A study published in the American Journal of Preventive Medicine looks at ultra-processed foods (UPFs) and premature mortality.
Prof Nita Forouhi, Professor of Population Health and Nutrition, MRC Epidemiology Unit, University of Cambridge, said:
“There are limitations to this paper, including the points the authors themselves raised. Nonetheless, evidence on the ‘health harms of UPF’ are accumulating and this paper does add to that body of evidence, and UPFs are unlikely to be healthful.
“We already know that correlation does not necessarily mean causation. But well conducted observational studies with long term prospective cohort data are often the best we are going to get realistically; we will not get randomised controlled trials (RCTs) of behaviours awaiting death or chronic disease events, and RCTs have their own biases and limitations, particularly for behavioural factors (different to taking medication vs placebo studies). So we should not ignore such findings, especially as the current research has reported consistently similar associations in several countries which increases the degree of confidence.
“In addition to the 8 countries they included for their population attributable fraction (PAF) estimates (Australia, Brazil, Canada, Chile, Colombia, Mexico, UK, USA), it would have been useful if they had also included the countries that provided the results on associations of UPFs with mortality but were not included (e.g. France, Italy, Spain).”
Prof Kevin McConway, Emeritus Professor of Applied Statistics, Open University, said:
“I’d be pretty cautious about the details and specific numerical estimates in this paper, for reasons I’ll explain. Also, some of the terminology in the paper and the press release appears, in my opinion, much more definite about what’s causing what than the evidence in the paper merits. That’s partly because some of the technical wording, even though it’s standard in this kind of research, doesn’t mean quite the same as it means in ordinary English.
“The problems of interpretation arise because the studies involved are observational, but they go further than that. The researchers have to make mathematical assumptions about exactly how UPF consumption is correlated with mortality risk, and even though they base these assumptions on data, there is at least one issue (described later). And in calculating what’s known as the attributable epidemiological burden, or population attributable fraction, of UPF consumptions, the researchers may appear to be making a simple comparison, but in fact it’s a lot more complicated than you might think.
“The data that the paper draws on for its conclusions, about consumption of UPFs and mortality, is all observational. The researchers are not reporting any new data here – they are taking data from previous studies, and population estimates for the countries concerned, and putting it all together. Nothing at all wrong with that – in fact in general it’s a good idea to review studies of the same things from different times and places, to see what overall picture emerges.
“The seven studies that the authors of this paper used, to find an overall pooled estimate of the association between UPF consumption and all-cause mortality, are all themselves observational. Again there’s nothing wrong with that – it’s pretty difficult, indeed impossible in most cases, to do a study linking diet to long-term health outcomes that is not observational. Such a study would have to allocate different individuals to different diets, and somehow ensure that they stuck to these diets for many years. So instead, researchers record what people eat, and then follow them up for a long time and record if and when they die.
“This all means that it’s impossible, for any one study like that, to be sure whether differences in mortality between people who consume different UPF amounts are actually caused by differences in their UPF consumption. There are bound to be many other differences between groups who consume different UPF amounts, in terms of other details of their diet, their lifestyle, their economic position, their sex and age, and so on. These differences might be, in part or in whole, the reason for the differences in the risk of early death. In other words, each individual study can find a correlation, an association, but can’t say for sure whether the association between UPF consumption and mortality is one of cause and effect. It might be, or it might not.
“The researchers in each of the studies reviewed in this new paper obviously are aware of this, and they all made statistical adjustments to allow for differences in other factors (though in different ways in different studies). But that doesn’t make the problem disappear – you still can’t be sure from any study of this kind exactly what’s causing what.
“The fact that the new paper puts together data from seven different observational studies does again help somewhat with the issue of what’s causing what, but it can’t deal with it entirely. There have been many criticisms of interpretation of observational studies involving UPFs and health outcomes, some of them on the basis that UPFs are defined in rather different ways by different writers, or on the grounds that the mechanisms by which UPFs might actually cause ill health haven’t been established clearly enough.
“I’m certainly not saying that there is no association between UPF consumption and ill health – just that it’s still far from clear whether consumption of just any UPF at all is bad for health, or of what aspect of UPFs might be involved.
