables, and smoking of meats have been a part of people’s dietary habits for centuries, and contribute to the availability of safe, affordable and healthy diets. More recently, other methods of food processing have been introduced, like those referred to as ultra-processing. This includes industrial processes such as extrusion, pre-frying, and the addition of substances such as colour, stabilisers, artificial preservatives, flavours and flavour enhancers. Ultra-processing differs from traditional food processing in a number of ways, including its purpose, which is to create convenient, non-perishable food products that are ready-to-eat or heat, like frozen pizzas, chicken nuggets and instant sauces [
4]. While
traditionally processed foods are a part of a healthy diet,
ultra-processed foods (UPFs) are generally energy dense, high in added sugar, fat and salt, and low in fibre, and, therefore, diminish diet quality [
5‐
10]. Evidence from a randomized controlled trial suggests that diets including many ultra-processed foods lead to higher energy intake and higher body weight even after adjustment for sugar, fat, fibre, and other macronutrients content [
11]. Recent research suggests that this increase in energy intake observed on diets rich in UPFs may be due to a faster energy intake rate (kcal/min) of UPFs [
12]. Although causal mechanisms liking UPFs to health outcomes still need to be better understood, the consumption of UPFs has been associated with adverse health outcomes including obesity, metabolic syndrome, cancer, type II diabetes, cardiovascular diseases, and all-cause mortality [
13‐
19].
Research based on national household budget surveys from nineteen European countries has shown that 26.4% of the total purchased dietary energy comes from ultra-processed foods. This percentage differs widely between countries, ranging from 10.2% in Portugal to 50.4% in the UK [
16]. UPFs are generally heavily marketed and convenient, which contribute to their popularity and high intake levels [
3,
20‐
22]. Another possible explanation for the high consumption of UPFs is its widespread availability in current food environments. It could well be that larger availability of UPFs pushes purchasing behaviour and results in higher consumption. The relation between the geographical availability and intake of UPFs has, however, not been explored in detail as of yet.
It is important to identify the individual and environmental factors that are associated with UPFs consumption to inform policymaking and design interventions to reduce the purchase and consumption of UPFs. Several studies that have focused on individual-level determinants have shown that higher consumption of UPFs was associated with male sex, younger age, lower education, and being a smoker [
5,
6,
23]. A limited number of studies have also focused on environmental-level factors. Two Brazilian studies have shown that higher perceived availability of fruits and vegetables in the residential neighbourhood was associated with lower UPFs consumption [
24]. Obtaining groceries in supermarkets rather than in local food shops has also been linked to higher UPFs purchasing [
25]. In New Zealand supermarkets, UPFs were found to be the most prevalent type of packaged foods and showed a worse nutrient profile [
7]. Although several studies provide evidence that the food environment is associated with dietary intake [
26‐
28], research on the association between the objectively measured food environment and consumption of UPFs is lacking. Thus, in this study we want to address this gap by considering a broad range of food retailers that may be a source of UPFs purchases in the Netherlands, namely fast-food restaurants, full-service restaurants, supermarkets, and convenience stores.
Therefore, in this study we aim to (1) describe the pattern of UPFs consumption among older adults in the Netherlands and to explore how UPFs consumption relates to overall diet quality; (2) explore whether or not the availability of and proximity to different types of food retailers near the home is associated with consumption of UPFs. Since higher consumption of UPFs has been associated with lower educational attainment [
6,
23]; and patterns of food consumption and health outcomes may differ for urban and rural areas [
29,
30], we also aim to (3) explore if the associations we identify differ across levels of educational attainment and neighbourhood urbanisation.
Results
Table
2 shows the descriptive characteristics of the participants. The mean age was 70 (± 10 Standard Deviation (SD)) years and the vast majority of participants were female (80.3%). Most participants were lower educated (57.4%) and lived with a partner (66.6%). The mean BMI was 25.8 (± 4.5 SD) kg/m
2; and was lower among lowest consumers of UPFs: 25.0 kg/m
2 (± 4.1 SD), as compared to the highest consumers: 26.7 kg/m
2 (± 4.9 SD). Mean contribution of UPFs to total consumption was 18% (± 9% SD) in grams and 37% (± 11% SD) in calories. Median consumption of processed meat, savoury snacks, and soft drinks was higher among the highest consumers of UPFs, as compared to the lowest consumers. Median consumption of fruit and vegetables was lower among the highest consumers of UPFs. The mean score for the Dutch Healthy Diet index was 80 (± 17 SD), and those in the lowest tertile of consumption of UPFs, i.e., consuming less UPFs, scored higher (84.1; ± 17 SD) than those in the highest tertile of consumption of UPFs (73.7; ± 17 SD).
