A methodological framework for deriving the German food-based dietary guidelines 2024: food groups, nutrient goals, and objective functions

Anne Carolin Schäfer, Heiner Boeing, Rozenn Gazan, Johanna Conrad, Kurt Gedrich, Christina Breidenassel, Hans Hauner, Anja Kroke, Jakob Linseisen, Stefan Lorkowski, Ute Nöthlings, Margrit Richter, Lukas Schwingshackl, Florent Vieux, Bernhard Watzl
doi: https://doi.org/10.1101/2024.10.24.24316069

Abstract

Background For a growing number of food-based dietary guidelines (FBDGs), diet optimization is the tool of choice to account for the complex demands of healthy and sustainable diets. However, decisions about such optimization models’ parameters are rarely reported nor systematically studied.

Objectives The objectives were to develop a framework for (i) the formulation of decision variables based on a hierarchical food classification system; (ii) the mathematical form of the objective function; and (iii) approaches to incorporate nutrient goals.

Methods To answer objective (i), food groups from FoodEx2 levels 3-7 were applied as decision variables in a model using acceptability constraints (5th and 95th percentile for food intakes of German adults (n=10,419)) and minimizing the deviation from the average observed dietary intakes. Building upon, to answer objectives (ii) and (iii), twelve models were run using decision variables from FoodEx2 level 3 (n=255), applying either a linear or squared and a relative or absolute way to deviate from observed dietary intakes, and three different lists of nutrient goals (allNUT-DRV, incorporating all nutrient goals; modNUT-DRV excluding nutrients with limited data quality; modNUT-AR using average requirements where applicable instead of recommended intakes).

Results FoodEx2 food groups proved suitable as diet optimization decision variables. Regarding deviation, the largest differences were between the four different objective function types, e.g. in the linear-relative modNUT-DRV model, 46 food groups of the observed diet were changed to reach the model’s goal, in linear-absolute 78 food groups, squared-relative 167, and squared-absolute 248. The nutrient goals were fulfilled in all models, but the number of binding nutrient constraints was highest in the linear-relative models (e.g. allNUT-DRV: 11 vs. 7 in linear-absolute).

Conclusion Considering the various possibilities to operationalize dietary aspects in an optimization model, this study offers valuable contributions to a framework for developing FBDGs via diet optimization.

1 Introduction

Food-based dietary guidelines (FBDGs) are typically developed to advice the public by which food intake dietary reference values (DRVs) can be met, and more recently, to reduce their risk of diet-related non-communicable diseases [1]. Nowadays, more advanced FBDGs integrate additional dimensions such as ecological sustainability [2], which increases the complexity of defining optimal intakes of different food groups [3,4].

The most holistic approach to manage the complex task of evaluating multiple dimensions of a sustainable diet is diet optimization, also referred to as diet modeling or linear/quadratic (squared) programming [5,6]. It aims to find the optimal quantitative combination of food groups (the decision variables), that may have both conflicting and/or complementary features, and that fulfills a set of constraints while minimizing or maximizing an objective function. Such a tool has been successfully applied in the context of FBDG development, for example in Australia, France, and the Netherlands [4,7–10].

However, each parameter in the optimization model can be represented in different (mathematical) options [11]. The food groups considered for FBDGs, and thus as decision variables for mathematical optimization, are often determined on an ad hoc basis [1,3,12,13]. The European Food Safety Authority (EFSA) has set up a detailed food classification system, FoodEx2, which is used to analyze and report survey data for European countries [14–17], what makes its use as decision variables favorable. However, it is unclear how such hierarchically organized food coding systems perform in optimization models. The objective function defines indicator(s), which should be minimized or maximized. In the literature using diet optimization, the most frequent indicator is the observed diet from which the deviation is minimized and thereby maximizes cultural acceptability of the model’s solution [18–20]. This fulfills the requirement that FBDGs should take the habitual diet of the population into account [1]. Constraints define the solution space in which the model can operate. When designing nutritionally adequate diets using diet optimization, they are usually set to fulfill nutrient recommendations using nationally available DRVs and to adhere to acceptability limits, where food group quantities are limited to realistic consumption levels observed in the population, e.g., the 95th percentile of consumption [5,8].

Although mathematical optimization has become an established method in nutritional sciences, hardly any systematic methodological analyses has been reported. With the goal of updating and expanding the German FBDG methodology, we undertook thorough analyses of these methodological choices. The present paper aimed to describe the use of FoodEx2 as decision variables for optimization models, the implication of the objective function’s mathematical form taking the observed dietary intake from the latest German nutrition survey as example, and the impact of different choices of nutrient goals based on DRVs on optimization results.

2 Methods

2.1 Decision variables

The decision variables are the variables that will be optimized. Optimization models for FBDGs typically focus on observed dietary intakes for a list of food groups as decision variables. For Germany, the food classification of EFSA, FoodEx2, provides internationally comparable food intake data and was therefore selected as the basis of our decision variables (version MTX 12.1, Exposure hierarchy [21]). Following a parent–child hierarchy, the FoodEx2 food groups are ordered into seven levels, with level 1 being the most aggregated (e.g., “Fruit and fruit products”) and the lower levels being more detailed (e.g., “Pome fruits” on level 3) (see Figure 1). Level 7 is the level at which intake data from surveys such as the most recent German National Nutrition Survey II [16,22] had been initially coded before being further aggregated for communication and comparisons. FoodEx2 facets, which further describe attributes of a food group, e.g., processing information, were not considered in this study.

Figure 1:
Figure 1: Example of the hierarchical structure of FoodEx2 showing food groups from level 1 “Fruit and fruit products”

However, FoodEx2 does not allow a clear distinction between whole grain and other grain products without the use of FoodEx2 facets. To work with well-defined food groups for whole grains, the food groups that belong to level 1 “Grains and grain-based products” were relabeled to obtain the food groups “whole grain” and “refined grain” as displayed in S1 Table. The relabeling was conducted by duplicating the initial “Grains and grain-based products” food group to create two different categories (“Whole grains” and “Refined grains”) that still adhered to the parent–child hierarchy. All initial food groups at level 7 were categorized as whole grain or refined grain and then aggregated until level 1.

Processed meat is also not clearly defined in the hierarchy of FoodEx2, as several level 2 food groups of the parent group “Meat and meat products” are a mix of processed meat products and unprocessed meat. Here, a new parent food group called “Processed meat” was established and the respective FoodEx2 food groups were assigned to this group. To differentiate between poultry and red meat for the food group “Mammals and birds meat” (level 3, food code A0EYH), the food code was duplicated, reassigned to either red meat or poultry, and the mean intake split between the two new variables. The percentiles used as acceptability constraints (chapter 2.2) were kept as observed.

The FoodEx2 level 1 groups “Other ingredients” and “Food products for young population” were excluded for their irrelevancy and low intakes by adults. After these adjustments, the number of food groups, and therefore decision variables, was 927 at level 7, which were aggregated into 926 at level 6, 857 at level 5, 593 at level 4, 255 at level 3, and 83 at level 2.

2.1.1 Reporting the whole diet

Considering that FoodEx2 level 2 already has 83 different food groups, the optimization results would be too detailed for establishing and communicating FBDGs. Thus, we decided that the reporting of results from our optimization models should be condensed into a defined list of food groups. This list was agreed upon by the working group of the German Nutrition Society responsible for the development of the scientific basis for the German FBDG (S2 Table). This list included the following groups: water, coffee and tea, vegetables, fruits, fruit and vegetable juices, legumes, nuts and seeds, potatoes, grain (products) and wholegrain (products) thereof, milk and dairy products, eggs, fish and seafood, poultry, red meat, processed meat, vegetable oils, and spreadable fats.

