UCL finds inconsistencies in food classification systems

Researchers led by University College London have identified substantial inconsistencies between the methods used to classify foods as unhealthy in the United Kingdom, finding that different official approaches can produce sharply different assessments of the same products.

The study, published in BMJ Medicine, compared 1,898 packaged food and drink products using models based on the government's Eatwell Guide, voluntary front-of-pack traffic light labelling, and the Food Standards Agency Nutrient Profiling Model.

Eatwell-aligned assessments classified more than 95% of the processed products and more than 98% of the ultra-processed products examined as unhealthy, while the existing labelling and regulatory methods identified substantially smaller proportions.

The voluntary traffic light system identified 48.2% of the surveyed products as unhealthy. The nutrient profiling model used to underpin high fat, salt, and sugar restrictions identified 57.6%.

These findings demonstrate that nutritional classifications depend heavily on the definition and scoring method applied. The researchers argue that inconsistent approaches can make it difficult for consumers to understand whether products align with government dietary recommendations.

The analysis was led by Sam Dicken of UCL's Institute of Epidemiology and Health Care, with Professor Chris van Tulleken serving as senior author. Researchers examined 262 processed products and 1,636 ultra-processed products drawn from a representative database of foods available in the UK.

The products were grouped using the Nova food processing classification system before different nutritional assessment methods were applied. Nova categorises foods according to the nature and extent of processing, rather than assessing nutritional composition alone.

That distinction is central to the study. A product's degree of processing does not, by itself, establish its score under a system designed to evaluate energy, saturated fat, sugars, salt, and other nutritional components.

Different classification frameworks can therefore reach different conclusions because they evaluate different characteristics or apply different thresholds.

The Eatwell Guide provides advice about the overall balance of foods and nutrients within a healthy diet. It is not an individual product certification or regulatory labelling scheme.

The UCL researchers developed nutritional assessment approaches based on the Guide's recommendations and compared their results with existing product-level classification methods.

Front-of-pack traffic light labelling operates differently. The voluntary system provides colour-coded information about individual nutrients, including fat, saturated fat, sugars, and salt.

Its presentation allows shoppers to identify particular nutritional characteristics, but the individual indicators are not equivalent to a comprehensive assessment of the product against all dietary recommendations.

The Food Standards Agency Nutrient Profiling Model has another purpose. It produces a numerical classification used within legislation restricting advertising and promotion of specified less healthy food and drink categories.

The system balances components considered less beneficial with certain favourable nutritional characteristics. Its classification can determine whether a qualifying product falls within particular regulatory restrictions.

However, the statutory position depends on both the product's nutrient score and its membership of a category covered by the legislation. A product being classified as unhealthy under a research model does not automatically make it subject to existing HFSS restrictions.

UCL's assessment found substantial differences between the three approaches when applied to the same dataset, including products that achieved acceptable results under the current labelling or regulatory methods but failed the Eatwell-aligned assessment.

Examples included canned pasta in tomato sauce, macaroni cheese prepared from packet mixes, and certain fat-free probiotic yoghurts.

The researchers reported that these products could receive an unfavourable classification under Eatwell-based models without being identified as unhealthy through traffic light labels or the existing HFSS definition.

The opposite type of discrepancy can also occur. Foods such as nuts or whole fruit may receive unfavourable indicators for particular nutrients, despite forming part of dietary recommendations encouraging their consumption.

Such differences arise because the nutritional context of a food is broader than the amount of an individual nutrient it contains.

For products manufactured using several ingredients, the final classification can also be influenced by how the assessment treats their relative contributions to the overall nutritional composition.

The study creates a distinction between reformulating a product to improve an individual nutritional indicator and changing its classification under a broader assessment method.

Reducing sugar or saturated fat can improve particular values, but the result under another model depends on the full calculation, the ingredients involved, and the threshold being applied.

Manufacturers considering nutritional reformulation consequently need accurate product composition data and an understanding of which classification system they are attempting to satisfy.

Technical development introduces further constraints. Ingredients can contribute to flavour, structure, stability, and processing performance as well as the nutritional composition of the finished product.

In bakery manufacturing, changing sugar or fat levels can affect dough handling, moisture behaviour, and texture. The consequences depend on the formulation and the processing conditions selected.

Chilled food and dairy products involve different requirements, including the effect of formulation changes on viscosity, flavour, fermentation, and shelf life.

The UCL research did not test alternative recipes or measure the manufacturing consequences of reformulation. Its findings concern the classifications produced by different systems when applied to existing product data.

The authors also acknowledge that the study does not establish which individual products should be brought under future regulation. Its purpose was to compare assessment methods, rather than construct a definitive list of foods requiring statutory restrictions.

Minimally processed foods were excluded from the analysis. The reported percentages therefore describe the processed and ultra-processed products examined, not every food and drink product available in the UK.

The researchers argue that greater consistency between dietary guidance, labelling, and regulatory assessment could improve the way nutritional information is communicated.

That question differs from the government's ongoing consideration of the updated Nutrient Profiling Model 2018, which concerns revisions to an established scoring system used for advertising and promotion restrictions.

The UCL study compares several approaches to defining unhealthy foods, while the proposed Nutrient Profiling Model change concerns the calculation and thresholds used within one regulatory framework.

No new statutory classification system has been adopted as a result of the research. Any future changes would require policy decisions and the appropriate regulatory processes.

Further analysis may also examine minimally processed foods and whether alternative definitions can provide greater consistency without creating other discrepancies.

The study provides additional evidence about the limitations of existing classification methods, with the next developments dependent on how policymakers respond to the differences identified.