Table 2
Descriptive characteristics of the EPIC-NL participants (follow up wave 4)—total analytical sample and according to the level of intake of ultra-processed foods (UPFs)
Age | 69.9 (10.0) | 70.7 (9.2) | 68.8 (10.7) | 70.4 (9.0) | 69.5 (10.6) |
Sex (%) | | | | | |
Female | 80.3% | 87.4% | 71.8% | 83.6% | 75.9% |
Region of residence (%) | | | | | |
Amsterdam | 20.1% | 21.8% | 18.8% | 22.9% | 17.4% |
Maastricht | 23.6% | 14.1% | 35.0% | 15.6% | 32.1% |
Utrecht | 56.3% | 64.1% | 46.2% | 61.5% | 50.5% |
Educational attainment (%) | | | | | |
Lower | 57.4% | 48.5% | 66.4% | 48.0% | 67.7% |
Middle | 11.0% | 11.6% | 10.2% | 11.4% | 10.2% |
Higher | 31.6% | 40.0% | 23.4% | 40.6% | 22.1% |
Marital status (%) | | | | | |
Living with partner | 66.6% | 64.2% | 69.2% | 63.3% | 67.8% |
BMI (kg/m2)a | 25.8 (4.5) | 25.0 (4.1) | 26.7 (4.9) | 25.1 (4.1) | 26.6 (4.9) |
Total calories (kcal) | 1896 (640) | 1679 (521) | 2091 (721) | 1731 (554) | 2037 (695) |
Percentage of calories from UPFs | 37 (11) | 28 (8) | 46 (9) | 25 (5) | 49 (7) |
Total grams | 2535 (770) | 2599 (746) | 2508 (835) | 2510 (760) | 2521 (760) |
Percentage of grams from UPFs | 18 (9) | 10 (2) | 28 (9) | 12 (5) | 25 (10) |
g/d of fruitb | 163 (75–236) | 211 (102–245) | 115 (58–220) | 208 (99–244) | 115 (57–223) |
g/d of vegetablesb | 119 (70–173) | 135 (84–193) | 102 (50–150) | 135 (87–193) | 100 (50–144) |
g/d of processed meatb | 23 (9–25) | 16 (4–31) | 32 (15–54) | 17 (4–33) | 31 (14–53) |
g/d of savoury snacksb | 7 (0–19) | 3 (0–12) | 12 (3–27) | 3 (0–10) | 13 (3–28) |
g/d soft drinksb | 0 (0–29) | 0 (0–1) | 27 (0–166) | 0 (0–6) | 15 (0–100) |
Dutch Healthy Diet index 2015c | 80 (17) | 84 (17) | 74 (17) | 82 (17) | 76 (16) |
Figure
2 shows the distribution of participants across the categories of the accessibility measure (shortest network distance from home). Convenience stores and candy stores and cafes were less prevalent food retailers as only half of participants had these stores available within 1500 m from their home. Contrary, other food retailers were much more prevalent as 37% and 48% of participants had supermarkets and fast-food restaurants at less than 500 m from their home. Figure
3 shows the distribution of participants for the availability measures (counts of food retailers and kernel density), where a similar pattern was observed. For instance, considering the kernel density measure, while almost 80% of the participants had zero convenience stores within 500 m from their home, for fast-food restaurants this percentage was around 30% within 500 m, and as low as 4% within 1500 m. Knowledge about the fact that the distribution of participants across categories is different according to different food retailers may be relevant while choosing buffer sizes. As indicated in Figs.
2 and
3, to avoid large amounts of zeros in the first category, larger buffers are suggested for less prevalent food retailers such as convenience stores and cafes, while smaller buffers maybe used for more prevalent ones, such as restaurants, fast-food restaurants and supermarkets. The prevalence of food retailers in each study setting should be determined beforehand as different prevalence are likely to be found for different areas. More details regarding Figs.
2 and
3 can be found in Supplementary Table 1. Supplementary Table 2 shows the minimum and maximum count of food retailers in each tertile according to different distances in a network path from the participant’s home to each food retailer. Great variation was observed on the calculated street-network distances for different food retailer types. For instance, a minimum of 13 and maximum of 727 restaurants were encountered for a calculated 1500 m network path from home. In contrast, while travelling this distance a minimum of 5 and maximum of 30 supermarkets were encountered. Supplementary Table 3 shows the Spearman rank correlation coefficients for the three exposure measures. Correlations between Kernel density estimates and proximity to the closest food retailers had the lowest coefficients (bottom left of the matrix). In contrast, correlations between Kernel density estimates and counts within a network path had the highest coefficients (bottom middle of the matrix). Correlations of different measures of exposure to the same food retailer types were generally moderate to strong (
ρ > 0.6).