These 18 FBDG food groups were matched to their most aggregated respective FoodEx2 food groups. Eight of the 18 food groups matched FoodEx2 level 1 food groups, six matched level 2 food groups, and four matched level 3 food groups (Table 1).

View this table:
Table 1: Food-based dietary guidelines (FBDG) food groups and respective FoodEx2 food group names and codes.

Many FBDGs aim to limit the intake of so-called discretionary foods [1], such as sweets and sugar-sweetened beverages. These food groups contribute to total energy intake and the intake of nutrients that should be limited, such as free sugars and saturated fatty acids. These heterogeneous and unfavorable food groups were reported as the summary of their energy share in addition to the 18 FBDG food groups. Discretionary food groups and their respective FoodEx2 food groups are displayed in Table 2.

View this table:
Table 2: Discretionary food groups according to FoodEx2 food group names and codes.

2.1.2 Selection of the appropriate FoodEx2 level for optimization

To determine which level of FoodEx2 should be selected as the source of decision variables for the optimization of the German FBDG, we concluded that the level should provide sufficient details to create a specific result for each food group reported in the FBDG. For example, if the FBDG should report the optimal consumption quantity of vegetable oils, as distinguished from other fats at level 3 (Table 1), level 3 of FoodEx2 would need to be selected as the source of decision variables. We found that FoodEx2 level 3 (255 decision variables) provided a list of food groups with sufficient detail for reporting in the German FBDG (Table 1). The list of decision variables at level 3 is shown in S3 Table.

2.1.3 Linking food groups to consumption and nutrient composition data

In the EFSA Comprehensive European Food Consumption Database, the FoodEx2 food codes are already linked to dietary intake data from various national surveys [15,16]. For our purpose, we retrieved data for adults (18–65 years) in Germany. The German intake data is based on two non-consecutive 24-hour dietary recalls from 10,419 men and women from the German National Nutrition Survey II (NVS II) (2005-2007) [24]. The data was weighted as described in Heuer et al. for age, sex, residential area, and other socioeconomic factors to represent the adult German population [25]. For each FoodEx2 food group and level, the mean intake and distribution percentiles in g/d were calculated, either for those individuals who consume the food group (consumers only) or among all individuals.

Information on nutrient and energy content was obtained from the German Nutrient Database (Bundeslebensmittelschlüssel (BLS) version 3.02 [26]). The BLS uses a different food group classification system than FoodEx2. To match the intake data with the national nutrient database, a matching of 1,288 foods from the BLS to FoodEx2 food groups, mainly on level 7, was applied (personal communication of Katja Sandfuchs from the Max Rubner-Institut, Karlsruhe, Germany). As the BLS lacks data for free sugars, this information was completed using mean values from 80 generic food categories of the LEBTAB database [27] that were matched to BLS food groups.

The nutrient values for all food groups (decision variables) were calculated following the parent–child hierarchy of FoodEx2 (see Figure 1). For a specific food group, the nutritional content was either the nutritional content of a BLS food if a direct matching existed, or an average nutritional content of its related food groups according to the FoodEx2 hierarchy that had a matching with the BLS database, weighted by intake data, as described by Gazan et al. [28].

2.2 Acceptability constraints

Acceptability constraints frame the solution space of the optimization model to ensure that optimized food intakes remain within the range observed in the target population. They were defined as follows: the 5th (minimum constraint) and 95th percentiles (maximum constraint) of the observed intake for all individuals for each food group at level 1 (see S4 Table) and, additionally, for each food group on level 3, the 5th (minimum constraint) among all individuals and 95th percentiles (maximum constraint) for consumers only.

For “Composite dishes” and “Seasoning and sauces,” which are part of the discretionary food groups, we made exceptions from the aforementioned rules. These food groups are not clearly described (e.g. potato-based dishes), but supply various nutrients within one decision variable, posing a very attractive group for the linear models which strive to achieve the best solution by changing least decision variables. Therefore, the optimized intake for these food groups could not exceed the corresponding observed mean intakes on level 3. Because the energy goal of the optimized diet is lower than the observed diet, the observed intakes used as upper acceptability constraints were matched to the energy goal of the optimized diet.

2.3 Objective function and nutrient goals

2.3.1 Different implementations of deviation from the observed diet

Four different mathematical implementations for minimizing deviation from observed dietary intakes in the objective function were investigated: linear as a percentage from the observed diet (linear-relative), linear in absolute quantities (linear-absolute), squared differences of the percentage from the observed diet (squared-relative), and squared differences of the absolute quantities (squared-absolute).

The mathematical formulas were as follows: Embedded Image where Dev is the deviation from the observed diet composed of n food groups, with Embedded Image the observed quantity of food group i and Embedded Image the optimized quantity of the same food group.

2.3.2 Three approaches to incorporate nutrient goals

The nutrient goals were based on the dietary reference values (DRVs) for Germany [29] and complemented by EFSA’s tolerable upper intake levels for nutrients [30]. DRVs are classified as either recommended intakes (RI), estimated values, or guiding values. Nutrients with an RI cover the nutrient requirements of 97.5% of the population [29]. The average requirement (AR) for these nutrients covers the requirements of 50% of the population. Three different lists of nutrient goals were defined by setting lower bounds, upper bounds, or specific target values. If not differently specified, reference values for adults (18–65 years, normal weight with moderate physical activity level (PAL) 1.4) were taken and subsequently weighted according to the proportion of sex and distribution of age groups in the German population (S5 Table).

The first list applied all nutrient goals (DRVs und EFSA upper intake levels) and where available the RI (allNUT-DRV). Selenium, chromium, and molybdenum were not included because there is no data in the nutrient database used (BLS version 3.02). Vitamin D was not included because the requirement is not to be covered by diet, but provided by the endogenous synthesis through exposure to sunlight or by supplements.

The second list (modNUT-DRV) built upon allNUT-DRV, but was modified in three different ways: (i) iodine, fluoride, copper, and manganese were not applied due to limited data quality in the nutrient database. Their contents vary greatly depending on their fortification in foodstuffs or animal feed or are not assessed at all; the contribution of other sources, e.g., fluoride intake from toothpaste, is unknown; and/or the bioavailability fluctuates (manganese) or is homeostatically regulated (copper). As the quantities tend to be underestimated rather than overestimated, the upper bounds were nevertheless used. (ii) Regarding total fat, an upper bound was applied (40% of energy intake) [31,32]. It was assumed that lower and upper bounds of fatty acids (saturated fatty acids (SFA), monounsaturated fatty acids (MUFA), and polyunsaturated fatty acids (PUFA)) would ensure adequate quality and intake of fat [29,32]. (iii) For iron, the DRV for premenopausal women was used for the entire population instead of a weighted mean for both sexes because this is the only DRV with a higher value for women and a higher risk for insufficient supply. For similar reasons, the upper limit for alcohol was taken from the women’s DRVs.

The third list of nutrient goals applied goals for the same nutrients as modNUT-DRV, except that here the AR was used instead of the RI (modNUT-AR). The remaining DRVs (11 estimated values, 5 guiding values, and 4 other recommended intakes) remained the same. A detailed overview of all lists is provided in Table 4 in the Results section.

