Expert- and public-led approaches

In order to address the two aspects ‘health’ and ‘social acceptance’ simultaneously, the expert- and public-led approaches make use of nutrition experts and focus groups. In the expert-led approach, nutrition experts compile healthy food baskets based on dietary recommendations. Where appropriate, the acceptability of the food baskets is verified through focus group discussions^{(e.g.(Reference Carrillo-Álvarez, Cussó-Parcerisas and Riera-Romaní8–Reference Storms and Deeming12))}. In the public-led approach, the process is in reverse order. First, menu plans are created through focus groups. Then, these are adjusted by nutrition experts for their compliance with the dietary recommendations^{(e.g.(Reference Bradshaw, Middleton and Davis14,Reference Mac Mahon, Thornton and Deeming15))} .

The expert-led approach was applied, for example, in a pilot project called ImPRovE-Approach, where European Reference Budgets were calculated^{(Reference Goedemé, Storms and Penne11)}, that is, in a first step nutrition experts complied food baskets, which were adjusted by focus groups in a second step. Although the approach of this project was generally considered promising, some limitations related to the expert-led approach and focus group discussions were mentioned in the literature. The composition of food baskets by experts is generally criticised for its normative character, as it could be influenced by the experts’ attitudes and opinions^{(Reference Austgulen, Elling Borgeraas and Deeming16)}. This is particularly critical because experts usually have a higher education and income and therefore may have a different perception of needs^{(Reference Austgulen, Elling Borgeraas and Deeming16)}. In the ImPRovE-project, the dietary recommendations used by experts also varied widely between countries, making it difficult to compare results across countries^{(Reference Goedemé, Storms and Penne11)}.

Focus group discussions are considered as an important tool for determining a socially acceptable minimum standards of living, as members of society determine what is essential for life. However, as a main limitation of the focus group discussions their low representativeness is mentioned^{(Reference Deeming17)}. Due to the limited number of people participating in the focus groups, the results do not necessarily reflect general public opinion^{(Reference Deeming17)} and therefore have a low robustness^{(Reference Mäkinen18)}. Furthermore, it is difficult to find a common consensus in focus group discussions^{(Reference Preuße19)}. This was shown, for example, by a Dutch study in which Reference Budgets differed considerably. For a couple with children, the focus groups calculated for the food sector monthly budgets between €491 and €730^{(Reference Hoff, Soede and Vrooman20)}.

Survey-led approaches

While the previously described approaches, when used in combination, consider both aspects ‘health’ and ‘public acceptance’, the survey-led approach usually focusses on the second aspect. Here, survey data are used to investigate households’ consumption patterns either in terms of quantities or expenditures. Often, expenditure data of low-income households are used to directly derive Reference Budgets. However, this approach is criticised because of its circularity: As the expenditure behaviour of low-income households is probably influenced by budget restrictions, an adequate living standard cannot be guaranteed^{(Reference Deeming and Deeming7)}. This is also of particular relevance for food, as it has been empirical observed that an unhealthier diet is associated with lower costs^{(Reference Darmon, Briend and Drewnowski21–Reference Drewnowski, Darmon and Brien24)} and is increasingly chosen by households with lower socio-economic status^{(Reference Darmon and Drewnowski23)}. Another limitation of expenditure data is that it does not account for the variability in prices arising from differences in brand, quality or point of sale (e.g. discount supermarket v. specialty store). Therefore, when deriving a Food Reference Budget based on the food expenditure of lower income groups, it cannot be guaranteed that the budget for a healthy diet is sufficiently high.

Linear programming – a survey-led approach

Against the background of the limitations of the methods used so far, Godemé et al. proposed to apply the Linear Programming (LP) method for the determination of Food Reference Budgets^{(Reference Goedemé, Storms and Penne11)}. LP is a scientifically recognised method that has already been used in international nutritional studies ^{(e.g.(Reference Conforti and D´Amicis25–Reference Masset, Monsivais and Maillot28))}. However, it has not yet been applied in the context of determining Reference Budgets. LP can be characterised as a form of the survey led approach, as in a first step, representative food consumption patterns can be derived from survey data, which in a second step can be adjusted using LP to integrate the health aspect. With LP, it is possible to transform an observed diet into a healthy one (for developing healthy food baskets) in such a way that the deviation from the current observed consumption is minimised while complying with restrictions (e.g. nutrient guidelines). Thus, the optimised food baskets are healthy and as close as possible to the observed consumption. After optimisation the healthy food baskets can be evaluated with prices to determine the Food Reference Budgets. In this study, we want to use LP for the determination of Food Reference Budgets. As a basis for the calculations, we need various datasets that are described below.

