What counts as ultra-processed food?

Under the NOVA system, ultra-processed foods are industrial formulations typically made from substances extracted from foods — protein isolates, modified starches, hydrogenated oils — combined with additives such as emulsifiers, colours, sweeteners and flavour enhancers that are not found in a domestic kitchen. Typical examples include soft drinks, packaged snacks, confectionery, reconstituted meat products, instant noodles, most commercial breakfast cereals and mass-produced packaged bread. The classification is about how a product is made rather than what nutrients it contains, which is both its distinctive contribution and the source of most criticism of it — since it places plain wholemeal supermarket bread and infant formula in the same category as sweets.

How strong is the evidence that ultra-processed food is harmful?

Stronger than for most nutritional exposures, but incomplete in an important way. Large meta-analyses consistently find dose-dependent associations with obesity, diabetes, heart disease, depression and death, replicated across many countries. More importantly, an NIH inpatient trial that provided all meals and matched them for calories, sugar, salt, fat and fibre found that people spontaneously ate around 500 more calories a day and gained weight on the ultra-processed diet — which shows a causal effect on eating behaviour, not merely a correlation. What remains genuinely unknown is why. No mechanism by which processing itself damages health has been established, and this gap is the reason many national authorities have not regulated on processing as opposed to nutrient content.

What is the criticism of the NOVA classification?

That it groups together foods with very different nutritional profiles and health implications. Plain wholemeal bread from a supermarket, tinned baked beans, plain yoghurt with added cultures and infant formula all fall into the ultra-processed category alongside sweets and soft drinks. Critics argue that the observed harm is driven by a nutritionally poor subset — sugary drinks, confectionery, processed meat — and that regulating an entire heterogeneous category is neither scientifically justified nor legally defensible. Defenders reply that nutrient profiling has been tried and has not captured the effect, and that the NIH trial matching nutrients and still finding a difference is precisely the evidence that something beyond nutrients is at work. Both arguments come from serious researchers, and this is a genuine open question rather than a manufactured one.

Could this just be about poverty rather than food?

It is a serious objection and it cannot be fully excluded. Ultra-processed food intake tracks closely with lower income, less education, longer or more irregular working hours, limited cooking facilities and living in areas with poor food access — all of which independently worsen health. Statistical adjustment reduces this problem but never eliminates it, because the measured variables are imperfect proxies for the underlying disadvantage. What makes this less than a complete explanation is the inpatient trial, where all participants were the same people in the same conditions eating provided food, and the effect appeared anyway. The honest position is that both are probably operating, and this constrains how far the evidence supports blaming individuals rather than changing the food environment.

So what should I actually do?

The useful thing about this debate is that the practical answer does not depend on winning it. Whether the harm comes from the processing, from the nutritional profile of the products, or from the circumstances of the people eating them, the same actions follow: build meals around minimally processed foods — vegetables, legumes, fruit, whole grains, eggs, fish, plain dairy, nuts; cook from basic ingredients where time and facilities allow; and reduce reliance on packaged products engineered to be eaten quickly and in quantity. It is worth avoiding the trap of treating the category as a purity test, since bread, tinned beans, plain yoghurt and frozen vegetables are not the problem. The products worth displacing first are sugary drinks, confectionery and savoury snacks.

How does GMJ News cover ultra-processed food?

GMJ News covers the observational evidence base and its consistency, the NIH inpatient crossover trial and its limitations, the unresolved mechanism question and the leading candidates including hyperpalatability and emulsifier effects on the intestinal mucus layer, the methodological criticism of NOVA as a policy target, residual socioeconomic confounding, front-of-pack labelling and fiscal policy evidence from Chile and elsewhere, the emerging mental health literature and its reverse causation problem, and the practical advice that holds regardless of how the scientific question resolves. Part of PHIG's contested science and nutrition policy programme.