View this table:
Table 3: Results of the deviation indicators between the observed diet and optimized models for the allNUT-DRV, modNUT-DRV, and modNUT-AR models in all four objective function types.
View this table:
Table 4: Nutrient goals for all analyses and nutrient contents for the observed diet and optimized diet in the linear-relative model for all nutrient goals (allNUT-DRV), the modified list of nutrient goals (modNUT-DRV), and the modified list of nutrient goals using the AR (modNUT-AR).

2.3.3 Mathematical implementation of nutrient goals in the optimization model

In addition to minimizing deviation from observed dietary intakes, the objective function of models considering nutrient goals had an additional component that only became active when a nutrient goal could not be fulfilled: Embedded Image where for nutrients k with a minimum nutrient goal, only the inadequate intake (negative deviations) Embedded Image is minimized (a), and for nutrients l having a maximum nutrient goal, only excess (positive deviations) Embedded Image is minimized (b). For nutrients with a target goal (e.g., a target of 2,000 kcal/d), the sum of negative and positive deviations is minimized. This allows deviation from nutrient goals if the model would otherwise find no feasible solution, which helps to identify the source of infeasibility. It should be noted that in the case of the squared function Embedded Image and Embedded Image are squared.

2.3.4 Complete objective function and technical aspects

The objective function used for investigations in the present work aims to minimize Fj, which is the sum of the total deviation from each specific objective: Embedded Image A high weight WN was assigned to penalize the objective function value if the solution deviated from nutrient goals.

The diet optimization models were developed using R version 4.1.3 [33], with an R package specifically designed for this project using the ROI (R Optimization Infrastructure) package version 0.3-3 [34], the solver lpsolve [35] for linear optimization, and quadprog version 1.5.8 [36] for quadratic/squared optimization. The data of the descriptors were stored in an SQL database and edited using MySQL Workbench version 8.0.

2.4 Analysis

To study the impact of the use of the hierarchical food code FoodEx2 as decision variables, a simple optimization model was used that only included the minimization from the observed diet in the objective function, acceptability constraints, and no nutrient constraints. To study the impact of the mathematical type of objective function (Devlr,la,sr,sa) and varying nutrient goals, 12 scenarios of the complete optimization model were run using decision variables from FoodEx2 level 3 with all four objective function types, acceptability constraints, and the allNUT-DRV, modNUT-DRV, and modNUT-AR lists for nutrient goals. To study how the mathematical implementation type used for objective function affected the results, the following indicators were used to measure deviation from the observed diet:

  • - The absolute sum of changes in food groups (grams/day):

  • Sum of absolute changes Embedded Image

  • - The sum of relative changes in food groups (%):

    Sum of relative changes Embedded Image

  • - The number of food groups that increased in the optimized diet

  • - The number of food groups that decreased in the optimized diet

  • - The number of food groups that disappeared in the optimized diet

  • - The number of food groups that did not change in the optimized diet

  • - The number of level 1 groups that were binding acceptability constraints.

Another criterion for the comparison between the models was the number of binding nutrient constraints defined as those constraints where the nutrient quantity in the optimized diet was equal to the imposed upper or lower bound [6].

3 Results

3.1 Use of a hierarchical food code as decision variables

Intuitively, a food code constructed according to a parent–child hierarchy will generate similar solutions for the decision variables independent of the applied level. We validated this principle by running optimization models with decision variable sources ranging from level 7 to level 2 (S6 Table). In this context, a low level of aggregation in FoodEx2 meant a higher number of decision variables was used for optimization; this was counterbalanced by uniform reporting of results, often on a highly aggregated level (see the Methods section). The optimization program had no difficulties in processing the 927 decision variables of level 7 or other levels and generated equal amounts for each reported food group regardless of the applied level. These findings suggest that the selected R programs represent a powerful tool for detailed food consumption data, such as those categorized in FoodEx2.

However, whether food group level affects the results when further parameters are applied (such as nutrient constraints) was not investigated, but was rated according to the experience with optimization models as minor. The relative independence of level may be explained by the fact that due to the clear parent–child relationship from level to level, each level contains the same information but is subdivided differently.

3.2 Different types of objective functions

With the outlined metrics we compared the optimized diet with the observed diet for four different optimization functions and three nutrient goal lists (see Table 3). In the allNUT-DRV models, the absolute and relative sum of changes in food groups was higher than those in the modNUT-DRV and modNUT-AR models. However, the largest differences in the deviation indicators were between the four different objective function types. As defined in the objective function, the sum of absolute quantitative changes was higher in the relative models than in the absolute models: for example, a total of 5,158 g of food for the linear-relative allNUT-DRV model vs. 493 g for the linear-absolute allNUT-DRV model.

Consequently, the relative changes were considerably higher in the absolute models than in the relative ones (e.g., an average of 965% per variable for the linear-absolute modNUT-DRV model vs. 23% for the linear-relative modNUT-DRV model).

The absolute models for allNUT-DRV, modNUT-DRV, and modNUT-AR reached none of the acceptability limits on FoodEx2 level 1 (limits shown in S4 Table). All relative models except for the modNUT-AR squared-relative model reached one to three acceptability limits. In the squared-relative allNUT-DRV model, for example, these were maximum limits for “Vegetables,” “Drinking water,” and “Coffee and tea.” The linear-relative modNUT-DRV model, which used high quantities of beverages as energy-independent variables to increase nutrient content, did not reach the acceptability constraints for these groups, but did for “Potatoes” and “Legumes and nuts.”

The linear models generated results that left the largest number of decision variables unchanged (relative (Devlr): 206–218; absolute (Devla): 177–182 out of 255 decision variables), whereas the squared-relative models left less than half of the variables (85–100) unchanged and the squared-absolute models left only 7–8 food groups unchanged. The latter also showed the highest number of food groups disappearing from the diet (96–106 food groups).

3.3 Nutrient goals

Table 44 shows the nutrient contents of the optimized diets for the linear-relative objective function. The optimized nutrient contents across the other objective function forms were similar and are shown in S6 Table. All nutrient goals were met in all models. In the allNUT-DRV, modNUT-DRV, and modNUT-AR models, the constraints for the six nutrients SFA and free sugars (upper limits), and MUFA, fiber, potassium, and calcium (lower limits) were binding, meaning that they reached exactly their upper or lower bound. In the allNUT-DRV model (15 upper and 33 lower bounds), cholesterol, sodium, iodine, fluoride, and zinc were also binding constraints. In the modNUT-DRV approach, there were four additional binding nutrient constraints (cholesterol, sodium, PUFA, and iron) from a total of 16 upper bounds and 28 lower bounds. The modNUT-AR model had the same number of nutrient goals, but only PUFA as another binding constraint. After excluding nutrient goals due to limited data quality in the modNUT-DRV and modNUT-AR models, the DRV was no longer reached for iodine and fluoride, and even the AR [29] was not reached for iodine in the modNUT-AR model. For copper and manganese, the DRV was met despite exclusion of their intake goals.

4 Discussion

In the present study, we showed different ways food groups as decision variables, the objective function, and nutrient goals of optimization models used for deriving FBDGs can be defined, and explored their implications on optimization results. We also investigated the implications of using a hierarchical food code as the source of decision variables for FBDG development, showing that selecting an appropriate food code, based on the desired level of detail for the FBDGs, is essential for the model. Further, we revealed the impact of using different mathematical forms of the objective function on dietary changes and provided reasons for making decisions about objective function types in the future. Lastly, we clarified which nutrient goals are suitable for use in the model. These analyses paved the way for constructing an optimization model that forms the basis of Germany’s FBDG 2024.