Consumption survey data

To identify the country-specific food consumption patterns, we made use of the EU Household Budget Survey (EU-HSB). This dataset contains information on expenditures and quantities and the foods groups are recorded in the European classification of consumption by purpose (ECOICOP). In the detailed 5-digit ECOICOP classification, the categorisation comprises sixty-two food groups, which are predominantly commodity based. Data on expenditure and quantities for out-of-home consumption (e.g. in restaurants or canteens) were not available. The dataset therefore reflects food consumption at home. The EU-HBS are compiled every 5 years based on national HBS surveys. These include one or more interviews/diaries depending on the country. Detailed information on the survey can be found in an Eurostat-report⁽²⁹⁾. The data used in this study are from 2015 (or 2010 for Malta, Portugal and Slovenia).

As consumption data for Germany were not included in the EU-HSB, the German Income and Consumption Survey (EVS) was used in addition. The EVS is part of the official statistics and is conducted every 5 years as a household survey. The participating households are selected based on a quota plan. Part of the overall survey consists of a survey on the purchases of food, beverages and tobacco products, in which households record their purchased quantities and expenditures over a 1-month period. At the time of the study’s preparation, the most recent survey available was the one from 2013 which covered 11 416 households with different household sizes and compositions. However, as nutrient and food recommendations (here European Food Safety Authority (EFSA) recommendations) are only given for single individuals, we made use of the OECD-modified equivalence scale to derive individual consumption quantities from the household consumption. Therefore, total household food quantities were divided by the respective weights of the equivalence scale.

While food and beverages are classified into 206 different groups in the German EVS consumption data, they were divided into sixty-two groups in the European Household Budget Survey. These 62 groups correspond to the 5-digit ECOICOP which is a commonly known classification system developed by the United Nations Statistics Division⁽³⁰⁾. In order to standardise the consumption datasets (EU-HBS and EVS), the 206 food and beverage groups of the German EVS were assigned to the sixty-two ECOICOP groups. However, since there were no foods in the EVS that could be assigned to the ECOICOP group ‘other tubers/ products of tuber vegetables’, consumption data were only available for 61 of the 62 ECOICOP groups for Germany. Furthermore, as two of the sixty-two ECOICOP groups, namely alcoholic beverages and baby food, were not of interest for the calculation of Food Reference Budgets, they were excluded from the analysis. In the end, the calculations were based on fifty-nine food groups for Germany and sixty for the other EU countries.

Nutrient data

To obtain information on nutrient contents, the German food composition database (Bundeslebensmittelschlüssel Version 3·01, BLS), which gives information on nutritional values for 14 814 foods available in the German market, was used. Each of the sixty food groups was assigned to the corresponding food groups of the nutrient database. If a clear assignment to a single food group was not possible, the average nutrient values of two or more food groups were used.

Price data

To price the food basket for Germany, data collected for the calculation of the German Consumer Price Index 2018 were used. The prices of the German Consumer Price Index are collected monthly in supermarkets and discounters throughout the country. Over the year 2018, a total of 1 199 968 individual food prices were gathered, which are summarised by the Federal Statistical Office in 170 groups. These 170 groups were assigned by us to the sixty ECOICOP groups. For each ECOICOP group, an average price per kilogram was calculated based on the available German Consumer Price Index prices of the assigned items. As prices within food groups can vary depending on factors such as brand or place of purchase (e.g. discount stores v. supermarkets), we calculated the Food Reference Budgets using several lower price percentiles (20th, 30th, 40th and 50th) to capture a realistic range of possible food costs.

In order to also get prices for the other EU countries Eurostat-Price level indices from 2018 were used. The price level indices reflect the relative price level of countries in relation to another country. The indices are subdivided in the ECOICOP 5-digit classification and were therefore directly available in the classification required for this study. The price level indices are available for all EU countries except Croatia. Therefore, the calculation of Food Reference Budgets based on the optimised food baskets could be calculated for twenty-five instead of twenty-six countries.

Income data

To assess the affordability of healthy food in each country, we determined the cost of the healthy optimised food baskets as a proportion of income. Specifically, we used the median net equivalised income for the 25–54 age group provided by Eurostat (Eurostat 2022)⁽³¹⁾. This indicator enables a cross-country comparison of disposable household income. By relating the cost of optimised healthy food baskets to income, we were able to assess the relative financial accessibility of a healthy diet.

Objective function

The objective function of the LP includes decision variables whose values are the outcome of the model. In this study, the outcome are optimal food quantities resulting from a minimisation of deviation to a current observed diet (objective function) while meeting constraints such as nutrient recommendations. The linear objective function Y has the form $Y = {a_0} + {a_1} \times {X_1} + {a_2} \times {X_2} + \ldots {a_n} \times {X_n}$ , where X ₁ , …X _n , are the decision variables and a ₀ , …a _n are constant.