“Then there are particular aspects of this new study that make the interpretation more complicated than it would be for other observational studies of UPFs and health.
“The authors begin by estimating the nature of the association between the consumption of UPFs and the risk of premature death. That is, they aren’t just trying to see whether high levels of UPF consumption are correlated with higher mortality. They want to know something more precise – exactly how much does the risk of dying increase, for every additional 10 per cent of a person’s calorie intake that comes from UPFs. (Again, no assumption here that the increase in risk is all caused by UPFs.) That sounds fine, but it involves assuming a particular mathematical form for the association (in the light of the data).
“After that, the authors use the estimate of that association between UPF consumption and risk of early death to calculate estimates of the population fraction of premature deaths (ages 30-69) attributable to UPF consumption, for 8 different countries including the UK. They use that to calculate estimates of the number of additional deaths in each of the 8 countries attributable to UPF consumption, and some of those numbers look pretty large.
“This is done by taking data on the number of people in different groups (defined by age and sex) in each country. This is then used to calculate how many would be expected to die at current levels of UPF consumption (using data from the estimate of the association between UPF consumption and premature death in all the studies that were put together in the first part of the work, so not just for the UK for example). Finally this is compared with the number that would be expected to die in a theoretical population where nobody consumes (or ever consumed) UPFs. No such population exists, not in a whole country, so this calculation has to be based on a statistical model. Then the deaths attributable to UPF consumption is the difference between these two expected numbers of deaths.
“What this sounds like, for the UK in 2018-19 for example, is that there would have been almost 18,000 fewer deaths of people aged between 30 and 69, if nobody in the country had consumed any UPFs (ever). However, that’s very far from the whole story, for a lot of reasons.
“First, it doesn’t mean that, because the studies involved are observational, and as the authors of the new paper rightly point out, there could be factors that could not be adjusted for in the original studies, that are involved in causes of early death. That’s why it’s called a population attributable fraction, rather than something even more definite, like population fraction caused by UPFs. Technically, it can’t mean that we know we could save those lives just by changing UPF consumption.
“But it’s deeper than that. There isn’t a whole population in the UK or in the other seven countries in the study, where nobody ever consumed any UPFs. So the comparison is being made between an estimate for current UPF consumption levels and an estimate for a theoretical population that can’t exist. Even if somehow all UPFs were banned today, it would take many decades before there was a population where nobody had ever consumed UPFs.
“And even if somehow we did get to that position, well, people have to eat something, and if they aren’t getting their calories from UPFs, they would need to get them from something else. They might well not get them all in the same way that people who consume very few UPFs do today. We just can’t tell.
“So it’s not the case that we could save 18,000 premature deaths annually in the UK by taking action to reduce UPF consumption. This doesn’t mean that taking such actions wouldn’t reduce early deaths – just that we can’t tell how much the reduction might be, or when it would occur, or how much longer the individuals concerned might have lived – not from the calculations in this paper.
“I have some other concerns.
“Several of the authors of the new paper collaborated on a previous paper, published in 2023 (reference 17 in the new paper, which is the reference given for the model used in the new paper for estimates of attributable deaths). The 2023 paper uses similar methodology to make an estimate of the premature deaths attributable to UPFs in Brazil in 2019. This uses similar data on the association between UPF consumption and premature mortality, from a systematic review and meta-analysis, to what’s used in the new paper, except that there are three additional studies reviewed in the new paper. The estimate is only for Brazil, and is 57,000 deaths in a year. The estimate for Brazil in the new paper is just over 25,000 deaths in a year.
“The big difference between the 2023 and the 2025 estimates for Brazil seems to be very largely because of a different assumption made in the two papers about the mathematical form of the association between UPF consumption and death risk. (In the jargon, they use a log-linear model in the 2023 paper but a linear model in the 2025 paper.) The new estimate is based on more data from more countries – but the big difference does emphasise the importance of mathematical modelling assumptions. Data can throw light on what assumptions are appropriate, but don’t tie things down very firmly at all in a situation like this.