The described optimization model could handle several hundred food groups (ranging from 255 at level 3 to 927 at level 7) without interfering with the internal parent–child hierarchy of the FoodEx2 food code (S6 Table). Subsequently, deciding on which level of FoodEx2 the optimization should be run is vital. We showed that the same optimized food intake quantities, independent of which food group level was initially selected, were obtained in an optimization model minimizing the distance from the observed intake data only and without nutritional constraints (S2 Table), demonstrating the hierarchical consistency of the food code data.

Previous studies on optimization have mostly used national food codes, selected codes of intake survey data, or a predefined food list to generate a list of decision variables [8,13,40,41]. The food classification system in this study supports a harmonized approach of data extrapolation, expansion, and use; it matches various food data sources, striving to harmonize nutritional research not only in European countries but also worldwide [42,43].

There are several reasons why a well-accepted and internationally used hierarchical system like FoodEx2 is useful for diet optimization: (i) it provides clearly defined food groups at each level; (ii) it provides representative and regularly updated food intake data, which can be easily incorporated and compared; (iii) gaps in the data can be calculated using dietary intake as weighting factors (see Methods section, [28]); and (iv) solutions can be evaluated by researchers of different disciplines working with such food code data [42]. One challenge in the application of FoodEx2 with regards to FBDGs is that the exposure-rooted classification does not always fit with a nutritional–physiological perspective (e.g., no clear distinction between whole grain and refined grain products); therefore, adjustments, like those described in the methods section, may be needed.

Another question addressed the selection of an appropriate aggregation level of food groups as decision variables. In this analysis, using FoodEx2 level 3 as the source of decision variables was adequate to calculate optimized intakes with the required detail to quantify each of the 18 food groups reported as output of the optimization model and basis of the FBDG, while also allowing for the highest possible level of aggregation. In the type of parent–child hierarchy seen in FoodEx2, the aggregated characteristics of food groups are similar across levels but have different degrees of detail. Therefore, the largest possible aggregation of food groups is required to minimize the chance of single food groups biasing the representativeness of the food group. As an example, increasing the intake of the nutrient-dense food group liver could lead to a significantly smaller amount of total meat than would have been the case without this variable. The preference for more aggregated food groups that better represent a general group over variables with specific properties has been described previously [44]. If a different degree of detail is required, e.g., if differences in citrus fruit intake needed to be addressed, choosing a different food group level would be necessary. Therefore, the food groups about which statements are to be made define the ideal FoodEx2 level of decision variables for optimization.

Further use of optimization results, e.g., for the development of FBDGs, usually target generic food groups that reflect the public’s understanding and are suited for easily understandable communication [12]. These food groups are in general much broader than the single estimates for food groups generated by the optimization model. Thus, we decided on a uniform reporting schema of the optimization results independent of the level and details of the model itself. We decided to also report the optimal intake of discretionary foods. In our case, they were essential to remain consistent with the FoodEx2 classification system in our optimization model. Some nutrient goals, especially upper limits for SFA, alcohol, and free sugars, particularly focus on discretionary food groups. These food groups are often not considered in diet modeling for FBDG derivation [9,45]. However, there are also examples of the integration of discretionary foods in food guides from the Netherlands and Belgium [9,46].

Above, we mentioned the tendency of the model to select specific foods that favor the fulfillment of optimization functions such as nutritional constraints due to their characteristics. Thus, it is important to set appropriate constraints for each food group at all levels. In our case, similar to earlier studies, the 5th and 95th percentiles of observed food intakes were selected as minimum and maximum constraints to keep optimized quantities within acceptable intake ranges [47]. To date, no other study has provided rational criteria to assess the acceptability of these limits [5,48,49]. Borgonjen et al. [41] found that the application of the 10th and 90th percentiles increased the number of nutrients that were difficult to fulfill and recommended the use of the 5th and 95th percentiles, while also stating that a narrower range of percentiles would more closely represent average food patterns, making the recommendations easier to adopt. A narrow range acceptability constraints can also limit the flexibility of the model to derive healthy diets, once indicators for diet–health relations are incorporated: using the 90th percentile for all individuals as acceptability constraints, the model could, for example, only use a maximum of 202 g/d for “Vegetables,” whereas the 95th percentile cutoff is 263 g/d (applied in this study), which is closer to an optimal vegetable intake regarding health [23]. Therefore, for our model we selected the 5th and 95th percentiles as a compromise between limiting unrealistic results and providing flexibility to fulfill the model’s goals and, in the future, health-oriented goals.

Regarding the mathematical form of the objective function, while linear approaches seem to be the most common choice for most optimization problems, the reasoning for this choice of objective function is rarely found [5]. Some studies claim that quadratic functions are superior in terms of cultural acceptability because smaller changes in more food groups would be easier to accept [9]. In the present study, we compared both relative and absolute linear and quadratic objective functions. However, our modeling could only partly confirm these theoretical considerations. The number of unchanged food groups was the highest in the linear-relative model, demonstrating that this model produces larger changes in fewer food groups, in accordance with the literature [5,9]. As expected, the quadratic approaches led to smaller changes in more food groups, but the sums of these changes, when considered both absolutely and relatively, were similar to or greater than the sums of changes from the linear models. Furthermore, particularly in the squared-absolute models, the quadratic functions led to several food groups being excluded from the diet on the decision variable level, which mostly affected food groups with dietary intakes <1 g/d. Previously, where a quadratic approach was used, the smallest observed intakes were a priori rounded down to 0 g/d [50].

When deviations from observed diets in absolute quantities were compared with relative quantities, the absolute models were expected to make smaller changes. The absolute model does not consider the scales of food group quantities, whereas the relative model gives each decision variable a similar weight and therefore tends to favor food groups consumed in high quantities. The results showed that these assumptions held true and indeed the deviation for the absolute models was smaller. However, a relative approach can assure that popular foods are used to meet the model’s requirements: in the linear-absolute allNUT-DRV model, “Tea leaves derivatives and tea ingredients,” as a dried and concentrated component, was used to fulfill the goal for fluoride. By contrast, large amounts of coffee beverages were used in the linear-relative model. Therefore, a lower deviation from the observed diet does not necessarily mean that the optimized diet is more practicable.

We observed that each type of objective function had specific properties that directed the model to find the optimal solutions. Which model performs best is therefore subject to the stated hypothesis and context of the analyses. In our case, we wanted to use the model results to guide the development of the new FBDGs for Germany. Therefore, we hypothesized that larger changes to often-used food groups would be the best strategy to communicate the new guidelines, rather than focusing on small changes to many food groups or eliminating some rarely used food groups. The type of function that met these criteria best was the linear-relative function, which was selected as our approach.

Considering the observed dietary intake via the objective function and as acceptability constraints fulfilled the EFSA framework’s requirement to consider this aspect for the derivation of FBDGs [1] and increases the acceptability of the optimization results. However, the term acceptability should be handled carefully: the acceptability of a diet in a population is not only represented by the smallest possible deviation from the mean observed diet, but can also be displayed in other characteristics such as financial cost [49]. Without application in real-life studies, as done previously [51], it is impossible to define which objective function is most acceptable. Further, the observed diet represents no single individual’s dietary pattern, but an average of all survey participants. Hence, even the starting point may not be acceptable to many people: the most acceptable option for an individual will always be a personal decision [6,44].

