“If they can get you asking the wrong questions, they don’t have to worry about the answers.”
— Thomas Pynchon
As noted in the recent column, The MAHA Report Card, recent and significant actions by the government on behalf of the MAHA movement over the first year have included recommendations to decrease “highly processed food” and to flip the pyramid to “eat more protein.” However, by not defining what is meant by “highly processed food” and by neglecting to address the potential role of the NOVA classification in categorizing ultraprocessed foods (NOVA classification group 4, UPFs), the result has been increasing confusion.
One of the persistent arguments against using the NOVA classification to determine whether a food is ultraprocessed or not is that, in terms of defining category boundaries, it is simply too “fuzzy” to be of practical use. The uncertainty created by the MAHA recommendations has reignited this debate. Critics argue that “The boundary between NOVA 3 and NOVA 4 foods is too vague. There are too many foods on the border that can be classified as either ultraprocessed (NOVA classification group 4) or not. There is too much inconsistency for this to be of any practical use.” In other words, if experts themselves can’t always agree whether something is ‘ultraprocessed,’ how can the system be scientifically valid?”
It sounds persuasive. After all, science is supposed to be precise. In current nutrition recommendations, Authority follows this precision: “Eat more protein,” with the downstream implication that this will improve overall health. Specifically, the federal guidelines link higher protein intake to improved muscle and bone health, better metabolic function, better weight management, and reduced chronic disease risk.
But both the criticism of NOVA classification and the certainty of blanket recommendations like “eat more protein” rest on a misunderstanding of how all meaningful categories work in biology, nutrition, and even everyday life. This concept was eloquently elucidated by the cognitive scientist Eleanor Rosch in the 1970s. And given the current discussions around this topic, it warrants revisiting.
Many critics of NOVA and nutrition experts assume that categories are defined by strict rules and by clearly identifiable necessary and sufficient conditions. Something either meets the criteria or it doesn’t. Rosch showed that this isn’t how human cognition actually works.
Instead, categories are organized around prototypes.
Think of the category “bird.”
A robin feels like a very good example of a bird.
A penguin? Less so.
A bat? Not a bird at all.
But what exactly makes something a bird? Feathers? Flying? Egg-laying? Wings? No single trait perfectly defines the category. Some birds don’t fly. Some lay eggs but aren’t birds. The category works because it is organized around a central prototype, with membership weakening toward the edges.
This “graded membership” model is now widely accepted across psychology and linguistics. Categories have:
- Clear central members
- Peripheral members
- And fuzzy boundaries.
Importantly, particularly in the discussion about the utility of NOVA classification, the existence of fuzzy edges does not make the category useless. It simply reflects how the real world operates.
Critics argue that NOVA 3 vs. NOVA 4 classifications can be ambiguous for certain foods, such as particular types of breads, flavored yogurts, and plant milks. But if this boundary fuzziness invalidates NOVA, then we also have to discard much of nutrition science along with its recommendations.
Consider the category of the moment: “protein.”
When public health guidelines tell us to “eat more protein,” what exactly does that mean?
Is collagen the same as whey?
Is gelatin equivalent to leucine-rich muscle protein?
Is LDL (low-density lipoprotein, often referred to as “bad cholesterol”) a fat or a protein?
LDL is actually a lipoprotein, a complex particle made of both fat and protein. It transports lipids in the bloodstream. It doesn’t fit neatly into a single macronutrient box.
Or consider “fat.”
Is cholesterol a fat?
Is a phospholipid the same as a triglyceride?
Is HDL a fat?
The answer depends on context and function. These categories are chemically defined but functionally heterogeneous.
The same applies to “carbohydrates.”
Is fiber the same as sugar?
Is refined starch the same as whole-grain starch?
They are all technically carbohydrates, yet their metabolic effects are profoundly different.
Despite these fuzzy boundaries, no one is suggesting that we abandon the concept of protein, fat, or carbohydrate. These categories are useful because they organize biological reality in a way that predicts outcomes. They are useful even if they are imperfect.
The NOVA system classifies foods based on their level of processing. Critics point out that there is disagreement at the margins, especially between minimally processed foods and ultraprocessed foods (NOVA 4). This is true. But that is not the correct question. The key question is, “How much fuzziness is there, and does it meaningfully affect health findings?”
In the United States and many other Western nations with significant UPF consumption, the bulk of what is consumed does not come from ambiguous foods. It comes from:
- Soda
- Candy
- Packaged snacks
- Industrial baked goods
- Fast food
- Ready-to-eat frozen meals.
These are not borderline cases. They are prototypical NOVA 4 foods.
Disagreement tends to occur in edge cases such as reformulated breads, certain yogurts, and some protein bars. Most ultraprocessed food consumption in Western diets comes from clearly industrial products rather than borderline cases.
In epidemiology, when misclassification errors are random, what researchers call “nondifferential misclassification,” they typically bias results toward zero. In other words, they weaken observed associations rather than artificially inflating them.
If NOVA’s fuzziness were overwhelming, we would expect:
- Weak associations
- Inconsistent results
- Non-monotonic patterns.
Instead, we see the opposite. Across multiple countries and large cohort studies, higher consumption of ultraprocessed foods is associated with:
- Increased obesity risk
- Higher incidence of type 2 diabetes
- Greater cardiovascular disease risk
- Increased all-cause mortality.
