What Does the DGE Recommend? Chapter 2
Since I strictly ignore the dietary recommendations of the Deutsche Gesellschaft für Ernährung - German Nutritional Society, a largely government funded NGO, designated as appointed authority on nutritional matters and have had exclusively good experiences doing so, I’m taking a closer look at those recommendations and their foundations in a loose series of articles. Each chapter I’ll pick out individual points from the available DGE documents and studies. Today: individual statements, taken out of context. Hence: almost a rant, but well sourced…
I’m drawing from the document “Scientific Foundations of Food-Based Dietary Recommendations for Germany” (1) and “A methodological framework for deriving the German food-based dietary guidelines 2024: Food groups, nutrient goals, and objective functions” (2), a publication linked (as preprint) by DGE.
Continuing from Chapter 1: the DGE neatly summarises its own methodology - and since I assume that every word in scientific communications from such an influential organisation is chosen deliberately, it’s worth taking a second look at the precision of the wording:
The objective function minimises the deviation from the observed dietary pattern in Germany and reduces the burden of disease and environmental impact. Within the solution space defined by the constraints, the solution is thus chosen that is closest to observed dietary intake while simultaneously being health-promoting and environmentally friendly.
What is minimised: deviation from the status quo. What is to be reduced: disease burden and environmental impact. In optimisation mathematics, this is not a stylistic difference. “Minimise” is a goal; “reduce” is a statement of intent. Whoever expected the maximisation of health to be the primary goal of Germany’s dietary guidelines: that was never on the agenda.
Limitations of the Model
At the same time, it appears that various upper limits are applied based on unspecified criteria (emphasis mine):
For five further food groups (soft drinks, composite dishes, alcoholic beverages, confectionery, sauces and condiments, other), optimised consumption quantities are also calculated. […] These food groups are included in the calculation in order to apply, for example, upper limits for nutrients such as saturated fatty acids or free sugars to the overall diet.
Even without a formal qualification in nutritional science, I fancy I can specify the healthy and optimal upper limit for soft drinks, confectionery, and alcoholic beverages with numerical precision. That optimal quantity certainly differs from actual consumption, or from the tolerably probably-won’t-do-too-much-damage-quantity: Optimal would be: Zero. That one would deploy an algorithm for this is either ambitious basic research, or a generous interpretation of the word “optimised”.
Chapter 3.3 of the publication (2) then states under “Nutrient goals”:
…All nutrient goals were met in all models. In the allNUT-DRV, modNUT-DRV, and modNUT-AR models, the constraints for the six nutrients Fatty acids with no reactive molecular component — normally solid at room temperature. Typically animal fats. → more and free sugars (upper limits), and Fatty acids with one double bond in the molecule — chemically more reactive than saturated fatty acids. Found in both plant oils (olive oil) and animal fats. → more, fiber, potassium, and calcium (lower limits) were binding, meaning that they reached exactly their upper or lower bound…
Given that multiple meta-analyses have by now established that there is no upper limit for saturated fats (3) - meaning no evidence of their harmfulness - this formulation suggests to me that the share of saturated fat was not determined by scientific methods, but rather that a desired maximum was entered as a parameter. The algorithm then naturally did not exceed it. That is, after all, the point of such an imposed ceiling.
And that, as regular readers of butterbei.de know, runs entirely counter to my personal experience.
When an optimisation algorithm finds the externally imposed upper limit for an ingredient as its solution, the obvious implication is that the actual optimum lies above that imposed ceiling, and the algorithm simply cannot reach it due to the explicit parameterisation.
Put differently: the algorithm did not discover that less saturated fat is better. It found the ceiling it was given. What science generally calls this is “confirmation of input premises”. It is not a new finding - it is the confirmation of what was entered beforehand.
The evidence and my own experience tell me that the upper limit for saturated fat is irrelevant (especially in situations with low carbohydrate intake, e.g. with low-carb or ketogenic diets). At the same time it’s fairly obvious that for sugar - a substance the body absolutely does not need - an even higher optimal value than the imposed ceiling would be pretty absurd.
In other words: I consider the algorithm to be scientifically sound - in mathematics, that is, not in nutritional science. The apparently arbitrary parameterisation that drives the results has thus (after Chapter 1) taken another hit. An optimisation algorithm only optimises what you let it optimise.
What Does the DGE Recommend? Chapter 1
Lumping Together Very Different Foods
The food classification is based on FoodEx2 (4), a system for classifying foods according to comprehensible properties and ingredients. Two problems jump out at me:
Butter vs. Margarine
Table 3 in the document groups “butter and margarine” together. Two products whose nutritional composition could not be more different.
Across the internet you’ll find countless comparisons of butter and margarine. All agree that both come from very different sources and have different fat compositions.
Where things diverge sharply is in the conclusions: some favour butter for its natural origin, flavour, and nutrient profile (butter naturally contains several fat-soluble vitamins, as well as saturated fats). Others favour margarine; fat-soluble vitamins are typically added artificially, it comes from plant sources, and it has a higher share of polyunsaturated fats.