The fulfillment of specific nutrient goals strongly drives optimization results and hence the scientific basis for FBDGs. Therefore, a thorough investigation of the selection and impact of different nutrient goals was undertaken. All nutrient goals could be fulfilled in the allNUT-DRV, modNUT-DRV, and modNUT-AR models. However, with more and higher nutrient goals (allNUT-DRV > modNUT-DRV > modNUT-AR), a greater number of binding nutrients constraints appeared. Binding nutrients are most difficult for the model to fulfill and strongly drive the optimization results. Most often, they appear when the observed diet failed to meet a nutrient goal. On the other hand, due to the mutual influence of the model’s parameters with rising complexity can also apparently uncritical nutrients or constraints cause challenges for the model: E.g., the upper bound for SFAs forced the model to change fat-providing foods to fulfil both the nutrient goal constraints for MUFAs and SFAs. Thereby, the lower bound for MUFAs became a binding constraint, although this goal was initially met in the observed diet. Another optimization study found that constraints for SFAs were difficult to fulfill with subsequent changes in fatty acid profile [52]; similar observations were found for vitamin E and sodium combined with potassium DRVs in the US [53–55].

Comparing the three different approaches to incorporate nutrient goals regarding deviation from the observed diet, we found the biggest difference between the modified lists and the first list using all nutrient goals. The latter led to solutions with higher deviations from the observed diet. These findings were expected because the first list is more restrictive. This is in line with other optimization studies that tend to exclude or adapt single nutrients that drive strong deviations [44,56]; modifying the list of nutrient goals ensures optimized diets do not increase rarely consumed food groups for nutrients that are not well-represented in the data anyways [6]. Furthermore, there often is a trade-off between incorporating a more restrictive nutrient constraint and acceptability. Less acceptable results can also stem from limited data quality for certain nutrients (see Chapter 2.4), which in this study was resolved by excluding the affected nutrient goals in the modNUT-DRV and modNUT-AR lists.

The allNUT-DRV models showed the largest deviations from the observed diet, especially in their high beverage intakes. Often, optimization studies suggest major increases in beverages like water and tea, as these food groups provide nutrients without contributing to energy or fats and sugars. This issue can be dealt with by constraining the total quantity of all foods or beverages [44,52,57]. Excluding nutrient goals from allNUT-DRV in the modified lists (modNUT-DRV and modNUT-AR), especially for nutrients frequently enriched (e.g., salt with iodine) or poorly represented in the data (e.g., fluoride), improved this acceptability issue without a constraint on total quantity.

The RI covers the individual needs of 97,5% of the healthy population but also overestimates the needs of half the individuals. The AR, on the other hand, only meets the needs of 50% of a defined group of people, meaning that its population-wide application may increase the risk of deficiencies or malnutrition [58]. Regarding diet modeling, the advisory committee of the US Dietary Guidelines 2020 concluded that the RI is preferred for planning individual dietary recommendations and the AR is preferred at the population level, but needs to be adjusted appropriately and subsequently evaluated [59]. By contrast, in France, the Netherlands, and the Nordic Nutrition Recommendations, the RI was used to derive FBDGs [8,9,60].

Another strategy can be to apply the AR for specific nutrients only. With regards to iron, Nordman et al. compared the AR and RI and found that “initial optimizations indicated a difficulty in fulfilling the high iron recommendations of pre-menopausal women without imposing large changes in the diet […]” [44]. To avoid compromising diet acceptability, a separate dietary pattern only for premenopausal women was calculated. This kind of target group segregation does not fit the goal of the German FBDG, which is supposed to include a dietary pattern applicable to the general adult population. Hence, the iron value for premenopausal women was used as the nutrient goal in the present model. Furthermore, the share of plant-based foods has increased in these simple models; the potential of lower iron absorption in a more plant-based dietary pattern [61] is another argument in favor of using the higher target (RI), especially for this critical nutrient.

Overall, excluding nutrient goals not well represented in the data led to more realistic results and gave greater flexibility to the model. This allowed to take other relevant aspects for FBDGs into account, such as diet–health relationships or environmental aspects [2]. However, a nutrient supply sufficient to meet the requirements of the population should be provided irrespective of these additional aspects. The modNUT-DRV approach seemed to provide the best compromise between the different requirements for nutrient goals in the German FBDG. The nutrients that were not regarded in the modified lists should still be considered in the final formulation of the FBDG, for example, recommendations for the use of salt fortified with iodine and fluoride.

Strengths and limitations

A variety of options are available to carry out diet modeling for FBDGs [62]; here, a systematic investigation of optimization options was undertaken. The main strength of the study is that it provides a rationale for decisions regarding the German FBDGs 2024 and could also guide other research using similar approaches. The study therefore adds to the body of data on the use of diet optimization for FBDGs. Although this simple model accounts only for nutrients and acceptability, these are the most common components of diet optimization studies [5] and this framework can easily be expanded and applied to other use cases.

However, this study could only address some of the options in the context of diet modeling for FBDGs. With increasing model complexity, further decisions about the (mathematical) options need to be made, which were not addressed here. Whether such decisions challenge the current conclusions is unknown. In this work, we introduced various indicators for deviation from the observed diet. Data in the literature on this aspect is scarce and further investigation of the indicators is needed to confirm which ones give robust and meaningful results to compare various objective functions. Next, there are limitations of databases themselves, whether it is the age of the data (e.g., the NVS II was conducted 2005-2007) or their methodologies (e.g., under-reporting in dietary surveys, variations in nutrient compositions).

Nonetheless, a diet optimization model may be easily updated whenever further or new data is available.

5 Conclusions

In this study, a novel diet optimization model was developed investigating crucial parameters such as the choice of decision variables, the mathematical form of the objective function, and nutrient goals. These insights will contribute to the framework not only for the German FBDG but also for decisions other FBDG-makers may face when using a diet modeling approach.

To answer the study objectives, only the nutritional and acceptability aspects for one target group of the German adult population were addressed. The integration of indicators for other dimensions of a sustainable diet, e.g., diet–health relations and environmental aspects, as a basis for the German FBDG 2024 will be part of a follow-up article.

Data Availability

The data underlying the results presented in the study are available from the European Food Safety Authority (https://www.efsa.europa.eu/sites/default/files/2023-08/pad-guidance-for-applicants.pdf) and from the Max Rubner-Institut (https://blsdb.de/license).

Short title

Framework for Germany’s FBDG update: Food groups, nutrient goals, and objective functions

Author contributions

ACS ran the analyses and wrote the first draft of the manuscript. ACS, HB, JC, UN, and BW conceptualized the optimization algorithm and analyses. RG and FV developed the code for the optimization algorithm. ACS and RG compiled the database. CB, JC, KG, AK, JL, UN, MR, SL, HH, LS, and BW contributed to the manuscript development, conceptualization, and interpretation of results. All authors discussed the procedure and results outlined in this manuscript and carefully revised the manuscript.

Supporting information captions

S1 Table. Classification of whole grain and refined grain on all levels of FoodEx2

S2 Table. Working group of the German Nutrition Society for the development of the scientific basis for the German food-based dietary guideline 2024.

S3 Table. All decision variables (food groups) on level 3 with according parent terms. S4 Table. Acceptability constraints on level 1.

S5 Table. Share of population sub-groups used to calculate mean DRVs for adults aged 18-64. S6 Table. Optimisation results for all analyses including food group quantities for the FBDG food groups, deviation indicators, and energy and nutrient contents.

