Ultra-Processed Food

TL;DR. The strongest single piece of evidence in this area is a controlled feeding trial in which people ate about 500 calories a day more on an ultra-processed diet than on a minimally processed one, despite the two diets being matched for calories offered, sugar, fat, fibre, and macronutrients. They ate faster and gained weight. That result, plus a large and consistent body of observational data, is why the category matters. What is much less settled is why, and whether the NOVA classification that defines it is a good tool or a blunt one that puts wholemeal bread and cola in the same box.

1. The classification

NOVA, developed by Carlos Monteiro's group at the University of São Paulo, classifies food by extent and purpose of processing, not by nutrient content:

The practical tell for group 4: it contains ingredients you would not find in a domestic kitchen. Protein isolates, maltodextrin, hydrolysed protein, invert sugar, high-fructose corn syrup, modified starches, hydrogenated oils, emulsifiers, humectants, bulking agents, colour stabilisers, flavour enhancers.

How much people eat: ultra-processed food supplies roughly 55 to 60 percent of energy in the UK and US, around 30 percent in France and Spain, and under 20 percent in Italy and much of Asia.

2. The key trial

Hall and colleagues, Cell Metabolism, 2019. This is the study everything else in the field is measured against.

Twenty adults lived in a metabolic ward for four weeks, spending two weeks on an ultra-processed diet and two on a minimally processed one, in random order. Both diets were matched for presented calories, energy density, sugar, fat, fibre, sodium, and macronutrients. Participants were told to eat as much or as little as they wanted.

Results:

• On the ultra-processed diet, participants ate about 500 kcal a day more.
• They gained about 0.9 kg over two weeks and lost about 0.9 kg on the minimally processed diet.
• They ate faster: more calories per minute.
• Appetite hormones shifted: PYY down and ghrelin up on the ultra-processed diet.

This is a small, tightly controlled, randomised crossover trial with a real causal design. It is the reason "ultra-processed" is taken seriously rather than dismissed as a food-purity argument.

3. The observational evidence

Large and consistent. Meta-analyses covering millions of participants associate higher ultra-processed food intake with:

• Higher all-cause mortality (roughly 20 to 30 percent at the highest intakes)
• Obesity and weight gain
• Type 2 diabetes
• Cardiovascular disease
• Depression
• Colorectal cancer
• Inflammatory bowel disease

A 2024 BMJ umbrella review assessed the evidence across 45 meta-analyses and found the strongest (and still, in its terms, only "convincing") associations for cardiovascular mortality, type 2 diabetes, and common mental disorders.

The confounding problem is severe and must be stated. People who eat the most ultra-processed food are, on average, poorer, less educated, more likely to smoke, less active, and eat fewer vegetables. Statistical adjustment helps and cannot fully resolve it.

4. The proposed mechanisms

Nobody knows which of these matters most, and the honest answer is probably several at once.

1. Energy density and eating rate. Ultra-processed food is soft, requires little chewing, and delivers more calories per minute of eating. Satiety signals take 15 to 20 minutes to register, so faster eating means more consumed before the brake applies. This is the mechanism the Hall trial most directly supports.

2. Texture and the destruction of food structure. As Chapter 11 argues throughout, the physical matrix of food governs how fast its nutrients are released. Industrial processing systematically destroys that matrix.

3. Hyperpalatability. Specific combinations of fat plus sugar, or fat plus salt, at levels that do not occur in nature, appear to drive intake beyond energy need. Whole foods rarely combine high fat and high sugar; ultra-processed foods routinely do.

4. Additives. The evidence is patchy and specific. Some emulsifiers (carboxymethylcellulose, polysorbate-80) alter the gut microbiome and thin the mucus layer in mouse models, and a small human trial found CMC altered microbiota and metabolites. Some artificial colours have been associated with behavioural effects in a subset of children (the 2007 Southampton study led to EU warning labelling). This is an active area, not a settled one.

5. Displacement. Ultra-processed food crowds out fruit, vegetables, pulses, and whole grains. This may be a large part of the whole effect.

6. Nutrient profile. Ultra-processed foods are, on average, higher in free sugars, sodium, and saturated fat and lower in fibre and micronutrients. The Hall trial matched for these and still found an effect, which suggests it is not the whole story.

7. Packaging contaminants. Migration of plasticisers and other compounds from packaging (Chapter 93).

5. The criticisms of the category

These are serious and worth taking seriously, because the classification is now driving policy.

It is not about nutrition, and that cuts both ways. NOVA deliberately ignores nutrient content, which is the source of both its insight and its problems. By the classification:

• Wholemeal supermarket bread is ultra-processed (emulsifiers, enzymes, added gluten), and lard is not (group 2).
• A sugar-free plant-based burger with pea protein and B12 fortification is ultra-processed; a home-made suet pudding is not.
• Plain live yoghurt is group 1; the same yoghurt with fruit and thickener is group 4.