“Finally, the systematic review and meta-analysis in the new paper is missing some of the technical details that one normally sees in this kind of work. The paper is very unclear on how the researchers chose the studies they included in their review, which after all drives all the estimates of attributable deaths. The authors write that studies were selected ‘on the basis of recently published systematic reviews’. That’s not normally the way it’s done, and in any case three of the included studies were not mentioned in the systematic reviews that are referred to in the new paper. I don’t know where the researchers got them. They may well be perfectly respectable studies – I haven’t had time to look at them – but really the authors of the new paper should have been much clearer about what they were doing, if we are to be confident about their conclusions. Also it’s usual in a systematic review to give some assessment of the quality of the research studies that were included, and that just isn’t done here. None of this increases trust in how the work was done.”
Dr Nerys Astbury, Associate Professor – Diet & Obesity, Nuffield Department of Primary Health Care Sciences, University of Oxford, said:
“Here Nilson and colleagues report findings from a study reporting associations between consumption of Ultra Processed Foods (UPF), defined by the NOVA classification system, and premature mortality.
“This study combines evidence on dietary intake of UPF from Columbia, Brazil, Australia, Canda, United Kingdom and USA and reports that for each 10% increase in proportion of UPF in the diet there was a 3% increase in all-cause mortality. The authors then used a mathematical formula to estimate the population attributable fraction, which is an estimate of the number of deaths which could be prevented if the exposure (consumption of UPF) was eliminated. It is important to note this does not mean that these deaths were caused by UPF consumption. The methods of this study simply cannot determine this.
“It’s been established for some time including in the Global Burden of Disease Consortium that consuming diets higher in energy, fat and sugar can have detrimental effects on health, including premature mortality. This study adds to the body of evidence on the association between UPF and ill health and disease. However, many UPF tend to be high in these nutrients, and studies to date have been unable to determine with certainty whether the effects of UPF are independent of the already established effects of diets high in foods which are energy dense and contain large amounts of fat and sugar.
“The authors of the study conclude that advice to reduce UPF consumption should be included in national dietary guideline recommendations and in public policies. However, rushing to add recommendations on UPF to these recommendations is not warranted based on this study in my opinion. Many national dietary guidelines and recommendations already advise the reduction of consumption of energy dense high-fat high-sugar foods, which typically fall into the UPF group. Adding additional recommendations based on UPF could cause consumer confusion – some foods may be considered unhealthy by nutrient standards, but not so by NOVA classification (and vice versa).
“This study and other similar studies that have explored the association between UPF and diet related disease, have used the NOVA classification system invented by Dr Carlos Monteiro (an author on this paper). In my view the NOVA system which defines foods according to different levels of food processing has many limitations, including arbitrary definitions and overly broad food categories, the over-emphasis of food ingredients opposed to the processing per se and the difficult practical application of the system in accurately classifying foods. This is especially notable when attempting to classify foods from dietary data collected in large cohort studies, as in this study.
“More research is needed to ascertain a causal link between UPF and disease and to establish the mechanisms involved.”
Dr Stephen Burgess, statistician in the MRC Biostatistics Unit, University of Cambridge, said:
“This study assesses observational associations rather than interventions, and so it is not able to make reliable causal claims. That is to say, it shows that individuals who consume higher levels of ultraprocessed foods have greater risk of premature mortality, rather than showing that increasing your consumption of ultraprocessed foods would increase your mortality risk. However, the similarity of findings across populations is notable, as consistent associations were seen in a variety of contexts, including those where high consumption of ultraprocessed foods is a sign of relative wealth and those where it is a sign of relative deprivation. This type of research cannot prove that consumption of ultraprocessed foods is harmful, but it does provide evidence linking consumption with poorer health outcomes. It is possible that the true causal risk factor is not ultraprocessed foods, but a related risk factor such as better physical fitness – and ultraprocessed foods is simply an innocent bystander. But, when we see these associations replicated across many countries and cultures, it raises suspicion that ultraprocessed foods may be more than a bystander.”
‘Premature Mortality Attributable to Ultraprocessed Food Consumption in 8 Countries’ by Eduardo A.F. Nilson et al. was published in the American Journal of Preventive Medicine at 05:05 UK time on Monday 28 April 2025.
DOI: 10.1016/j.amepre.2025.02.018
Prof Nita Forouhi: “No conflicts of interest to declare.”
Prof Kevin McConway: “Previously a Trustee of the SMC and a member of its Advisory Committee.”