Acknowledgments

We thank Arno Lellmann for the thorough compilation of DRV values into a database, Theresa Maria Ting for her contribution to the calculation of the metrics for the deviation of dietary habits, and Julia Haardt for her contribution to the classification of whole grains food groups in FoodEx2. We also thank Ute Alexy for her support with the LEBTAB data, Maike Benz for support in the dietary intake data analyses, and Katja Sandfuchs for support with the BLS-FoodEx2 matching.

References

  1. 1.↵
    EFSA (European Food Safety Authority). Scientific opinion on establishing food-based dietary guidelines. EFSA J. 2010; 8:1460. doi: 10.2903/j.efsa.2010.1460.
    OpenUrlCrossRef
  2. 2.↵
    FAO (Food and Agriculture Organization), WHO (World Health Organization), editors. Sustainable healthy diets - Guiding principles. Rome; 2019.
  3. 3.↵
    . Bechthold A, Boeing H, Tetens I, Schwingshackl L, Nöthlings U. Perspective: food-based dietary guidelines in Europe-scientific concepts, current status, and perspectives. Adv Nutr. 2018; 9:544–60. doi: 10.1093/advances/nmy033 PMID: 30107475.
    OpenUrlCrossRefPubMed
  4. 4.↵
    . Schäfer AC, Schmidt A, Bechthold A, Boeing H, Watzl B, Darmon N, et al. Integration of various dimensions in food-based dietary guidelines via mathematical approaches Report of a DGE/FENS Workshop in Bonn, Germany, 23-24 September 2019. Br J Nutr. 2020:1–18. doi: 10.1017/S0007114520004857 PMID: 33272337.
    OpenUrlCrossRefPubMed
  5. 5.↵
    . van Dooren C. A review of the use of linear programming to optimize diets, nutritiously, economically and environmentally. Front Nutr. 2018; 5:48. doi: 10.3389/fnut.2018.00048 PMID: 29977894.
    OpenUrlCrossRefPubMed
  6. 6.↵
    . Gazan R, Brouzes CMC, Vieux F, Maillot M, Lluch A, Darmon N. Mathematical optimization to explore tomorrow’s sustainable diets: a narrative review. Adv Nutr. 2018; 9:602–16. doi: 10.1093/advances/nmy049 PMID: 30239584.
    OpenUrlCrossRefPubMed
  7. 7.↵
    Australian Government, National Health and Medical Research Council. A modelling system to inform the revision of the Australian guide to healthy eating. 2011 [cited 5 May 2022]. Available from: https://www.eatforhealth.gov.au/sites/default/files/files/public_consultation/n55a_dietary_guidelines_food_modelling_111216.pdf.
  8. 8.↵
    . Mariotti F, Havard S, Morise A, Nadaud P, Sirot V, Wetzler S, et al. Perspective: modeling healthy eating patterns for food-based dietary guidelines-scientific concepts, methodological processes, limitations, and lessons. Adv Nutr. 2021; 12:590–9. doi: 10.1093/advances/nmaa176 PMID: 33508130.
    OpenUrlCrossRefPubMed
  9. 9.↵
    . Brink E, van Rossum C, Postma-Smeets A, Stafleu A, Wolvers D, van Dooren C, et al. Development of healthy and sustainable food-based dietary guidelines for the Netherlands. Public Health Nutr. 2019; 22:2419–35. doi: 10.1017/S1368980019001435 PMID: 31262374.
    OpenUrlCrossRefPubMed
  10. 10.↵
    . Ferguson EL, Darmon N, Briend A, Premachandra IM. Food-based dietary guidelines can be developed and tested using linear programming analysis. J Nutr. 2004; 134:951– 7.
    OpenUrlAbstract/FREE Full Text
  11. 11.↵
    1. Cochran JJ,
    2. Cox LA,
    3. Keskinocak P,
    4. Kharoufeh JP,
    5. Smith JC
    Banciu M. Dual simplex. In: Cochran JJ, Cox LA, Keskinocak P, Kharoufeh JP, Smith JC, editors. Wiley Encyclopedia of Operations Research and Management Science. Hoboken: Wiley; 2010.
  12. 12.↵
    . Herforth A, Arimond M, Álvarez-Sánchez C, Coates J, Christianson K, Muehlhoff E. A global review of food-based dietary guidelines. Adv Nutr. 2019; 10:590–605. doi: 10.1093/advances/nmy130 PMID: 31041447.
    OpenUrlCrossRefPubMed
  13. 13.↵
    . Monteiro JS, Botelho RBA, Zandonadi RP, Araujo WMC. Is there a convergence between the food classification adopted by food-based dietary guidelines and food science and technology. Foods. 2023; 12:3824. Epub 2023/10/18. doi: 10.3390/foods12203824 PMID: 37893716.
    OpenUrlCrossRefPubMed
  14. 14.↵
    EFSA (European Food Safety Authority). The food classification and description system FoodEx2 (revision 2). 2015 [cited 25 Oct 2021]. Available from: https://efsa.onlinelibrary.wiley.com/doi/pdf/10.2903/sp.efsa.2015.EN-804.
  15. 15.↵
    . Merten C, Ferrari P, Bakker M, Boss A, Hearty A, Leclercq C, et al. Methodological characteristics of the national dietary surveys carried out in the European Union as included in the European Food Safety Authority (EFSA) Comprehensive European Food Consumption Database. Food Addit Contam Part A Chem Anal Control Expo Risk Assess. 2011; 28:975–95. doi: 10.1080/19440049.2011.576440 PMID: 21732710.
    OpenUrlCrossRefPubMed
  16. 16.↵
    EFSA (European Food Safety Authority). Use of the EFSA comprehensive European food consumption database in exposure assessment. EFSA J. 2011; 9:2097. doi: 10.2903/j.efsa.2011.2097.
    OpenUrlCrossRef
  17. 17.↵
    EFSA (European Food Safety Authority), Nikolic M, Ioannidou S. FoodEx2 maintenance 2020. EFSA Supporting publication. 2021:EN–6507. doi: 10.2903/sp.efsa.2021.EN-6507.
    OpenUrlCrossRef
  18. 18.↵
    . Gedrich K, Hensel A, Binder I, Karg G. How optimal are computer-calculated optimal diets. Eur J Clin Nutr. 1999; 53:309–18. doi: 10.1038/sj.ejcn.1600727 PMID: 10334657.
    OpenUrlCrossRefPubMed
  19. 19.
    . Maillot M, Darmon N, Drewnowski A. Are the lowest-cost healthful food plans culturally and socially acceptable. Public Health Nutr. 2010; 13:1178–85. doi: 10.1017/S1368980009993028 PMID: 20105388.
    OpenUrlCrossRefPubMedWeb of Science
  20. 20.↵
    . Eustachio Colombo P, Patterson E, Schäfer Elinder L, Lindroos AK, Sonesson U, Darmon N, et al. Optimizing school food supply: integrating environmental, health, economic, and cultural dimensions of diet sustainability with linear programming. Int J Environ Res Public Health. 2019; 16:3019. Epub 2019/08/21. doi: 10.3390/ijerph16173019 PMID: 31438517.
    OpenUrlCrossRefPubMed
  21. 21.↵
    . Niforou K, Livaniou A, Ioannidou S. FoodEx2 maintenance 2023. EFSA Supporting publications. 2024:EN–8813. doi: 10.2903/sp.efsa.2024.EN-8813.
