What 'Ultra-Processed' Can and Cannot Prove

The field in brief

Nutritional epidemiology mostly watches rather than experiments. Its workhorse is the observational cohort study, in which tens of thousands of people record what they eat and are then followed for years while researchers count who develops diabetes, heart disease, or death [8]. The method’s curse is confounding: the person who eats more packaged snacks also tends to smoke more, move less, earn less, and sleep worse, so any illness that follows could belong to the company the food keeps rather than the food itself [8]. Statisticians adjust for the confounders they can measure, but healthy-user bias, the tendency of health-conscious people to do many healthy things at once, and residual confounding, the pull of everything left unmeasured, never fully clear [8]. The one design that would settle causation, the randomized controlled trial or RCT, is brutally hard for diet, because you cannot lock thousands of people in a lab and govern every bite for years [1].

Into this uncertainty came a classification that ignored nutrients entirely. Around 2009, Carlos Monteiro and colleagues at the University of São Paulo proposed NOVA, a scheme that sorts food not by its salt or calories but by how much industrial transformation it has undergone [3]. Its four groups run from unprocessed or minimally processed foods, through processed culinary ingredients and processed foods, to ultra-processed foods, the contested fourth group [3]. NOVA defines that group as industrial formulations built largely from substances extracted from foods, such as protein isolates and modified starches, plus additives like emulsifiers, dyes, and flavorings of no or rare culinary use, engineered for palatability and shelf life [3]. A home-baked loaf counts as processed; a wrapped supermarket loaf with emulsifiers counts as ultra-processed, even when their nutrition labels look nearly identical. That is the crux the whole fight turns on: NOVA claims processing carries risk beyond what the nutrients explain [6].

The fight

For years the ultra-processed idea lived mostly in correlations, easy to wave away. Then in 2019 it acquired an experiment. Kevin Hall, a physicist turned metabolism researcher at the US National Institutes of Health, admitted twenty adults to a metabolic ward, a hospital setting where every calorie eaten is weighed [1]. For two weeks each, in random order, they were offered an ultra-processed diet or a minimally processed one, the two matched for calories, sugar, fat, fiber, and salt as presented, and told to eat as much or as little as they liked [1]. On the ultra-processed weeks people ate about 508 calories a day more and gained roughly two pounds; on the minimally processed weeks they lost about the same [2]. Nothing in the matched nutrients explained it, and something about the processing itself seemed to drive the overeating [1]. It was small, but it was the rare prize the field almost never wins: a controlled cause [1].

The result made Hall the reluctant champion of a framework he had not invented, and it drew fire. In early 2021 independent data sleuths publicly questioned the trial’s suspiciously tidy weight figures; Hall traced the oddities to pre-weighed hospital pajamas subtracted at each weigh-in, and both sides agreed most concerns dissolved on inspection [4]. The deeper objection was never about arithmetic. It was about the category.

Its most prominent critic is Arne Astrup, a Danish physician and nutrition scientist, who in a 2022 point-counterpoint in the American Journal of Clinical Nutrition argued no: that NOVA adds little to existing nutrient-based advice and misfires badly at its edges [5]. The same food, he noted, can land in different groups depending on where it was assembled, while wholesome items such as whole-grain bread, plant-based milks, and infant formula get tarred as ultra-processed alongside candy [5]. Across the table sat Monteiro himself, arguing yes: that nationally representative studies keep linking ultra-processed intake to worse diets and worse health, and that guidelines should say so plainly [6][7].

The observational evidence has only grown heavier. A 2024 umbrella review in The BMJ, led by Melissa Lane, pooled forty-five meta-analyses covering nearly ten million people and tied higher ultra-processed intake to thirty-two health outcomes, with the strongest, convincing-grade evidence for cardiovascular death, anxiety, and type 2 diabetes [8]. The authors were careful to note that umbrella reviews sit atop observational data, so unmeasured confounding cannot be excluded [8]. That caveat is the entire argument in miniature.

Then the fight left the journals. In the United States, ultra-processed foods now supply about 55 percent of the calories people eat, and more than 60 percent of what children eat, by a 2025 analysis from the Centers for Disease Control and Prevention [9]. Yet when the 2025 Dietary Guidelines Advisory Committee assembled its scientific report, it declined to make ultra-processed foods a formal recommendation, citing the lack of a single agreed definition [9]. The absence of a definition had become a policy veto.

The dispute then reached inside the government that funds it. In April 2025 Kevin Hall took early retirement from the National Institutes of Health at fifty-four, saying officials had constrained how he could discuss his own findings, because the results did not fully support a favored account of food as addictive [10]. The agency disputed that characterization [10]. The field’s leading experimentalist on ultra-processing left public science just as federal food policy was taking the subject up.

His last major experiment complicated everyone. A 2025 crossover trial in Nature Medicine, run at University College London with Hall’s help, fed fifty people two diets that both met UK healthy-eating guidelines, one minimally processed and one ultra-processed [11]. Both produced weight loss, but the minimally processed diet produced modestly more, about two percent of body weight against one, alongside fewer cravings [12]. Processing mattered, and yet a well-formulated ultra-processed diet was no catastrophe, a result neither camp could bank as a clean win [11].

What the fight reveals

The dispute exposes the central bind of nutritional epidemiology: its most abundant evidence is its least conclusive. Observation can watch millions but never fully isolate a cause; experiment can isolate a cause but only in twenty or fifty people for a few weeks [8][1]. NOVA forced the field to argue about something it had long ignored, that how food is built might matter apart from what it contains, and that reframing has proven useful even where the category stays blurry at its edges [6].

On the narrow question, the evidence currently leans one way. The controlled trials, small as they are, point in the same direction as the mountains of correlation: matched for nutrients, more processed diets lead people to eat more and fare somewhat worse [1][11]. That convergence between the confounded observational data and the clean but tiny experiments is why defenders feel vindicated and why the concept has survived its critics.

What remains genuinely unresolved is the mechanism and the boundary. No one has yet shown what, precisely, about ultra-processing drives the overeating, with texture, energy density, additives, and sheer eating speed all still in contention, nor drawn a line that reliably separates a harmful formulation from a harmless one [1]. Until that line exists, ‘ultra-processed’ will remain a category that predicts harm without yet explaining it.