Ultra-Processed Food Debate: Scientists Urge Caution Before Rewriting Dietary Policy

A fierce scientific dispute over ultra-processed foods has erupted into the open, with a team of epidemiologists and causal inference specialists challenging one of the most influential nutrition publications of recent years. Writing in BMC Medicine, researchers led by Daniel B. Ibsen of Aarhus University argue that a high-profile three-part series published in The Lancet in November 2025 overstated the certainty of the evidence linking ultra-processed foods to chronic disease, and that public health policy built on that evidence risks repeating some of nutrition science’s most embarrassing historical mistakes. The response, published as a Debate article, does not dispute that diet is a major determinant of health. Instead, it takes aim at a more specific and technically consequential question: whether the category of ultra-processed food, as currently defined, is a coherent enough exposure to support strong causal claims and sweeping policy interventions.

The Lancet series, spearheaded by Carlos Monteiro and colleagues, advanced the hypothesis that ultra-processed foods are displacing long-established dietary patterns, degrading diet quality and driving up rates of obesity, type 2 diabetes, cardiovascular disease and other chronic conditions. The authors judged that the evidence fulfilled seven of the nine Bradford Hill criteria, the classical framework used to infer causation from observational data, and called for policies promoting dietary patterns based on whole foods. Crucially, they urged that the need for further evidence should not delay public health action. Ibsen and his co-authors, including Michael Fridén, Georgia D. Tomova, Natalia Ortega and Peter W.G. Tennant, counter that these are strong claims which underestimate the importance of grounding policy in established evidence that transparently acknowledges uncertainty. Their argument is not a defence of industrial food production but a methodological critique with real stakes for how governments regulate diets.

The core of the critique concerns the Nova classification system, the four-category framework that places foods in group 4, the ultra-processed category, described by Monteiro and colleagues as branded, commercial formulations made from cheap ingredients extracted or derived from whole foods and combined with additives. The BMC Medicine authors accept that this describes a commercially coherent cluster of products, but insist it is not biochemically coherent. The category sweeps together foods with long-established evidence of harm, such as processed meat, sugar-sweetened beverages and sugary foods, alongside items with far weaker evidence of harm, including many dairy products and breads. It also conflates well-characterised nutritional problems, such as excess fat, salt, sugar and calorie intake, with speculative and poorly understood concerns about food texture, industrial processing and additives. In causal inference terms, ultra-processed food is a multidimensional exposure, and the authors argue it is arguably fundamentally unsuitable for causal enquiry in its current form.

This heterogeneity has practical consequences that ripple through the research literature. Because studies of ultra-processed foods typically compare broad intake categories, they produce averaged, context-specific results that may dilute the large effects of genuinely harmful foods while simultaneously incriminating foods or food features that are neutral or even beneficial. A statistical association observed for the category as a whole may be driven almost entirely by its worst members, yet the resulting headline implicates everything from fizzy drinks to wholegrain industrial bread. The authors warn that this makes it extremely difficult to judge how past research translates into policy in different food environments, particularly where certain foods and features, such as salt content or sugary drinks, are already subject to dietary guidelines and legislation. In other words, the bluntness of the exposure measure risks misdirecting interventions that could otherwise be sharply targeted.

The second pillar of the critique concerns mechanism. Monteiro and colleagues argued that firm mechanistic knowledge is not necessary for immediate public health action, invoking historical examples where beneficial interventions preceded complete biological understanding. The response does not reject that logic outright, but points out an awkward fact: many of the mechanisms proposed to explain the harms of ultra-processed foods are not unique to ultra-processing. High energy density, one of the leading proposed pathways to weight gain and chronic disease, is found in homemade chocolate cakes just as much as in factory-made ones. Soft food texture, another candidate mechanism that may accelerate eating speed and increase energy intake, characterises fruit smoothies and homemade bread as well as industrial products. Evidence from randomised trials suggests that texture, rather than ultra-processing itself, could be a key driver of higher energy intake, and a controlled feeding study in which ultra-processed and non-ultra-processed meals were matched for hyper-palatability, energy density and texture found no difference in ad libitum energy intake among US adolescents eating from a buffet.