Table
3 shows the results from the linear regression analysis using proximity to closest food retailers as exposure measure and the percentage of grams and calories consumed per day from UPFs as outcomes. In general, participants that lived closer to any food retailer, as compared to those that lived further away, consumed a lower percentage of grams and calories from UPFs. Regression coefficients were generally small, with strongest associations observed for restaurants (
β = − 1.6%, 95%CI − 2.6; − 0.6), and supermarkets (
β = − 2.2%, 95%CI − 3.3; − 1.1) when using the percentage of calories from UPFs as the outcome. Table
4 shows the results of the linear regression analyses using the counts of food retailers across a network distance of 1000 m, and kernel density estimates within a 1000 m radius, as exposure measures. Similar to the analysis using the proximity measure, living in areas that had
any food retailers present was, in general, associated with a lower percentage of consumption in grams and calories from UPFs. More consistent trends with somewhat stronger coefficients were observed for counts of supermarkets and restaurants. Kernel density estimates, in turn, showed a slightly different pattern as associations were less often significant, effect sizes were smaller in some instances and the direction of the association for fast-food restaurants was positive, though no clear trend was observed across the categories.
Table 3
Regression coefficients (β) and 95% confidence intervals (95% CI) resulting from linear regression analyses with network distance to closest food retailers as exposure measure and the two outcomes: percentage of consumption in grams from ultra-processed food (UPFs) and percentage of consumption in kilocalories from UPFs (n = 8104)
Fast-food restaurant | | |
> 1500 m | Ref.a | Ref.a |
1001–1500 m | − 0.3 (− 1.4; 0.8) | − 1.4 (− 2.8; − 0.0) |
501–1000 m | − 0.5 (− 1.4; 0.4) | − 0.6 (− 1.7; 0.5) |
0–500 m | − 0.3 (− 1.2; 0.6) | − 0.8 (− 1.9; 0.4) |
Convenience stores | | |
> 1500 m | Ref. | Ref. |
1001–1500 m | 0.2 (− 0.5; 0.8) | − 0.1 (− 0.8; 0.7) |
501–1000 m | − 0.4 (− 0.9; 0.2) | − 0.8 (− 1.5; − 0.0) |
0–500 m | − 0.4 (− 1.1; 0.2) | − 1.1 (− 2.0; − 0.3) |
Restaurants | | |
> 1500 m | Ref.a | Ref.a |
1001–1500 m | − 0.1 (− 1.0; 0.7) | − 0.5 (− 1.6; 0.6) |
501–1000 m | − 0.8 (− 1.5; 0.0) | − 0.6 (− 1.5; 0.4) |
0–500 m | − 1.6 (− 2.4; − 0.8) | − 1.6 (− 2.6; − 0.6) |
Supermarket | | |
> 1500 m | Ref. | Ref. |
1001–1500 m | − 1.2 (− 2.2; − 0.3) | − 1.1 (− 2.3; 0.1) |
501–1000 m | − 1.8 (− 2.7; − 0.9) | − 1.8 (− 2.8; − 0.7) |
0–500 m | − 2.1 (− 3.0; − 1.2) | − 2.2 (− 3.3; − 1.1) |
Candy stores and cafés | | |
> 1500 m | Ref.a | Ref.a |
1001–1500 m | 0.1 (− 0.5; 0.7) | 0.4 (− 0.3; 1.2) |
501–1000 m | − 0.3 (− 0.8; 0.3) | − 0.3 (− 1.0; 0.3) |
0–500 m | − 1.2 (− 1.9; − 0.6) | − 1.3 (− 2.1; − 0.4) |
Table 4
Regression coefficients (β) and 95% confidence intervals (95% CI) resulting from linear regression analyses with counts of food retailers within a network distance of 1000 m and kernel density estimates as exposure measure and the two outcomes: percentage of consumption in grams from ultra-processed food (UPFs) and percentage of consumption in kilocalories from UPFs (n = 8104)
Fast-food restaurant | a | a, b | a | a |
Zero | Ref. | Ref. | Ref. | Ref. |
First tertile | − 0.3 (− 0.9; 0.4) | − 0.2 (− 1.0; 0.6) | 0.8 (− 0.2; 1.9) | 0.9 (− 0.3; 2.2) |
