Understanding ultra processed foods and human health: a journey without a plan?
Rapid increases in obesity and diet-related diseases in countries undergoing nutritional transition from traditional food practices to use of more processed foods has prompted development of a food classification system which characterises people’s diets by the amount and type of processed foods they eat, rather than by nutrient content. Most prominent of these, the Nova classification system, assigns foods into four groups (Nova 1-Nova 4) according to increasing degree of processing and has been widely used in recent years to assess potential relationships between so-called ultra processed foods (UPFs), the most heavily processed category (Nova 4), and human health.
Reports of relationships between intakes of UPFs and higher risk of obesity and chronic disease have, not surprisingly, caused concern and have also resulted in a debate about the nature and quality of the scientific evidence. This was the topic of an expert group consensus meeting convened by the British Nutrition Foundation (BNF), the outputs from which have been reported in Nutrition Bulletin. The consensus findings included a number of key questions that need to be addressed: i) the validity of the Nova classification itself and its definitions, ii) the mechanisms underlying effects of UPFs on health and, iii) unintended consequences that might arise by the use in policy-making of a classification that has not been validated for purpose of use and is not yet uniformly accepted.
The elegant editorial by Forde which accompanied the consensus meeting report, welcomed the willingness of the meeting to go beyond repetitive scientific arguments about the pros and cons of individual association studies that has tended to dominate the debate so far. The editorial emphasises the need for mechanistic studies to test the wide variety of mechanisms put forward to explain the health effects of processed foods, but which as yet remain largely unexplored.
In the Academy of Nutrition Sciences’ first peer reviewed Position Paper on the ‘Evidence Base in Nutrition’, we described how advances in methodology for epidemiological association studies had enabled causal criteria, including consistency of findings, large effect sizes and dose responses to be accurately determined. However we noted that ‘biological plausibility’, also considered an important causal criterion, requires testable mechanistic studies which, when available, can provide support to the possibility that observed associations from epidemiological studies reflect causal relationships. For this reason our Position Paper, and a related paper, included recommendations that more emphasis be placed on the use of high quality mechanistic findings in policy-making in the area of diet and health.
As noted in the BNF consensus meeting report (Lockyer et al), there has been much debate about the strength and consistency of findings from association studies of UPFs and health and the lack of data from randomised controlled trials (RCTs). Supporters of the UPF approach argue that observed associations from epidemiological studies already provide compelling evidence for adverse effects of processed foods on health, with urgent public health action needed now. They consider that findings from two recent studies of UPFs and cancer (Chang et al 2023, Kliemann et al 2023), which show 10% increments (or decrements) in UPF consumption to be associated with differences in risk of cancers in the region of 2-27%, and other reported associations of UPFs with type 2 diabetes and cardiovascular diseases of similar magnitudes, could explain between- and within-country differences in rates of diet-related diseases.
However, another viewpoint articulated by Forde considers the present status of nutritional epidemiology research discussed by Lockyer et al. in Nutrition Bulletin to be insufficient in quantity, type and quality to justify policy implementation at the present time, with current research findings also unable to provide answers as to what types of intervention are needed. The Academy agrees with this viewpoint and, despite recognising the many advances in research quality of association studies, considers the dietary assessment methods used and the historical nature of the diet records raise particular questions about the accuracy of estimates of habitual consumption of processed foods.
In our first Position Paper we acknowledged the major developments in the field of nutritional epidemiology that have taken place over the past 50 years. Advances in study design and statistical analysis of association studies have reduced problems of bias and confounding, and the use of systematic reviews and meta-analysis have allowed examination of combined data to test evidence for causality of reported associations. Most of the association studies of UPFs and diet-related disease have deployed sophisticated statistical analyses to correct for a wide range of potentially confounding factors (using over 30 different covariates across a number of models), which address issues of confounding and estimates of effect size.
However the elephant in the room lies in the methods used for dietary recording and, in the present context, lack of information on how the estimates of processed food intakes were achieved using diet records completed 10-20 years earlier.
Food frequency questionnaires (and to a lesser extent multiple 24-hour dietary records) can provide reasonable estimates of habitual food and nutrient intakes and are the pragmatic methods of choice in most association studies. However we question whether they can accurately reflect habitual intakes (and compositions) of a wide range of processed foods, eaten both in and out of the home, and over prolonged periods of time when rapid changes in their consumption and composition will have taken place. These types of concern are heightened by noting that, in the UK Biobank studyreported above, although the aim was to record 5 separate 24-hour records, 35% of the participants only collected one 24-hour record, with only 19% providing five records. In another study, the majority of the cohort provided only two 24-hour records. We question whether a dietary record covering 24- or 48-hrs can truly reflect habitual exposure over 10-20 years?
The issue of under-reporting of diet is rarely dealt with in any detail in many of the reports on UPFs (or indeed many other dietary studies). National diet surveys estimate 25% of participants to show under-reporting of energy intakes in the region of 30% and estimates are higher in subjects with high BMI, in women, and for those consuming foods high in fat and sugar. These data indicate a potential for systematic bias in the diet records of consumers who fall into these categories and this may include high consumers of processed foods. Most association studies now deal with this complex matter by stating they remove records from outliers reporting > 5000 kcals/d or < 500 kcals/d as this indicates significant over- or under-reporting (e.g. Chang et al 2023, Kliemann et al 2023, Srour et al 2019). This approach cannot deal with the great majority of under-reporting consumers who fall inside these ‘safe’ values. For example, an individual who reports consuming 2500 kcals/d but under-reports by 25% will actually take ^3,000 kcals/d, which is well within these arbitrary limits. Furthermore any correction factors can at best only ‘correct’ for energy consumption and are not able to account for mis-reporting of specific foods or food types (e.g. high fat foods).
In his editorial, Forde proposed the intractable challenges of association studies in this area require a focus on randomised controlled trials (RCTs) and studies to address the mechanisms underlying the reported associations. The Academy agrees, and considers the present status of the research is insufficient to demonstrate a causal relationship between UPFs and diet-related diseases.
As a starting point we consider an RCT would be a good foundation study, designed along the lines of the DASH study,which was an intervention that examined the effects of fruit, vegetables and dairy products on blood pressure. As with DASH, many outcomes, including circulating markers of nutrition, metabolic status, processed foods consumption and markers of risk of diseases such as diabetes, cardiovascular disease and cancer, could be measured. Controlled metabolic studies for measurement of energy balance, appetite hormones, body composition and postprandial test studies, could be included, as part or separately, from the main RCT. Numbers of participants would need to be large, requiring a multi-centre investigation with need for public and private funding. One of the challenges the Academy recognised in its first Position paper was the lack of funding for nutrition studies of the type described above. This has been exacerbated since the transfer of nutrition from the Food Standards Agency where mechanistic research and RCTs had been well supported.
The risk of unintended dietary consequences of premature policy concerning UPFs is high, given that the classification system itself has been widely criticised for failure to discriminate foods that contribute beneficial nutrition from those that do not, and the proposed mechanisms involved are multiple and mostly under-investigated. The solutions will not be found in repetitive association studies but require a strategic approach to funding of nutrition research that is currently lacking in the UK.
Professor Christine Williams
Some of our posts are designed to share news on a recent development. The aim of others, such as this, is to encourage debate or reflection on controversial issues, focusing on the strength of the evidence base underpinning these issues. Unlike our Position Papers, such articles do not present an official opinion of the Academy or its Members.