Most of us know, in a general way, that heavily processed food is not great for health. But “not great” is vague, and vague warnings rarely change behaviour at the supermarket checkout. What changed the conversation was a 2018 study published in the BMJ that put a precise number on the risk – and that number was striking enough to prompt governments, dietitians, and food regulators worldwide to take a second look at how food is actually classified.
The study, led by Thibault Fiolet and colleagues at the French National Institute of Health and Medical Research (INSERM), followed 104,980 adults in the NutriNet-Sante cohort from 2009 to 2017. Using repeated 24-hour dietary records covering more than 3,300 individual food items, the researchers categorised every item each participant ate according to a four-tier food processing framework called NOVA. The headline finding: each 10% increase in the proportion of ultra-processed food in the diet was associated with a 12% higher overall cancer incidence (hazard ratio 1.12, 95% CI 1.06-1.18) and an 11% higher breast cancer risk (HR 1.11, 95% CI 1.02-1.22). These were statistically robust associations that held after adjusting for total calorie intake, physical activity, BMI, smoking, and educational level.
Understanding what that finding means in practice starts with understanding the NOVA system itself – because the word “processed” is doing a lot of heavy lifting in public health conversations, and it is frequently used to mean very different things.
What NOVA Actually Classifies – and Why Group 4 Is Different
NOVA was developed by Brazilian epidemiologist Carlos Monteiro at the University of Sao Paulo and formalised in a landmark 2019 paper in Public Health Nutrition. Unlike older classification systems that sort foods by nutrient content or food group, NOVA classifies by the nature, extent, and purpose of industrial processing. It divides foods into four groups.
Group 1 is unprocessed or minimally processed foods: fresh vegetables, fruit, meat, fish, eggs, legumes, plain milk. Group 2 covers processed culinary ingredients: butter, flour, sugar, salt, oils – substances extracted from Group 1 foods and used in home cooking. Group 3 is processed foods: tinned fish, cheese, cured meats, pickled vegetables – foods made by adding Group 2 ingredients to Group 1 foods, typically to extend shelf life. Groups 1 through 3 have all existed throughout human culinary history.
Group 4 is the outlier. Ultra-processed foods are defined as industrial formulations made mostly or entirely from substances derived from foods – often chemically modified – with additives whose purpose is to make the final product hyper-palatable, visually appealing, and shelf-stable. These include emulsifiers, stabilisers, humectants, artificial flavours, colour enhancers, and non-sugar sweeteners. Critically, many of these additives have no culinary equivalent and would never appear in a home kitchen. The product is designed in a laboratory, not a kitchen.
Two Additives Under the Microscope: Carrageenan and Carboxymethylcellulose
Among the additives that have attracted scientific scrutiny, carrageenan (E407) and carboxymethylcellulose (CMC, E466) stand out because of the volume of mechanistic research linking them to gut inflammation.
Carrageenan is a polysaccharide derived from red seaweed. It is widely used as a thickener and stabiliser in dairy alternatives, chocolate milk, infant formula, deli meats, and salad dressings. Research published in Frontiers in Pediatrics in 2017 by Bhatt and colleagues reviewed both animal and cell-culture evidence and found that carrageenan disrupts the intestinal epithelial barrier by reducing the density of zonula occludens-1 (ZO-1) tight junction proteins, activates pro-inflammatory pathways including toll-like receptor 4 (TLR4) and nuclear factor kappaB (NF-kB), and stimulates interleukin-8 secretion. In animal models, degraded carrageenan produced ulcerative colitis-like lesions in 100% of guinea pigs within 30 days. Germ-free animals developed no lesions at all – pointing to the gut microbiome as an essential mediator of the inflammatory response.
CMC, used as a thickener in ice cream, breads, and plant-based foods, works through a related but distinct mechanism. Mouse studies found that CMC erodes the mucus layer separating luminal bacteria from the intestinal epithelium, allowing bacteria to colonise territory they normally cannot reach. This microbiota-mediated barrier disruption promotes low-grade systemic inflammation. In genetically susceptible mice lacking the anti-inflammatory cytokine IL-10, CMC exposure produced Crohn’s-like pathology. Human trial data remains limited, but the mechanistic picture is consistent across multiple experimental systems.