    OpenUrlCrossRef
  22. 22.↵
    Krems C, Walter C, Heuer T, Hoffmann I. Lebensmittelverzehr und Nährstoffzufuhr - Ergebnisse der Nationalen Verzehrsstudie II. In: Deutsche Gesellschaft für Ernährunge. V., editor. 12. Ernährungsbericht 2012. Bonn; 2012. pp. 40–85.
    OpenUrl
  23. 23.↵
    . Schwingshackl L, Knüppel S, Michels N, Schwedhelm C, Hoffmann G, Iqbal K, et al. Intake of 12 food groups and disability-adjusted life years from coronary heart disease, stroke, type 2 diabetes, and colorectal cancer in 16 European countries. Eur J Epidemiol. 2019; 34:765–75. doi: 10.1007/s10654-019-00523-4 PMID: 31030306.
    OpenUrlCrossRefPubMed
  24. 24.↵
    . Max Rubner-Institut, Bundesforschungsinstitut für Ernährung und Lebensmittel (MRI). Nationale Verzehrsstudie II. Ergebnisbericht, Teil 2. Die bundesweite Befragung zur Ernährung von Jugendlichen und Erwachsenen. Karlsruhe; 2008.
  25. 25.↵
    . Heuer T, Krems C, Moon K, Brombach C, Hoffmann I. Food consumption of adults in Germany: results of the German National Nutrition Survey II based on diet history interviews. Br J Nutr. 2015; 113:1603–14. Epub 2015/04/13. doi: 10.1017/S0007114515000744 PMID: 25866161.
    OpenUrlCrossRefPubMed
  26. 26.↵
    . Hartmann B, Vasquez-Calcedo A-L, Bell S, Krems C, Brombach C. The German nutrient database basis for analysis of the nutritional status of the German population. J Food Compost Anal. 2008; 21:S115–S118.
    OpenUrl
  27. 27.↵
    . Sichert-Hellert W, Kersting M, Chahda C, Schäfer R, Kroke A. German food composition database for dietary evaluations in children and adolescents. J Food Compost Anal. 2007; 20:63–70. doi: 10.1016/j.jfca.2006.05.004.
    OpenUrlCrossRef
  28. 28.↵
    . Gazan R, Barré T, Perignon M, Maillot M, Darmon N, Vieux F. A methodology to compile food metrics related to diet sustainability into a single food database: application to the French case. Food Chem. 2016; 238:125–33. doi: 10.1016/j.foodchem.2016.11.083 PMID: 28867082.
    OpenUrlCrossRefPubMed
  29. 29.↵
    Deutsche Gesellschaft für Ernährung e. V. (DGE), Österreichische Gesellschaft für Ernährung, Schweizerische Gesellschaft für Ernährung, editors. Referenzwerte für die Nährstoffzufuhr. 2nd ed. Bonn; 2021.
  30. 30.↵
    EFSA (European Food Safety Authority). Overview on tolerable upper intake levels as derived by the Scientific Committee on Food (SCF) and the EFSA Panel on Dietetic Products, Nutrition and Allergies (NDA). 2018 [cited 22 Oct 2021]. Available from: https://www.efsa.europa.eu/sites/default/files/assets/UL_Summary_tables.pdf.
  31. 31.↵
    . Yu-Poth S, Zhao G, Etherton T, Naglak M, Jonnalagadda S, Kris-Etherton PM. Effects of the national cholesterol education program’s step I and step II dietary intervention programs on cardiovascular disease risk factors: a meta-analysis. Am J Clin Nutr. 1999; 69:632–46.
    OpenUrlAbstract/FREE Full Text
  32. 32.↵
    . Wolfram G, Bechthold A, Boeing H, Ellinger S, Hauner H, Kroke A, et al. Evidence-based guideline of the German Nutrition Society: fat intake and prevention of selected nutrition-related diseases. Ann Nutr Metab. 2015; 67:141–204. doi: 10.1159/000437243 PMID: 26414007.
    OpenUrlCrossRefPubMed
  33. 33.↵
    R Core Team. R: A language and environment for statistical computing. Vienna, Austria: 2022 [cited 25 Apr 2022]. Available from: https://www.R-project.org/.
  34. 34.↵
    . Theußl S, Schwendinger F, Hornik K. ROI : an extensible R optimization infrastructure. J Stat Soft. 2020; 94. doi: 10.18637/jss.v094.i15.
    OpenUrlCrossRef
  35. 35.↵
    Berkelaar M. lpSolve: Interface to ‘Lp_solve’ v. 5.5 to Solve Linear/Integer Programs. 2023 [cited 3 May 2024]. Available from: https://rdrr.io/cran/lpSolve/.
  36. 36.↵
    Berwin A, Turlach R, Weingessel A. quadprog: functions to solve quadratic programming problems. Comprehensive R Archive Network (CRAN) 2019 [updated 30 Jun 2023; cited 12 Jul 2023]. Available from: https://cran.r-project.org/web/packages/quadprog/index.html.
  37. 37.
    Deutsche Gesellschaft für Ernährung e. V. (DGE). Fettkonsum und Prävention ausgewählter ernährungsmitbedingter Krankheiten - Evidenzbasierte Leitlinie. Bonn; 2006.
  38. 38.
    . Strohm D, Boeing H, Leschik-Bonnet E, Heseker H, Arens-Azevêdo U, Bechthold A, et al. Sonderdruck: Speisesalzzufuhr in Deutschland, gesundheitliche Folgen und resultierende Handlungsempfehlung. Ernährungs Umschau. 2016; 63:2–12. doi: 10.4455/eu.2016.012.
    OpenUrlCrossRef
  39. 39.
    . Ernst JB, Arens-Azevêdo U, Bitzer B, Bosy-Westphal A, Zwaan M de, Egert S, et al. Quantitative Empfehlung zur Zuckerzufuhr in Deutschland. Kurzfassung des Konsensuspapiers der Deutschen Adipositas-Gesellschaft e. V. (DAG), der Deutschen Diabetes Gesellschaft e. V. (DDG) und der Deutschen Gesellschaft für Ernährung e. V. (DGE). Ernährungs Umschau. 2019; 66:26–34. doi: 10.4455/eu.2019.006.
    OpenUrlCrossRef
  40. 40.↵
    . Parlesak A, Tetens I, Dejgård Jensen J, Smed S, Gabrijelčič Blenkuš M, Rayner M, et al. Use of linear programming to develop cost-minimized nutritionally adequate health promoting food baskets. PLoS One. 2016; 11:e0163411. doi: 10.1371/journal.pone.0163411 PMID: 27760131.
    OpenUrlCrossRefPubMed
  41. 41.↵
    . Borgonjen-van den Berg KJ, Vries JHM de, Chopera P, Feskens EJM, Brouwer ID. Sensitivity of food-based recommendations developed using linear programming to model input data in young Kenyan children. Nutrients. 2021; 13:3485. Epub 2021/09/30. doi: 10.3390/nu13103485 PMID: 34684486.
    OpenUrlCrossRefPubMed
  42. 42.↵
    FAO (Food and Agriculture Organization), WHO (World Health Organization). FAO/WHO Global Individual Food consumption data Tool (FAO/WHO GIFT). Project brief updated as per August 10th, 2017. 2017 [updated 20 Sep 2017]. Available from: http://www.fao.org/fileadmin/user_upload/nutrition/docs/assessment/2017-08-10_FAO-WHO_GIFT_Briefing_document_mod.pdf.
  43. 43.↵
    Tufts University. GDD 2018 Dietary Data Harmonization. Global Dietary Database. 2019 [updated 3 Sep 2021; cited 3 Sep 2021]. Available from: https://www.globaldietarydatabase.org/foodex2-data-harmonization.
  44. 44.↵
