the new state of play

Reflecting on a year of UPF and HFSS

Over the past year, we have both found ourselves in countless (and heated) conversations about ultra-processed foods (UPFs) and foods high in fat, sugar and/or salt (HFSS). Be it presenting at conferences, supporting researchers and organisations to make sense of emerging evidence, or helping manufacturers and retailers to navigate shifting regulatory requirements, the science and policy of UPF and HFSS are major discussion points. Much of our communication has involved translating new science, classification systems and policy developments for our networks, as the boundaries between ‘processing’, ‘formulation’ and ‘nutrient composition’ continue to blur.

The end of 2024 felt like a turning point. The concepts of UPF and HFSS were repeatedly pitted against one another, raising questions about whether nutrient-based models, which have long underpinned policy, are the only method of classifying food and drink for scientific or policy purposes.

This year has brought no shortage of developments, including new trials, policy impact studies and technology-driven classification methods, alongside growing momentum around mandatory reporting, data infrastructure and the operationalisation of concepts like UPFs, ‘markers of processing’ and ‘free sugars’.

After a year of speaking about UPF and HFSS at academic and industry events, to rooms often filled with professionals across sectors, it feels timely to reflect on how the narrative has evolved, which developments have shaped this year, and what questions are now emerging as these discussions move into 2026.

A year of scientific developments

Early in the year, two open letters captured the widening polarisation of the UPF debate. One, signed by academics and health organisation stakeholders, was directed at researchers who proposed a project to update the NOVA classification system calling for them to not use the term UPF. The other, signed largely by food scientists/technologists, dismissed the NOVA classification entirely, claiming it to be unscientific and unsuitable for evaluating health outcomes. The exchange highlighted a growing divide on whether NOVA should be refined or replaced, and more broadly, whether ‘processing’ should remain a metric of interest or be redefined through measurable attributes. Part of this debate surrounds differing interpretations and meanings of ‘processing’ across society, public health and food technology.

Shortly afterwards, the International Union of Food Science and Technology (IUFoST) published work proposing a new scientific definition of food processing, emphasising the measurable effects of individual processing steps on product attributes (eg, nutritional composition) while distinguishing the concept of ‘formulation’ as meaning ingredient selection. The work focused specifically on developing a quantitative metric for the degree of processing, largely excluding the purpose of processing captured by NOVA. Rather than placing foods into single categories, this approach moves towards analysing specific impacts.

Numerous other classifications have been and are being developed based on different underlying rationales. This raises the question of whether a large number of diverging classifications will ultimately limit the applicability and uptake of any one particular new model.

In April, the UK Scientific Advisory Committee on Nutrition (SACN) published its updated rapid evidence review on processed foods and health. It reaffirmed consistent associations between UPF consumption and adverse health outcomes, but reiterated that most evidence remains observational, with potential confounding by energy intake, socioeconomic status and lifestyle. A key theme was the importance of UPF subgroup analysis. Sweetened drinks, processed meats and animal products showed consistent risk associations, while vegetarian alternatives showed little or no risk and categories such as bread, dairy and sauces showed mixed findings.

SACN’s research recommendations signalled a shift towards sub-classifying UPFs not only by processing but also by nutritional composition, data transparency on additives and processing methods, and refining National Diet and Nutrition Survey (NDNS) methods to improve UPF intake estimation without creating unnecessary data burdens.

In May, WHO announced plans to convene an expert group to develop formal guidance on UPF consumption, suggesting global engagement with processing-based concepts, not only in research, but in monitoring and policy. The composition of the group, published in November, largely encompassed academics across public health, potentially to focus on the ‘purpose of processing’ aspect. However, calls have been made to consider a wider breadth of experts, particularly food technologists, to better reflect backgrounds in the ‘degree of processing’.

Emerging certification schemes

During the year, two US-based non-profit certification schemes – Non-UPF Verified and Non-UPF Certified – tested different approaches to communicating processing to consumers. While one adopts detailed criteria based on processing steps, ingredients and nutrient thresholds, the other remains closer to NOVA, primarily focusing on the presence or absence of ingredients/additives “not commonly found in your kitchen”. Their emergence reflects growing demand and regulatory pressure for product-level transparency and the use of processing-based markers in consumer communication.