Recent randomised evidence adds further nuance. In a crossover trial comparing two diets that both adhered to UK dietary guidelines, one based mostly on minimally processed foods and one mostly on ultra-processed foods, both groups lost weight over eight weeks, with the minimally processed group achieving a modestly greater loss of 0.96 kilograms. That result suggests processing level is not the dominant determinant of outcome when overall diet quality is held constant. The authors also challenge a hypothesis advanced by Monteiro and colleagues in response to earlier criticisms: that while a Nova 4 food is not automatically worse than any food in a lower category, the ultra-processed version of a particular food will always be worse than its less processed equivalent. The response notes this hypothesis is currently unsupported by evidence, because relatively few studies compare individual ultra-processed foods with direct non-ultra-processed equivalents, and not all of those that do find the ultra-processed option worse. Comparisons of plant-based meat with unprocessed red meat, plant-based milk with cow’s milk, and artificially sweetened beverages with water have observed no short-term differences in health outcomes.

Underlying all of this, the authors identify a deeper structural problem with the evidence base: most studies compare Nova 4 foods with Nova 1 minimally processed foods, a contrast that bundles processing effects together with profound differences in nutrient composition, energy density, cost and social context. If ultra-processing and industrial additives are truly independent causes of adverse health, the authors argue, that effect should be detectable in comparisons between Nova 4 and similar Nova 3 processed foods, which differ more narrowly in degree of processing. Such research would also be far more directly useful for population health, offering straightforward paths to reformulation and dietary guidance rather than asking populations to reverse the deep social and economic changes that have reshaped food environments. Expecting people to substitute shop-bought food with home-cooked meals from minimally processed ingredients, the authors suggest, is unrealistic for those whose livelihoods do not support it, and any isolated policy premised on that assumption is unlikely to succeed.

The way forward, they propose, is disaggregation. Future research should move away from treating ultra-processed foods as a monolithic category and instead examine specific subtypes and features, such as texture, energy density and particular additives. Monteiro and colleagues have explicitly argued against subtype analyses, citing methodological challenges including confounding and measurement bias, but the response contends these challenges are not solved by studying monolithic variables. Indeed, the opposite may be true. Eliminating confounding between groups with very different exposure profiles, such as the highest versus lowest quartile of ultra-processed food intake, is far harder than between more modest contrasts, such as processed versus unprocessed red meat. Moreover, creating a summary variable from a series of mismeasured components can exacerbate measurement error when the errors are correlated, as is typical in dietary data, meaning the composite Nova variable may be less reliable than its parts.

The authors are careful to draw a clear line between critique and complacency. They state that the consumption of soft, hyper-palatable, energy-dense foods high in fat, sugar and salt is almost certain to be harmful, that many such foods are ultra-processed, and that for many people reducing ultra-processed food intake will probably be beneficial by cutting exposure to exactly those products. What remains unclear, they emphasise, is whether ultra-processing confers additional, unique harms beyond these well-established properties. They also invoke research on science communication showing that acknowledging uncertainty protects rather than undermines public trust, and that unwarranted certainty is what most generates scepticism. Given nutrition science’s history of confident but mistaken causal claims, from antioxidant supplements onward, they argue that candour about the limits of the ultra-processed food evidence is a scientific necessity, not a rhetorical weakness.

The policy implications are concrete. Rather than a generic targeting of ultra-processed foods, the authors point to taxes on sugar-sweetened beverages, salt reformulation targets and front-of-pack labelling schemes as examples of acting on specific, well-characterised exposures that have meaningfully improved population health without requiring resolution of the contested question of whether processing itself is harmful. These approaches, they argue, offer clearer and more actionable paths than asking consumers to navigate the uncertain boundaries of the Nova classification on their own. The debate is unlikely to be settled quickly, but its resolution will shape dietary guidelines, food regulation and labelling policy worldwide, determining whether the next generation of nutrition policy is built on the broad brush of processing categories or on the sharper, slower work of disentangling which specific foods and features actually drive disease.

Subject of Research: Causal inference limitations of ultra-processed food research and implications for public health nutrition policy

Article Title: Public health policy should be grounded in established evidence that acknowledges uncertainty: a response to The Lancet series on ultra-processed foods

Article References: Ibsen, D. B., Fridén, M., Tomova, G. D., Ortega, N., & Tennant, P. W. (2026). Public health policy should be grounded in established evidence that acknowledges uncertainty: a response to The Lancet series on ultra-processed foods. BMC Medicine, 24(1), Article 522. https://doi.org/10.1186/s12916-026-05214-y

Image Credits: AI Generated

DOI: 10.1186/s12916-026-05214-y

Keywords: ultra-processed foods, Nova classification, nutrition epidemiology, causal inference, public health policy, dietary guidelines, food processing, energy density, food texture, The Lancet, BMC Medicine, diet-related chronic disease