Reading the Label Through a NOVA Lens
The practical challenge is that food labels are not designed to make NOVA classification easy. Regulatory requirements vary, and nutrient panels do not distinguish between a food that is simply salted (Group 3) and one that contains ten synthetic additives (Group 4). The most reliable signal is the ingredients list, not the nutrition facts panel.
Group 4 foods typically reveal themselves through specific markers in the ingredients list. Any of the following, particularly in combination, place a product firmly in the ultra-processed category: modified starch (any variety), protein isolate or hydrolysate, high-fructose corn syrup or glucose-fructose syrup, hydrogenated or interesterified fats, artificial flavours or “natural flavours” produced by industrial extraction, emulsifiers such as soy lecithin (E322), polysorbate 80 (E433), carrageenan (E407) or CMC (E466), and synthetic colour additives (any E number in the 100-199 range). A useful working rule: if the ingredients list contains more than five items, or any item that sounds like a chemistry laboratory rather than a kitchen, the product is likely ultra-processed.
This is not the same as saying all additives are equally harmful, or that ultra-processing is the only dietary variable that matters. The NOVA framework is a classification system, not a toxicological ranking. But as Monteiro’s team argued in Public Health Nutrition, when a product type is consistently associated with adverse health outcomes across multiple large cohort studies, the classification itself becomes epidemiologically useful, regardless of whether the individual additive mechanism is fully established.
Beyond Cancer: The Broader Evidence Base
The Fiolet BMJ study was not the first to link ultra-processed food consumption to health harm, and the evidence has continued to accumulate since. A 2023 systematic review and meta-analysis published in eClinicalMedicine (The Lancet family) drawing on the UK Biobank cohort found associations between ultra-processed food intake and higher risks of type 2 diabetes, cardiovascular disease, obesity, and overall mortality – not just cancer. A 2023 Lancet Regional Health Europe multinational study across nine European countries confirmed that ultra-processed food consumption raised the risk of multimorbidity, meaning the simultaneous occurrence of cancer alongside cardiometabolic conditions.
What is notable about all of these studies is that the associations hold after adjustment for overall dietary quality scores. In other words, ultra-processed foods appear to cause harm over and above whatever nutrient imbalances they introduce. The additives, the degree of industrial manipulation, or some combination of characteristics unique to Group 4 foods may be driving independent risk. That is precisely the mechanistic hypothesis that the carrageenan and CMC research is attempting to test at the cellular level.
Where This Leaves Practical Dietary Choices
Brazil, which is where NOVA was developed, updated its national dietary guidelines in 2014 to make “avoid ultra-processed foods” one of its ten core recommendations – a move widely praised by nutritional epidemiologists and criticised by the food industry. The UK, France, and Canada have since issued guidance that references ultra-processed food as a category to limit, even without using the NOVA label explicitly.
For anyone trying to act on this evidence in daily life, a few shifts have outsized impact. Replacing ultra-processed breakfast cereals with oats, fresh fruit, or eggs removes several Group 4 products without requiring calorie counting or macro tracking. Choosing plain tinned fish, legumes, or cheese over packaged deli-style products swaps Group 4 for Group 3 with no loss of convenience and considerable gain in protein quality. Reading the ingredients list rather than the front-of-pack claims shifts attention away from marketing language (“high protein,” “low fat,” “wholegrain”) toward the actual industrial inputs. None of this demands perfection – occasional ultra-processed food consumption in an otherwise whole-food diet is unlikely to carry the risks observed in people for whom these products form the majority of daily calories. What the EPIC data tells us is that the proportion matters, and that reducing it consistently over time is what moves the needle on long-term cancer incidence.
This article is for general informational purposes only and does not constitute medical or dietary advice. Consult a qualified healthcare professional or registered dietitian for personalised guidance.