Public views on UPF

Given the rapid increase in media exposure regarding UPF, understanding public views is important. Therefore, UK Research and Innovation (UKRI) conducted a public dialogue on UPF. The aim of this report was to understand public opinions about UPF, including how they affect our health as well as their governance and regulation. The dialogue was completed this year, with draft findings shared with stakeholders across academia, government, public dialogue participants and health organisations in September (which Dr Dicken attended). The full report is due to be published in early 2026 and is intended to help shape future research and policy.

From dietary surveys to product databases: scaling food classification

One of the clearest challenges this year has been bridging the gap between two fundamentally different approaches to food classification. On one hand, expert-led classification of dietary surveys based on category-level or qualitative descriptors (eg, coding dietary survey data with the NOVA classification or Eatwell Guide recommendations) remains central to epidemiological research. On the other, product-level classification systems are emerging that use quantitative, industry-generated data such as nutrient composition, ingredient lists, additives and markers of processing, enabling food classification at scale.

Technology-driven approaches to classification

Recent years have seen advances in machine learning applied to food classification, leading to the development of AI-based systems such as FProX, which in 2025 was applied to a large dataset over 50,000 food items. Building on this momentum, newer technology-driven tools are emerging that integrate data from ingredients lists and nutrition labels to generate more nuanced classifications. For example, WiseCode differentiated the level of processing for 100,000 commercially available foods and identified distinct subgroups within the UPF category.

Alternative approaches have also been proposed. Ares et al. (2025) used data-driven clustering, grouping products based on nutritional composition and food additive disclosures by identifying clusters associated with the use of specific additives. However, the clusters did not consistently align with nutrient profiling approaches, showing the complexity of bringing together processing-based and nutrient-based classification.

Meanwhile, mobile food-profiling apps continue to proliferate, though their scoring systems often diverge from one another and from recognised nutrition policies. For example, ZOE released a new processed-food scoring method that evaluates energy intake rate (how quickly calories are consumed), and hyperpalatable- (the likelihood a food promotes overeating) and additive-related risks, placing products into one of five “risk” categories.

Untangling potential mechanisms

In the second half of 2025, we also saw a number of multi-day randomised controlled trials (RCT) assessing the health impact of UPF being published or results shared at scientific conferences. These include Preston et al. (2025), Rego et al. (2025) and Vaezi et al. (2025).

UPDATE was the first free-living trial to assess the health impact of UPF vs minimally processed food (MPF), the longest trial to date (eight weeks on each diet), and the first to assess the health impact of UPF in the context of healthy dietary guidance. Both diets resulted in significant weight loss from the nutrient-poor, high UPF baseline diets, but the MPF diet resulted in significantly greater weight loss, significant fat mass loss and favourable changes in appetite and craving measures.

This year also saw a move towards mechanisms of UPF, including the RESTRUCTURE trial and long-awaited results of Kevin Hall’s second metabolic ward trial.

At the American Society for Nutrition conference in Florida, the RESTRUCTURE trial highlighted how differences in texture and eating rate can significantly moderate intake from UPF across two-week diets, although there were notable differences between diets in macronutrients, including 50 percent differences in fat, 133 percent in saturated fat, 25 percent protein and a carbohydrate:fat ratio of 0.63 versus 1.07.

Kevin Hall presented the final results of his second UPF trial at ObesityWeek in Atlanta. The results highlighted the key role of energy density and hyperpalatable food in driving excess energy intake from UPF across seven-day diets. However, there was no difference in eating rate seen across UPF and MPF arms, and only the MPF diet resulted in fat loss, similar to the UPDATE trial. The results are due to be published in 2026.

This raises the question of whether these attributes can be quantified and scaled up to large datasets or nutrient databases to better understand how they link with the purpose of processing. For example, Dicken and Brown recently reported that in the UK National Diet and Nutrition Survey nutrient database, UPFs were significantly more energy dense than MPFs, which was consistent across different types of food, as well as when only comparing foods without red front-of-pack traffic lights.

In November, The Lancet launched a UPF series of three key papers. The first addressed the health impact of UPF, the second outlined policy options and the third covered barriers to change, highlighting financial aspects of the food system and corporate political activity of multinational food corporations. The series focused on the wider food systems aspect of the NOVA classification and called for action to prioritise UPF as a global issue, including the development of multi-level coalitions and collective action.