Second tertile | − 0.5 (− 1.3; 0.3) | 0.1 (− 0.8; 1.1) | 0.9 (− 0.2; 2.0) | 1.9 (0.5; 3.2) |
Third tertile | − 0.6 (− 1.6; 0.4) | − 0.7 (− 1.9; 0.5) | 0.3 (− 0.9; 1.6) | 0.9 (− 0.6; 2.5) |
Convenience stores | a | a | a | a |
Zero | Ref. | Ref. | Ref. | Ref. |
First tertile | 0.2 (− 0.4; 0.7) | − 0.2 (− 0.9; 0.5) | 0.0 (− 0.6; 0.6) | 0.1 (− 0.6; 0.8) |
Second tertile | − 0.3 (− 1.0; 0.4) | − 0.6 (− 1.5; 0.3) | − 0.1 (− 0.7; 0.5) | − 0.3 (− 1.0; 0.4) |
Third tertile | − 0.8 (− 1.7; 0.1) | − 1.7 (− 2.8; − 0.6) | − 0.3 (− 1.1; 0.4) | − 0.9 (− 1.8; − 0.1) |
Restaurants | | | a, b | a, b |
Zero | Ref. | Ref. | Ref. | Ref. |
First tertile | − 0.8 (− 1.4; − 0.3) | − 0.4 (− 1.1; 0.3) | − 0.2 (− 1.0; 0.6) | 0.5 (− 0.5; 1.5) |
Second tertile | − 1.1 (− 18; − 0.3) | − 1.0 (− 1.9; − 0.1) | − 0.7 (− 1.6; 0.2) | − 0.0 (− 1.1; 1.1) |
Third tertile | − 2.2 (− 3.0; − 1.3) | − 2.4 (− 3.4; − 1.4) | − 1.9 (− 2.9; − 1.0) | − 1.7 (− 3.0 − 0.5) |
Supermarkets | a,b | a,b | a | a |
Zero | Ref. | Ref. | Ref. | Ref. |
First tertile | − 1.0 (− 1.6; − 0.4) | − 1.1 (− 1.9; − 0.3) | − 0.6 (− 1.6; 0.3) | − 0.1 (− 1.2; 1.1) |
Second tertile | − 1.1 (− 1.8; − 0.4) | − 1.2 (− 2.1; − 0.3) | − 0.1 (− 1.2; 0.9) | 0.4 (− 0.9; 1.6) |
Third tertile | − 1.4 (− 2.2; − 0.5) | − 1.7 (− 2.7; − 0.7) | − 0.7 (− 1.8; 0.4) | − 0.3 (− 1.6; 1.1) |
Candy stores and cafés | | a | a, b | a, b |
Zero | Ref. | Ref. | Ref. | Ref. |
First tertile | − 0.2 (− 0.8; − 0.4) | − 0.2 (− 1.0; 0.5) | − 0.1 (− 0.7; 0.4) | 0.3 (− 0.4; 1.0) |
Second tertile | − 0.2 (− 1.0; 0.5) | − 0.1 (− 1.1; 0.8) | − 0.3 (− 0.9; 0.3) | 0.1 (− 0.6; 0.9) |
Third tertile | − 0.7 (− 1.5; 0.1) | − 1.7 (− 2.1; − 0.2) | − 0.5 (− 1.2; 0.2) | − 0.6 (− 1.5; 0.3) |
A sensitivity analysis using the counts of food retailers and kernel density estimates within a distance of 500 and 1500 m showed a similar direction and strength of associations to the main analysis (Supplementary Table 4). Interaction terms with educational attainment were not significant for the analysis with proximity as exposure measure. However, a significant interaction was found with urbanization level in the models including proximity to restaurants and fast-food restaurants (Table
3). Significant interaction terms with educational attainment and urbanisation were also found in some analysis with counts and kernel density estimates (Table
4). However, analyses stratified for both urbanisation and education were mostly non-significant or similar to the general analysis (Supplementary Tables 5 to 9).
Discussion
In this study, we describe patterns of UPFs consumption among a predominantly elderly and female Dutch population, and explored how the objectively measured residential food environment was associated with consumption of UPFs. Based on descriptive statistics, we found that participants that consume more UPFs were younger, more likely to be male, and lower educated. Furthermore, these individuals had a higher BMI, higher energy intake, consumed less fruits and vegetables, and more processed meats, savoury snacks and soft drinks. Closer proximity and larger availability to any type of food retailer was, in most instances, found to be associated with a lower consumption of UPFs, with somewhat stronger associations with exposure to restaurants and supermarkets. None of the food environment exposure variables were significantly associated with higher consumption of UPFs.