To make margarine solid despite this fat composition, the plant oils are hydrogenated, chemically converted into a different form.
The first purely plant-based margarine was produced in 1952, and the manufacturing process doesn’t sound very appetising to me (5). Butter, by comparison, is documented for at least 8,000 years. We’ve lived with margarine for just over 70 years, and with butter for millennia. Same category. Same table column. Same basis for the recommendations.
There are many further aspects one could raise, but only one matters to me here: grouping two such different products together solely because of their shared use as a “spread” is roughly equivalent to aggregating bicycles and aircraft under “means of transport”: Technically correct, analytically worthless. That discredits the data foundation for me. And conveniently, a sentence appearing just after this table reads: “Finally, the quality of the results can only be as good as the underlying data”. That sentence I agree with unreservedly. Regrettably, it also undermines the findings presented in the document, in my view.
A Hypothesis: Oversimplification
If I’m misunderstanding the system, I welcome corrections, but I suspect this is where the source of a recurring problem lies: if I report eating a “hamburger” on a questionnaire, it gets counted as “meat”. Never mind that this meat comes with a generous serving of grain, sugary sauce, and highly processed vegetables - usually accompanied by fried potatoes and a not-insignificant sweetened soft drink.
Nobody considers a hamburger health-promoting, but when I look at the nutrients, ingredients, and their bioavailability, the meat comes out best compared to all the other components. Yet in the typical studies that rely on questionnaires to measure dietary habits, the simplification “hamburger == meat” attributes the effects of every single ingredient to the meat.
FoodEx2 states:
Top-down order of priorities for the choice of food groups for composite dishes (and similar cases). This pre-defined order facilitates the choice of the base term in cases where there are multiple characterising ingredients, none of which has a clear predominance (Table 7:)
Food group Assumed intention when choosing the food Meat based Source of protein Fish based Alternative source of protein Cheese or dairy based Source of protein and fat Egg based Source of protein and fat Legume based Source of protein and carbohydrates / energy Potato based Source of carbohydrates / energy Cereal based Source of carbohydrates / energy Fruit based Source of micronutrients Vegetable based Source of micronutrients
Since the classification explicitly works top-down, a hamburger lands under “Meat based”. The white bun: incidental. The sugary sauce: incidental. The fried potatoes: not even part of it. The litre of sweetened drink on the side: different category, different problem.
Anyone who then statistically determines that “meat eaters” have certain health problems has possibly said more about the accompanying carbohydrates, sauces, and soft drinks than about the meat itself. But “meat is dangerous” makes for a more compelling headline than “hamburgers have lots of ingredients”.
Also worth noting: the difference in how the two common fuels are labelled. Carbohydrates appear as “Carbohydrates / Energy”; fat appears simply as “Fat”, with no “/Energy” appended. Fat contains more than twice the calories per gram of carbohydrates. Classifying it without reference to energy demonstrates at least an interesting worldview: fat is (exclusively) a nutrient, carbohydrates are fuel. My body apparently hasn’t read this table - it continues burning predominantly fat, quite unimpressed.
The Preferred Fuel
One Final Confirmation
Before I close this chapter: one further aspect that categorically prevents a ketogenic diet - as I practise it - from ever being identified as optimal by the DGE’s “scientific approach” appears on page 4:
Energy and Nutrient Goals
The energy and nutrient targets specified for the model are based on the DGE’s reference values for nutrient intake. These must be met by the optimised dietary pattern.
And the document referenced there contains some revealing figures:
| Nutrient | Lower limit | Upper limit |
|---|---|---|
| Fat | – | 40 % |
| Fatty acids with no reactive molecular component — normally solid at room temperature. Typically animal fats. → more | – | 10 % |
| Fatty acids with one double bond in the molecule — chemically more reactive than saturated fatty acids. Found in both plant oils (olive oil) and animal fats. → more | 10 % | – |
| Fatty acids with multiple double bonds in the molecule — chemically more reactive than saturated fatty acids. Typically (but not exclusively) plant oils. Omega-3 and omega-6 fall into this category. → more | 7 % | 10 % |
| Carbohydrates | – | – |
| Free sugars | – | 10 % |
| Fibre | 30 g/day | – |
| Alcohol | – | 10 g/day |
Keto-friendly, and in line with nutritional science, there is no lower limit for Nonessential nutrients are those that the body can synthesise itself out of other nutrients. Therefor these do not have to be provided with food. carbohydrates - but since there is an upper limit for fat at 40% of energy intake, and nobody seriously wants to derive energy from protein, none is needed anyway.
The DGE thus uses an algorithm it has actively prohibited from ever considering this dietary pattern. This isn’t science versus keto. It’s keto versus a calculator with a predetermined outcome.
Solid science? Solid mathematics! Solid nutritional science would mean: not excluding what your own parameterisation prevents you from finding.
But, as a reminder, an already mentioned quote from the DGE document itself, with my own addition:
Finally, the quality of the results can only be as good as the underlying data
(DGE)
… and the arbitrary parameterisation of the algorithm employed
(Olaf)