UPF & HFSS: exploring overlap

In September, at the Society for Social Medicine Conference in Bradford, new analyses using product-level nutrition and ingredient data further explored the overlap between UPF and HFSS classifications. Wallis et al. (2025) found that most (88 percent) of ~900 products sampled from categories in scope of UK HFSS promotion, placement and advertising restrictions were considered NOVA 4 (UPF), yet only 52 percent failed the UK Nutrient Profiling Model (NPM) and would be subject to regulation.

Complementing this, Williams et al. (2025) presented similar product-level data to a much larger dataset of ~90,000 UK food products, finding that 86 percent were classified as UPFs, yet only 41 percent were both HFSS and UPF – again highlighting divergence between the two systems, particularly in categories such as canned/tinned foods, fish/seafood and fruit juices/smoothies.

At an even larger scale, Nesta published analysis of over 3.3 million food transactions, similarly demonstrating partial overlap – almost two-thirds of UPF calories purchased (64 percent) come from HFSS products.

Policy momentum and the shifting role of data

Under the new US administration, the US Food and Drug Administration and US Department of Agriculture jointly launched a federal Request for Information on defining UPF, marking the first time that US agencies have formally engaged with the concept in a regulatory context. This followed a CDC published report on UPF intake from NHANES. California went further, passing legislation that over the next decade will restrict certain UPFs from schools, using a combined definition that incorporates both markers of ultra-processing and thresholds for nutrients of public health concern. New York took a different approach for meals and snacks served by public institutions: rather than trying to define UPF, they instead focused on what must be included as a minimum (whole, minimally processed foods), implicitly excluding UPF. And just in December, the city of San Francisco is suing a number of large ultra-processed food companies.

In the UK, HFSS policy advanced to the implementation stage. New volume-based placement restrictions came into force in October 2025, with advertising restrictions expected early 2026. The DIO Food team at The University of Leeds published the first researcher-led evaluation of implementation; the legislation reduced sales of in-scope HFSS products as a proportion of total sales. When scaled nationally, this equates to around 2 million fewer in-scope HFSS items sold per day. In addition, work from Dhuria et al. (2025) captured industry perspectives. Both highlighted how the legislation had the potential to shift retailer priorities from solely profit maximisation, to also supporting public health.

Alongside this, the government confirmed its intent to introduce a Healthy Food Standard, requiring businesses to report on HFSS sales and health metrics. This move brings to the spotlight the role of data infrastructure and the need for consistent underlying classification systems; a conversation that is expected to be a big part of 2026, alongside data challenges on potential updates to the UK Nutrient Profiling Model (NPM).

Updating the UK Nutrient Profiling Model

The proposal to adopt a new UK NPM, potentially based on the 2018 review incorporating free sugars, has been suggested as part of government plans. While scientifically aligned with SACN’s recommendations on free sugars and fibre, it would introduce significant operational challenges. Free sugars cannot be directly analysed and may need to be estimated using category assumptions based on ingredient lists. Manufacturer-level calculations may therefore diverge from retailer or researcher-level approaches due to the number of assumptions that the suggested draft model relies upon. Implementation at scale would require advanced data science approaches and agreed methods of estimation and this could become the defining technical bottleneck of 2026 for UK policy.

Where are we headed?

Looking ahead to 2026, we expect data challenges to come into sharper focus. The conversation on classification is shifting from conceptual debates to questions of feasibility and implementation. It is less about how we define UPF or HFSS, and more about applying these definitions in practice. What data do we have? What data do we need? And how do we build systems that allow different models to be applied consistently at scale?

Core questions are becoming clearer: can NOVA, or any processing-based model, be operationalised across retail, manufacturing, purchase or consumption datasets? How do markers of (ultra-)processing align with nutrient profiling and other established metrics? Do ingredient metrics and the presence of specific additives provide meaningful insight into product healthfulness, or are we now at a stage where we need AI-driven methods applied to large datasets to uncover more complex signals and variations in product attributes? And fundamentally, do we need more comprehensive datasets and transparency on product ingredient and sales reporting to do this?

These questions are no longer confined to academic discussions; they are now shaping regulatory approaches, industry strategies and upcoming policy developments.

One thing is for sure, the UPF concept doesn’t seem to be going anywhere. But it feels clear that the next chapter of UPF and HFSS will be data-led. The challenge now is not simply how to classify, but how to do so consistently, based on robust underlying data, in a way that is transparent, reliable and scalable across datasets that are currently disconnected.