    . Nordman M, Lassen AD, Stockmarr A, van ‘t Veer P, Biesbroek S, Trolle E. Exploring healthy and climate-friendly diets for Danish adults: an optimization study using quadratic programming. Front Nutr. 2023; 10:1158257. doi: 10.3389/fnut.2023.1158257.
    OpenUrlCrossRef
  45. 45.↵
    45. Ministry of Agriculture Zambia. Zambia Food-Based Dietary Guidelines. Technical recommendations. 2021 [cited 13 Jul 2023]. Available from: https://www.nfnc.org.zm/download/zambia-food-based-dietary-guidelines-technical-recommendations-2021-produced-by-the-ministry-of-agriculture/?wpdmdl=2526&refresh=64afbf813ae801689239425.
  46. 46.↵
    46. Rubens K, Neven L, Jonckheere J. Food and environmentally responsible consumption: towards healthy food patterns for a healthy planet. Background document for the food triangle recommendations. 2021 [cited 6 Jan 2023]. Available from: https://www.gezondleven.be/files/Background-food-and-environment-EN.pdf.
  47. 47.↵
    . Maillot M, Vieux F, Ferguson EF, Volatier J-L, Amiot MJ, Darmon N. To meet nutrient recommendations, most French adults need to expand their habitual food repertoire. J Nutr. 2009; 139:1721–7. doi: 10.3945/jn.109.107318 PMID: 19625703.
    OpenUrlAbstract/FREE Full Text
  48. 48.↵
    . Babb AM, Knudsen DC, Robeson SM. A critique of the objective function utilized in calculating the Thrifty Food Plan. PLoS One. 2019; 14:e0219895. doi: 10.1371/journal.pone.0219895 PMID: 31329629.
    OpenUrlCrossRefPubMed
  49. 49.↵
    . Perignon M, Darmon N. Advantages and limitations of the methodological approaches used to study dietary shifts towards improved nutrition and sustainability. Nutr Rev. 2022; 80:579–97. doi: 10.1093/nutrit/nuab091 PMID: 35142357.
    OpenUrlCrossRefPubMed
  50. 50.↵
    . Broekema R, Tyszler M, van ‘t Veer P, Kok FJ, Martin A, Lluch A, et al. Future-proof and sustainable healthy diets based on current eating patterns in the Netherlands. Am J Clin Nutr. 2020; 393:447. doi: 10.1093/ajcn/nqaa217 PMID: 32766880.
    OpenUrlCrossRefPubMed
  51. 51.↵
    . Hlaing LM, Fahmida U, Htet MK, Utomo B, Firmansyah A, Ferguson EL. Local food-based complementary feeding recommendations developed by the linear programming approach to improve the intake of problem nutrients among 12-23-month-old Myanmar children. Br J Nutr. 2016; 116, Suppl 1:S16–S26. Epub 2015/12/23. doi: 10.1017/S000711451500481X PMID: 26696232.
    OpenUrlCrossRefPubMed
  52. 52.↵
    . Maillot M, Issa C, Vieux F, Lairon D, Darmon N. The shortest way to reach nutritional goals is to adopt Mediterranean food choices: evidence from computer-generated personalized diets. Am J Clin Nutr. 2011; 94:1127–37. doi: 10.3945/ajcn.111.016501. PMID: 21900460.
    OpenUrlAbstract/FREE Full Text
  53. 53.↵
    . Gao X, Wilde PE, Lichtenstein AH, Bermudez OI, Tucker KL. The maximal amount of dietary alpha-tocopherol intake in U.S. adults (NHANES 2001-2002). J Nutr. 2006; 136:1021–6. doi: 10.1093/jn/136.4.1021 PMID: 16549468.
    OpenUrlAbstract/FREE Full Text
  54. 54.
    . Maillot M, Monsivais P, Drewnowski A. Food pattern modeling shows that the 2010 Dietary Guidelines for sodium and potassium cannot be met simultaneously. Nutr Res. 2013; 33:188–94. doi: 10.1016/j.nutres.2013.01.004 PMID: 23507224.
    OpenUrlCrossRefPubMed
  55. 55.↵
    . Maillot M, Drewnowski A. A conflict between nutritionally adequate diets and meeting the 2010 dietary guidelines for sodium. Am J Prev Med. 2012; 42:174–9. doi: 10.1016/j.amepre.2011.10.009 PMID: 22261214.
    OpenUrlCrossRefPubMed
  56. 56.↵
    . Mazac R, Meinilä J, Korkalo L, Järviö N, Jalava M, Tuomisto HL. Incorporation of novel foods in European diets can reduce global warming potential, water use and land use by over 80%. Nat Food. 2022; 3:286–93. doi: 10.1038/s43016-022-00489-9.
    OpenUrlCrossRefPubMed
  57. 57.↵
    . Perignon M, Masset G, Ferrari G, Barre T, Vieux F, Maillot M, et al. How low can dietary greenhouse gas emissions be reduced without impairing nutritional adequacy, affordability and acceptability of the diet? A modelling study to guide sustainable food choices. Public Health Nutr. 2016; 19:2662–74. doi: 10.1017/S1368980016000653 PMID: 27049598.
    OpenUrlCrossRefPubMed
  58. 58.↵
    . Yates AA. Which dietary reference intake is best suited to serve as the basis for nutrition labeling for daily values. J Nutr. 2006; 136:2457–62. doi: 10.1093/jn/136.10.2457 PMID: 16988109.
    OpenUrlAbstract/FREE Full Text
  59. 59.↵
    Dietary Guidelines Advisory Committee, editor. Scientific Report of the 2020 Dietary Guidelines Advisory Committee: Advisory Report to the Secretary of Health and Human Services and the Secretary of Agriculture. Washington, D.C.; 2020.
  60. 60.↵
    Blomhoff R, Andersen R, Arnesen EK, Christensen JJ, Eneroth H, Erkkola M, et al. Nordic Nutrition Recommendations 2023. Copenhagen: Nordic Council of Ministers; 2023.
  61. 61.↵
    . van Wonderen D, Melse-Boonstra A, Gerdessen JC. Iron bioavailability should be considered when modeling omnivorous, vegetarian, and vegan diets. J Nutr. 2023; 153:2125–32. Epub 2023/05/12. doi: 10.1016/j.tjnut.2023.05.011 PMID: 37182693.
    OpenUrlCrossRefPubMed
  62. 62.↵
    FAO (Food and Agriculture Organization of the United Nations). Food-based dietary guidelines. Third webinar: Diet modelling for food-based dietary guidelines. 2019 [updated 26 Oct 2020; cited 26 Oct 2020]. Available from: http://www.fao.org/nutrition/education/food-dietary-guidelines/capacity-development/webinar-series/third-webinar/en/.
Back to top
Posted October 25, 2024.
Download PDF
Data/Code
Email
A methodological framework for deriving the German food-based dietary guidelines 2024: food groups, nutrient goals, and objective functions
Anne Carolin Schäfer, Heiner Boeing, Rozenn Gazan, Johanna Conrad, Kurt Gedrich, Christina Breidenassel, Hans Hauner, Anja Kroke, Jakob Linseisen, Stefan Lorkowski, Ute Nöthlings, Margrit Richter, Lukas Schwingshackl, Florent Vieux, Bernhard Watzl
medRxiv 2024.10.24.24316069; doi: https://doi.org/10.1101/2024.10.24.24316069
Twitter logo Facebook logo LinkedIn logo Mendeley logo
Citation Tools

Subject Area