Home › Topics › Ultra-Processed Food

Ultra-Processed Food

GMJ News knowledge hub · last reviewed September 2026 · Georgian Medical Journal

Summary Statistics Glossary GMJ news FAQ Documents Organizations Research

Ultra-processed food is the rare contested nutrition topic where the disagreement is genuinely scientific rather than manufactured: meta-analyses of prospective cohorts consistently associate high UPF intake with obesity, cardiovascular disease, type 2 diabetes, depression and all-cause mortality, and a landmark inpatient crossover trial found that people ate around 500 additional calories per day and gained weight on an ultra-processed diet matched for macronutrients, sugar, salt and fibre (WHO). Yet serious methodological objections remain unresolved: the NOVA classification groups foods by degree of industrial processing rather than nutritional composition, placing wholemeal supermarket bread and confectionery in the same category; the mechanism by which processing itself causes harm is still not established, which is precisely why many national authorities have declined to act on processing as opposed to nutrient content; and residual confounding by socioeconomic position is difficult to exclude in observational data. This is a hub about how to reason when the evidence is strong but incomplete.

Key messages

SETTLED: high UPF intake is consistently associated with worse outcomes

Umbrella reviews of meta-analyses covering millions of participants find consistent associations between high ultra-processed food consumption and obesity, type 2 diabetes, cardiovascular disease, depression, adverse gastrointestinal outcomes and all-cause mortality. The associations are dose-dependent, replicated across countries and cohorts, and persist after adjustment for the obvious confounders. The consistency across outcomes and populations is stronger than for most nutritional exposures, and this is not seriously disputed by anyone in the field.

STRONG: the Hall inpatient trial demonstrated a causal effect on intake

The most important single study is a small but rigorous NIH inpatient crossover trial in which 20 adults ate ultra-processed or unprocessed diets, matched for presented calories, macronutrients, sugar, sodium and fibre, with all food provided and intake precisely measured. On the ultra-processed diet, participants spontaneously ate around 500 additional calories per day and gained approximately 0.9kg in two weeks, reversing on the unprocessed diet. This moves the question beyond correlation for at least one outcome, and the trial has been replicated in direction if not in magnitude.

GENUINELY OPEN: the mechanism by which processing itself causes harm

This is the crux of the scientific disagreement, and it is unresolved. Candidate mechanisms include energy density and hyperpalatability driving passive overconsumption; textural softness increasing eating rate; disruption of satiety signalling; effects of emulsifiers and additives on the intestinal mucus layer and microbiota; loss of food matrix structure altering nutrient absorption; and packaging-derived contaminants. Each has supporting evidence and none is established. Until a mechanism is demonstrated, many national authorities have declined to regulate on processing as distinct from nutrient composition — a defensible position, not obstruction.

CONTESTED: whether NOVA is fit for purpose as a policy instrument

NOVA classifies by degree and purpose of industrial processing rather than by nutritional composition, which produces genuinely awkward results: wholemeal supermarket bread, plain yoghurt with added cultures, tinned baked beans and infant formula sit in the same category as confectionery and soft drinks. Critics argue the category is too heterogeneous to support regulation and that the observed harm is driven by a nutritionally poor subset. Defenders argue that the grouping captures something real that nutrient profiling misses, and that the Hall trial supports this by matching nutrients and still finding an effect. Both positions are held by serious researchers.

UNRESOLVED: residual confounding by socioeconomic position

Ultra-processed food intake correlates strongly with income, education, food environment, working hours, cooking facilities and time poverty — all of which independently affect health. Statistical adjustment reduces but cannot eliminate this, and the possibility that UPF is partly a marker for disadvantage rather than solely a cause of disease remains live. This does not dismiss the findings; it constrains how confidently they can be converted into individual dietary advice as opposed to structural food policy.

WHAT FOLLOWS PRACTICALLY: the advice is robust even where the science is not

The useful feature of this debate is that the practical conclusion does not depend on resolving it. Whether the harm comes from processing itself, from the nutrient profile of the products, or from the circumstances of the people eating them, the same action follows: eat more minimally processed food, cook more where circumstances allow, and reduce reliance on packaged products designed for passive overconsumption. Policy is harder — taxation, marketing restriction and labelling all require defining the target category, which is exactly what remains contested.

Key statistics

~500 kcal/day

additional spontaneous intake on an ultra-processed diet in the NIH inpatient crossover trial

Cell Metab 2019

Dose-dependent

associations with obesity, T2DM, CVD, depression and mortality across umbrella reviews

BMJ/umbrella review

Mechanism unknown

no established mechanism by which processing per se causes harm — the central open question

Nature Food/Lancet

NOVA contested

groups wholemeal bread and confectionery in one category — the core methodological objection

Am J Clin Nutr

50%+

of dietary energy from UPF in several high-income countries, and rising rapidly in LMICs

PAHO/WHO

Advice robust

the practical conclusion holds regardless of which explanation proves correct

Editorial consensus

Ultra-processed food — where the disagreement actually lies

Source: Bars show strength of supporting evidence. This is a topic where the association is strong and the explanation is genuinely unresolved.