Crucially, these associations are often dose-dependent. That means the more UPF someone consumes, the greater the risk. The relationship isn’t random. It isn’t all-or-nothing. It forms a gradient. Similar epidemiologic findings provided the first clues that linked cigarette smoking to the risk of developing lung cancer. Dose-response patterns are powerful in science because they suggest an underlying biological gradient, not just a statistical artifact.
If NOVA classification were so fuzzy as to be meaningless, such consistent dose-dependent findings would be highly unlikely. Instead, what we see is that even with boundary ambiguity, the central pattern holds.
This is where the Food-as-Information (FAI) framework adds an even deeper layer. Under FAI, food is not just calories or nutrients. It is a structured signal delivered to a complex adaptive biological system. The body processes food through multiple nested boundaries:
- The microbiome
- The gut
- The gut-brain axis
- The organism as a whole.
From this perspective, ultraprocessing is not merely a social label. It represents increasing informational distortion. In other words, NOVA 4 foods tend to deliver biological signals that are less coherent relative to evolutionary priors. That distortion is not binary. It exists along a gradient. In other words, if the biological reality is a gradient, then any external category drawn across it will have blurred edges. Because fuzziness is exactly what we expect if the underlying phenomenon is continuous.
Critically, in evaluating biological systems, the existence of fuzzy boundaries does not invalidate a category. It means the category is compressing a continuous reality into usable bins. This is exactly what we do when we categorize proteins, inflammatory states, cholesterol risk categories, and blood pressure treatment cutoffs. NOVA does this, too.
At the end of the day, scientific categories are tools we create. Their value depends not on perfect edges but on predictive power. And the predictive signal around ultraprocessed foods is strong.
The real scientific debate should not focus on boundary ambiguity. We should instead be asking:
- Are the associations causal?
- Are they independent of energy density?
- Do they persist after adjusting for nutrients?
- What mechanisms explain them?
Arguing that NOVA is useless because of fuzzy boundaries misunderstands how science, especially biology, works. Biology is inherently graded. Boundaries are probabilistic. Stability and disruption exist along spectra. While the criticism that “NOVA has fuzzy boundaries” is technically correct, it reflects a misunderstanding of how scientific categories function.
All biologically meaningful categories are graded and prototype-based. Macronutrient categories themselves are heterogeneous and context-dependent. Yet we use them because they help us predict outcomes. NOVA functions similarly.
The central prototype of ultraprocessed food is clear. The edge cases are fewer. The epidemiologic findings are dose-dependent and reproducible. Misclassification would more likely weaken than fabricate the signal. From a Food-as-Information perspective, this makes sense. Ultraprocessing reflects increasing informational distortion across biological systems, which is a gradient phenomenon.
Fuzzy boundaries do not invalidate NOVA. They reflect the nature of living systems and the reality of existence. The scientific question is not whether edges blur. It is whether the gradient predicts health. And so far, the data suggest that it does.
References:
Copeland, K. T., Checkoway, H., McMichael, A. J., & Holbrook, R. H. (1977). Bias due to misclassification in the estimation of relative risk. American Journal of Epidemiology, 105(5), 488–495. https://doi.org/10.1093/oxfordjournals.aje.a112408
Lakoff, G. (1987). Women, fire, and dangerous things: What categories reveal about the mind. University of Chicago Press.
Monteiro, C. A., Cannon, G., Levy, R. B., Moubarac, J.-C., Louzada, M. L. C., Rauber, F., Khandpur, N., Cediel, G., Neri, D., Martinez-Steele, E., Baraldi, L. G., & Jaime, P. C. (2019). Ultra-processed foods: What they are and how to identify them. Public Health Nutrition, 22(5), 936–941. https://doi.org/10.1017/S1368980018003762
Monteiro, C. A., Cannon, G., Moubarac, J.-C., Levy, R. B., Louzada, M. L. C., & Jaime, P. C. (2018). The UN Decade of Nutrition, the NOVA food classification and the trouble with ultra-processing. Public Health Nutrition, 21(1), 5–17. https://doi.org/10.1017/S1368980017000234
Pagliai, G., Dinu, M., Madarena, M. P., Bonaccio, M., Iacoviello, L., & Sofi, F. (2021). Consumption of ultra-processed foods and health status: A systematic review and meta-analysis. British Journal of Nutrition, 125(3), 308–318. https://doi.org/10.1017/S0007114520002688
Rosch, E. (1973). Natural categories. Cognitive Psychology, 4(3), 328–350.https://doi.org/10.1016/0010-0285(73)90017-0
Rosch, E., & Mervis, C. B. (1975). Family resemblances: Studies in the internal structure of categories. Cognitive Psychology, 7(4), 573–605. https://doi.org/10.1016/0010-0285(75)90024-9
Rothman, K. J., Greenland, S., & Lash, T. L. (2021). Modern epidemiology (4th ed.). Wolters Kluwer.
Schnabel, L., Kesse-Guyot, E., Allès, B., Touvier, M., Srour, B., Hercberg, S., & Buscail, C. (2019). Association between ultra-processed food consumption and risk of mortality among middle-aged adults. JAMA Internal Medicine, 179(4), 490–498. https://doi.org/10.1001/jamainternmed.2018.7289
Srour, B., Fezeu, L. K., Kesse-Guyot, E., Allès, B., Méjean, C., Andrianasolo, R. M., Chazelas, E., Deschasaux, M., Hercberg, S., Galan, P., & Touvier, M. (2019). Ultra-processed food intake and risk of cardiovascular disease: Prospective cohort study (NutriNet-Santé). BMJ, 365, l1451. https://doi.org/10.1136/bmj.l1451