Classification is unreliable. Studies giving the same food lists to different trained coders find substantial disagreement, particularly for groups 3 and 4.

"Processing" is doing enormous work as a word. Pasteurisation, freezing, canning, and fortification are processing, and they are among the great public health achievements: they made food safe, available year-round, and affordable. Folic acid fortification prevents neural tube defects; iodised salt prevented intellectual disability at population scale.

The counterfactual matters. Telling people to avoid ultra-processed food is a recommendation with a cost: fresh cooking takes time, skill, equipment, and money, all of which are unequally distributed. A category that indicts most affordable, shelf-stable, convenient food without providing an accessible alternative risks being advice that only the comfortable can follow.

Not all group 4 foods are equal, and treating them as one thing is the central weakness. Wholemeal bread, baked beans, fortified breakfast cereal, and tinned soup sit in the same box as cola and confectionery. Several researchers have argued for subdividing the category, and some national guidelines have declined to adopt NOVA for exactly this reason.

6. What to do with all this

A defensible reading of the evidence:

The category is a useful signal, not a rule. High ultra-processed intake reliably predicts a diet that is worse in several ways at once. That makes it a good screening question, and a poor absolute prohibition.

Prioritise within the category. The evidence is strongest and most consistent for:

1. Sugar-sweetened drinks. The clearest single harm.
2. Processed meat. IARC Group 1 (Chapter 55).
3. Packaged snacks and confectionery, which are engineered for overconsumption and displace real food.
4. Ready meals and fast food, largely for sodium, energy density, and portion.

And weakest for:

• Tinned pulses, tinned fish, tinned tomatoes (all excellent food that happens to be processed).
• Wholemeal bread, plain yoghurt with added fruit, fortified breakfast cereals with good fibre.
• Frozen vegetables (Chapter 9).
• Baked beans, which are the classic edge case: group 4 by classification, and a genuinely useful source of fibre and protein.

Practical heuristics that survive the criticism:

• Read the ingredients list, not the front of the pack. Length and unfamiliarity are a reasonable proxy. Five recognisable ingredients is a different product from twenty-five.
• Ask what it displaces. A ready meal instead of a takeaway is a step up; a ready meal instead of a home-cooked meal is a step down.
• Slow down. Eating rate is the mechanism with the best direct evidence, and it is free to change.
• Judge drinks most harshly. Liquid calories are the least compensated for.
• Do not moralise about it. Ultra-processed food is not a moral failing, and the strongest predictors of eating a lot of it are time, money, and food environment rather than willpower.

7. The bottom line

• The NOVA classification sorts food by how industrially it was made rather than by what is in it, and ultra-processed food supplies more than half the calories in the UK and US.
• The Hall metabolic ward trial found people ate about 500 kcal a day more on an ultra-processed diet matched for nutrients, and gained weight. That is the strongest causal evidence in the field.
• Observational data consistently associate high intake with higher mortality, diabetes, and cardiovascular disease, with substantial and unresolvable confounding by income and lifestyle.
• The likeliest mechanisms are eating rate, destroyed food structure, hyperpalatability, and displacement of whole foods, probably acting together. Additives are an open question, not a settled harm.
• The category's weakness is that it puts wholemeal bread, baked beans, and cola in the same box. Use it as a signal, prioritise sugary drinks and processed meat, and read ingredient lists rather than applying the label as a rule.

Sources and notes

The NOVA classification is Monteiro et al., Public Health Nutrition, 2019. The metabolic ward trial is Hall et al., Cell Metabolism, 2019. Population intake shares follow national survey analyses, including Rauber et al. for the UK and Martinez Steele et al. for the US. The umbrella review of associated outcomes is Lane et al., BMJ, 2024, which graded evidence strength by outcome. Eating rate and energy density mechanisms follow Forde and colleagues' work at Wageningen. Emulsifier effects follow Chassaing et al., Nature, 2015, in mice and Chassaing et al., Gastroenterology, 2022, the small human CMC feeding trial. Southampton azo colour findings are McCann et al., The Lancet, 2007. Classification reliability problems follow Braesco et al., European Journal of Clinical Nutrition, 2022. Critiques of NOVA's construction follow Gibney's and Astrup's published objections. Phthalate levels in fast food follow Zota, Phillips, and Mitro, Environmental Health Perspectives, 2016.

Open questions. Whether ultra-processing itself causes harm, or whether the category simply tracks nutrient profile plus displacement plus socioeconomic position, is the central unresolved question. Inter- rater reliability of NOVA classification is poor enough that different research groups reach different intake estimates for the same populations.

👉 That completes the food. Next: what you drink.