Glossary of key terms

NOVA classification
Nutritional scienceA four-group system classifying foods by the extent and purpose of industrial processing rather than by nutrient content. Group 1: unprocessed or minimally processed — fruit, vegetables, meat, milk, grains, eggs. Group 2: processed culinary ingredients — oils, butter, sugar, salt. Group 3: processed foods — tinned vegetables, cheese, freshly made bread, cured meats, made by combining groups 1 and 2. Group 4: ultra-processed — industrial formulations typically containing five or more ingredients including substances not used in domestic kitchens such as protein isolates, modified starches, emulsifiers, colours and flavour enhancers. The classification was developed for population monitoring in Brazil and has been adopted by PAHO and several national guidelines, while remaining contested as a basis for individual advice or regulation.

The Hall inpatient trial
Clinical researchConducted at the NIH Clinical Center, this is the study that changed the debate. Twenty weight-stable adults lived in a metabolic ward and were randomised to two weeks of ultra-processed and two weeks of unprocessed diets in crossover, with meals matched for presented energy, macronutrients, sugar, sodium and fibre, and were instructed to eat as much or as little as they wished. On the ultra-processed diet they ate around 500 kcal per day more and gained weight; the effect reversed on switching. The design controls for nutrient composition, which is why it is cited as evidence that something about processing matters beyond nutrients. Limitations are real: small sample, short duration, single centre, and the diets differed in ways that are difficult to match completely — but no comparably controlled study contradicts it.

Hyperpalatability and passive overconsumption
MechanismThe leading candidate mechanism. Ultra-processed products are formulated to combine energy density, softness, rapid oral processing and specific fat-carbohydrate-salt combinations that occur rarely in whole foods. Consequences include faster eating rate, which is independently associated with higher intake because satiety signalling lags ingestion by 15-20 minutes; reduced chewing and oral exposure time; and diminished post-ingestive satiety relative to energy delivered. This explains the Hall trial result without requiring any toxicological mechanism, and it is the explanation most consistent with the available data — though it also implies that the target should be product formulation rather than processing as a category.

Emulsifiers and the intestinal mucus layer
Emerging scienceThe most scientifically interesting of the additive hypotheses. Animal work has shown that common emulsifiers including carboxymethylcellulose and polysorbate-80 can thin the intestinal mucus layer, allow bacteria closer to the epithelium, alter microbiota composition and promote low-grade inflammation and metabolic syndrome in mice. Small human feeding studies have found microbiota and metabolomic changes with carboxymethylcellulose. This is genuinely preliminary: the animal doses and the human relevance are debated, the human studies are small and short, and clinical outcome evidence does not exist. It is the right kind of hypothesis — mechanistically specific and testable — and it should be described as promising rather than established.

Front-of-pack labelling and fiscal policy
PolicyChile's 2016 law combining black octagonal warning labels, marketing restrictions and school sales bans is the most studied natural experiment, with evaluations showing substantial reductions in purchases of labelled products and in advertising exposure. Mexico, Brazil, Peru, Israel and others have followed with variations. Sugar-sweetened beverage taxes have consistently reduced purchasing where implemented and evaluated. The unresolved policy question is whether regulation should target nutrient thresholds, which are objective and defensible in law, or processing categories, which may capture more of the harm but rest on a contested classification — a distinction that has shaped how each country has legislated.

Ultra-processed food and mental health
Emerging scienceA rapidly growing literature associates UPF intake with depression, anxiety and adverse cognitive outcomes, with dose-dependent relationships reported in several large cohorts. Proposed mechanisms include microbiota-gut-brain signalling, systemic inflammation, glycaemic variability and displacement of nutrients relevant to neurotransmitter synthesis. The interpretive difficulty is unusually severe here because the confounding runs in both directions: depression alters appetite, food choice, cooking capacity and income, so reverse causation is not merely possible but likely to contribute. The SMILES trial and subsequent dietary intervention studies in depression provide some support for a causal component, but the specific attribution to processing rather than diet quality remains unproven.