The Dutch dietary guidelines, whose adherence is measured by the Dutch Healthy Diet index, recommends the avoidance of foods such as processed meats and sugar-sweetened beverages. Therefore, we did expect that UPFs consumption and the Dutch Healthy Diet index would be inversely associated. However, it could also have been the case that those consuming UPFs such as sugar-sweetened beverages and ultra-processed meats, on the whole did have an adequate consumption of vegetables, whole grains and dairy thereby resulting in a relatively healthy overall diet quality. Nonetheless, descriptive statistics indicated that those who consumed diets containing a high percentage of UPFs had a lower score on the Dutch Healthy Diet index and consumed less fruit and vegetables, suggesting an overall lower diet quality. This finding is in line with previous research conducted in other countries [
5,
6,
23], although what mechanism explains this association requires further investigation [
12]. The proportion of UPFs in the diet in this population is comparable to that of populations from other European countries including France, Austria, and Norway [
6,
16], but is lower than that of populations in the UK, Germany, and the USA [
16,
23,
46]. However, it needs to be noted that the EPIC-NL cohort is, on average, an older and mostly female population. The dietary contribution of UPFs in younger adults, or in a more general adult population, is likely to be higher, as has been demonstrated by previous research [
5,
6,
23].
We presumed that greater accessibility and availability of the food retailers included in the analysis might be a potential source of UPFs and would, therefore, be associated with higher UPFs consumption. However, the associations found were largely counter-intuitive, especially for food retailers such as candy stores and fast-food restaurants. Regarding supermarkets and restaurants, it could well be that our hypotheses were wrong and that the association for higher exposure to these food retailers and lower ultra-processed foods consumption would be the true association. Besides, the fact that the effect sizes found were generally small, could be an indication of a general null finding. However, a more likely explanation is that, despite the fact that we accounted for neighbourhood characteristics such as urbanisation and the presence of other food retailers, unmeasured/residual confounding could still have played a role [
47,
48].
Inconsistent associations between the food environment and dietary intake have been reported previously [
33] and have been attributed to factors such as the general use of low-quality instruments for dietary assessment, and oversimplification of the definition of exposure. That is an analysis that is restricted to the residential environment, which usually consider only simple measures that do not accurately reflect individuals’ exposure, and does not account for the broader food environment [
33,
44,
49‐
51]. In this study, we attempted to avoid these potential pitfalls as much as possible. For instance, we used more comprehensive dietary data (i.e., data from a validated FFQ), employed different measures to define exposure, and used different distance categories. Because one cannot always determine in advance what measure would better capture different dimensions of exposure, we used several measures [
23,
50]. Nonetheless, our results remained counterintuitive, thus, issues regarding our definitions of exposure do not seem to account for the unexpected findings of this study.
Correlation analysis showed that coefficients for different measures of exposure to the same food retailers were generally very strong, which could indicate that the various measures used represent exposure in the same way. However, when looking across different food retailers, we observed that kernel density estimates and counts within a network path correlated more strongly with each other than kernel density and distance to the closest food retailer. Indeed, Kernel density and counts within a street-network path are more complex measures than the closest distance measure, as the latter only takes into account proximity to one food retailer.
This study has both strengths and limitations. The fact that we only considered the residential environment may be a limitation [
52]. However, the EPIC-NL cohort consists of a predominantly older female population, many of whom may be housewives or retired individuals. The residential setting is, therefore, more likely to be representative of exposure to their food environment than it would be for a younger population with greater mobility. As demonstrated by Kirkpatrick et al.’s systematic review, most studies that analyse the relationship between food environment and diet make use of short questionnaires for dietary assessment, which introduces a considerable source of bias in terms of assessing intake, thereby affecting the results of subsequent analyses [
51]. In this context, the use of a comprehensive FFQ to obtain nutritional data is a strength of our study. However, any self-reported dietary data is still prone to bias (e.g., social desirability) and to both under and over reporting of dietary intake. The fact that we use different exposure measures of the food environment, accounting for both constructs of proximity and availability, and applied them to different types of food retailers, is also a strength. As has been suggested by previous reviews, more multi-method studies are needed to build a strong evidence base that identifies which measures apply to various contexts [
35,
53]. Additional strengths of this study include the large sample size and the innovative aspects of the study, including being the first to report the consumption patterns of UPFs in the Netherlands, and the first to analyse the relationship with the objectively measured food environment.
In conclusion, in the present study among predominantly elderly and female participants, we found that those who consumed more UPFs had higher total energy intake and had a poorer overall diet quality. We did not find evidence that the accessibility or availability of five types of food retailers that offer many opportunities to purchase UPFs was associated with higher UPFs consumption in this population. On the contrary, using various measures of exposure to the food environment, we found evidence that exposure to some types of food retailers, especially restaurants and supermarkets, was consistently associated with somewhat lower consumption of UPFs.