The aim in this study is to build the objective function in such a way that it minimises the absolute deviation between a current observed diet and the optimised healthy diet calculated by the LP. As a starting point for optimisation, first the current observed mean consumption C _i (with i = 1 to n, where n is the number of food groups) was calculated based on the consumption data. However, a calculation of the total consumed energy content of the food groups showed that the amount of energy in some countries deviates significantly from the recommendations. Therefore, the percentage energy structure between the food groups was used to proportionally adjust the food quantities so that the energy content of the total food consumption corresponds to the EFSA recommended calorie intake. This energy scaled consumption (sC _i ) represents the countries’ current observed diet at the level of the recommended energy intake and is the starting point for optimisation. The scaled consumption is defined as follows:

$$s{C_i} = {{{rE}}\over{{cE}}} \times {C_i}$$

C _i observed mean consumption of i

sC _i scaled mean consumption of i

rE recommend energy intake

cE observed mean energy intake of a country

with i = 1 to n, where n is the number of food groups

The absolute deviation is defined as the difference between the quantities of each food item X _i calculated by the LP (with i = 1…n, where n is the number of food groups) and the observed scaled consumption sC _i , divided by sC _i . Dividing by the quantity sC _i aims to standardise the differences between the food groups. The approach follows Darmon et al. ^{(Reference Darmon, Ferguson and Briend26)} and is based on the principle of the goal programming approach^{(Reference Anderson and Earle32)}. The function of the ‘total departure of mean food intake’ is thus as follows:

$$\begin{align}TDMI = &{{ABS\left( {s{C_1} - {X_1}} \right)}\over{s{C_1}}} + {{ABS\left( {s{C_2} - {X_2}} \right)}\over{s{C_2}}} + \ldots \\&+ {{ABS\left( {s{C_n} - {X_n}} \right)}\over{s{C_n}}}\end{align}$$

TDMI is the total departure of mean food intake

X _i decision variable: optimised amount of i

ABS absolute value

By introducing further decision variables and restrictions, this non-linear objective function can be transformed into a linear objective function of X _n ^{(described in detail in (Reference Darmon, Ferguson and Briend26))}. With the help of this transformation, the function can be solved as a Linear Programming problem using OpenSolver in Excel.

Restrictions

The restrictions used in this study are divided into two parts. On the one hand, health-related restrictions on nutrient and food recommendations were considered to model a healthy diet. On the other hand, consumption restrictions were included to ensure that all food groups consumed by the population, including those considered rather unhealthy, such as sweets, are included in the optimised diet.

Health-related-restrictions

The food baskets in this study were calculated for adult men and women with medium activity levels. To account for nutrient-related constrains, the reference values for adequate nutrient intake of the EFSA were chosen, using a physical activity level of 1·4 corresponding to sedentary work with little or no strenuous leisure activities⁽³³⁾. To ensure age-appropriate nutrient intake for adults, the reference values for 25–49-year-olds were used. In addition to lower limits, the EFSA also gives upper limits for some nutrients that should only be consumed to a limited extent⁽³⁴⁾. The lower and upper limits used as restrictions in this study are listed in Table 1.

Table 1.

Nutrition-related constraints used in the linear programming model

Source for lower and upper recommendation:⁽³³⁾ and⁽³⁴⁾.

Source Cholesterol:⁽³⁵⁾.

* Recommmendation made in % of energy intake, in g/kg bodyweight or in mg/MJ are converted in g or mg considering the following assumptions: body weight: 82 kg (man), 66 kg (women)⁽³⁶⁾; PAL: 1·4 (man and woman); energy intake: 9·4 MJ (man), 7·55 MJ (woman)⁽³³⁾; Energy contents of energy-providing nutrients according to EC Directive (90/496/EEC) were used.

† As Vitamin D can be ingested through food as well as produced by the body through endogenous synthesis a lower limit of 5 µg per day was used.

In addition to the reference values for nutrient intake, there are also some food-related recommendations that are important for a healthy diet. This applies to fruits, vegetables, water, salt, sugar and caffeine. The food-related recommendations used in this study, their sources and more detailed explanations can be found in Table 2.

Table 2.

Food-related constraints used in the linear programming model

* Source: The recommendation that fruits and vegetables should provide between 7 and 14 % of energy intake⁽³⁷⁾ were converted in g/d. For this following information were used: Recommendation that fruits and vegetables should be consumed in the following proportions: 2 portions of fruits and 3 portions of vegetables⁽³⁸⁾; Average calorie content of fruits/vegetable: 50 kcal/ 20 kcal⁽³⁷⁾, which, considering the recommended proportions, corresponds to an average caloric content of 32 kcal for vegetables and fruits.

† Source:⁽³⁹⁾.

‡ Source: The recommendation of reducing the intake of free sugars to less than 10 % of total energy intake⁽⁴⁰⁾ were converted in g.

^§ Source: The recommendation includes water from beverages and from food⁽³³⁾. To avoid beverages being represented in disproportionate amounts, an upper limit was set that allows for a maximum deviation of 30 % above the recommended water amount.

^|| Source: As caffeine is not included in the nutrient database, the limitation of caffeine intake had to be done by limiting the amounts of coffee and tea. Up to 400 mg of caffeine per day is regarded as safe⁽⁴¹⁾. With a caffeine content of less than 100 mg per cup of coffee (200 ml)⁽⁴¹⁾, 4 cups are within the safe intake level. Since the recommendations for caffeine intake are based on body weight, the limit for women was scaled accordingly based on body weight.