A study published in Thorax looks at ultra-processed food (UPF) consumption and the risk of lung cancer.
Prof Kevin McConway, Emeritus Professor of Applied Statistics, The Open University, said:
“I have to confess that my heart rather sank when I got the request from SMC to comment on this study. That’s not because it’s complicated, and not even because it’s a particularly bad study of its type. It’s just that it’s yet another of a class of studies about ultra-processed foods (UPFs) that, in my view, are doing nothing much to advance what is known about associations between the consumption of UPFs and human health. I’m well aware that studies with other kinds of methodology are going on – I just wish that researchers would concentrate more on those other types of research, instead of repeatedly cranking the handle of doing studies like this one.
“The type of study that I’m getting sick of seeing goes like this. The researchers find an existing large observational study, that involves a cohort of people, and for which the data have been made available to other researchers on application. Many of the cohort studies involved took place in the UK or the USA. Of course it’s generally a good thing that data from big studies is made available for other research, but that doesn’t mean that all the research you can do with this secondary data is necessarily very useful.
“Typically, the researchers wouldn’t have been involved in the original study whose data they are using, and often are based in an entirely different country. In this case, the original study was in the USA, and all but one of the researchers is based in China (with one being at Harvard).
“To be useable for a study aiming to examine the association between consumption of UPFs and some health outcome, the original study has to have recorded information on what the participants eat, and then has to follow them up reasonably systematically, and record the health outcome. It also has to record other potentially relevant information about the participants.
“The outcome – in this case a diagnosis of lung cancer, but in other studies it might be death from a particular cause or from all causes – typically might take a long time to develop, so the participants have to be followed up for a long time. (In the case of this new study, the average length of follow-up was over 12 years.)
“Make no mistake – although the cohort used in this new study comes from a randomised trial of cancer screening, the so-called PLCO Cancer Screening Trial, the new study is observational. That’s because the exposure of interest – how many UPFs the participants said they consumed – wasn’t allocated to them at random. The participants themselves chose what to eat and drink.
“The trouble with observational studies like this is that each one of them, on its own, might find an association, that is, a correlation between what people eat in the way of UPFs, and the risk of the health outcome of interest. But it can’t tell you whether that association is one of cause and effect. That’s because there will typically be many other factors that differ between the groups that consume different amounts of UPFs. These are called potential confounders or potential confounding factors – if one or more of them happens to be associated also with people’s risk of the health outcome, then the potential confounders might be the cause of the association with the health outcome, and not the participants’ UPF consumption at all.
“It’s possible to make statistical adjustments to try to allow for potential confounders, but that process isn’t definitive – you can never be sure that you have adjusted for all the potential confounders, so you can never be sure about cause and effect. That’s why this study, like any decent study of this general type, points out that it can’t establish what is causing what.
“In this particular new study, for instance, although the researchers made adjustments for participants’ smoking status (that is, whether they were current smokers, past smokers, or never smoked), they did not adjust for the amount that people smoked, and they point out that this is a limitation of the study and may potentially lead to biases. (Actually it seems strange that they did not make that adjustment, since my reading of the information about the underlying clinical trail (at https://cdas.cancer.gov/plco/) indicates that information on the amount people smoked was collected and available.) And we certainly can’t assume that no other potential confounders were omitted from the adjustments. (See also point 1 in Further Information below.)
“If there were lots of other studies looking at associations between UPF consumption and lung cancer risk, including studies on the actual mechanisms by which the cancer might be caused by the foods, we might eventually get to a conclusion about cause and effect. But the researchers on the new study are proud to point out its novelty, so we’re clearly nowhere near that position yet.
“The new study certainly doesn’t rule out the possibility that eating larger quantities of UPFs cause increases lung cancer risk, but it doesn’t come near to establishing that this cause and effect really exist. That’s why I feel it has told us rather little.
“Studies of this general type very often have two other important limitations, to do with the measurement of people’s consumption of different foods and drinks, and they both apply to this new study (and indeed are mentioned as limitations by the researchers). Because a long follow-up period is needed, and because the definition of ultra-processed foods did not emerge until 2009, the questionnaires used to measure people’s consumption of foods and drinks used in the underlying cohorts were very often not designed to measure UPF consumption. That certainly applies to the PLCO Cancer Screening Trial that provided data for this new study. It recruited participants between 1993 and 2001, and recorded their diets between 1998 and 2001. (See also point 2 in Further Information below.) Also, often, and certainly in this case, the participants’ diets were recorded only once during their participation, so the new study cannot take account of any changes of diet.
“I have done some calculations (using https://realrisk.wintoncentre.uk/) to give what, I hope, is a clearer view of the actual size of the risk associated with consuming greater amounts of UPFs. A relevant point is that, despite the fact that the new study uses data from over 100,000 participants in the underlying PLCO Trial, most of them were not diagnosed with lung cancer during the 12 years of follow-up. Somewhere between 1 and 2 in every 100 of them got lung cancer during follow-up. Thus the estimates of difference in lung cancer risk between the groups that consumed different amounts of UPFs are subject to quite large statistical margins of error.
“In a group of 1,000 people who are just like the quarter of participants that consumed the lowest levels of UPFs, less than one serving per day on average, about 11 would get a lung cancer diagnosis in a 10-year period. Now consider another group of 1,000 people, who are just like the first 1,000 in terms of all the factors used for statistical adjustment in the researchers’ calculations, except that they all consumer UPFs at a rate like those of the highest quarter in the study (so more than 3.7 servings per day, on average). If the differences in lung cancer risk are due to the higher UPF consumption, and I repeat that this study just can’t establish that, then we’d expect about 15 cases of lung cancer in the second, high-UPF group of 1,000. That’s just 4 more than in the low-UPF group, despite the large difference in UPF consumption between the groups. The statistical uncertainty means that that number could plausibly be greater or smaller – between 3 and 6 more lung cancer cases in the high-UPF group than the low-UPF group of 1,000 people. So even if the higher UPF consumption was somehow known to be what causes the increased risk (and, remember, this can’t be known from this type of study), arguably the increase isn’t all that large anyway.
“Studies of this kind also usually can do little or nothing directly to investigate how high UPF consumption might possibly cause an increase in health risk, and that’s the case in this new study. The authors make suggestions on how high UPF consumption could potentially increase the risk of lung cancer, but these suggestions are based on other previous studies at best, and don’t come from the new data analysis.
“The press release repeats the suggestion in the research paper that the compound acrolein is known to be a toxic component of cigarette smoke, and that (of course) it is well known that cigarette smoke increases lung cancer risk. It also mentions that acrolein is found in grilled sausages and caramel sweets, which would generally be counted as UPFs. But it appears* that acrolein is found in many food substances, including some that would not be counted as UPFs. In any case, the data behind the new research did not make any measurements of acrolein.”
Further information
The following points are more technical.
- The researchers calculated what’s called an E-value, to give a measure of how strong the effect of a single potential confounder, not included in their adjustments, would have to be in order to wipe out the observed association between UPF consumption and lung cancer risk. They point out that the E-value implies that a single potential confounder would need to have a stronger effect than smoking status to overturn their observed association. However, you have to remember that they left out smoking amount as a factor to adjust for, and that might have a strong effect in addition to smoking status. Also E-values look only at the effect of one omitted confounder at a time – there could be several of them, whose combined effect could be great enough to overturn the observed association. We simply can’t tell.
- The new research paper reports that the Diet History Questionnaire (DHQ) used in the underlying study to measure diet, was validated in four 24-hour dietary recalls. But the study on the 24-hour recalls, reference 16 in the new research paper, was not part of either the underlying PLCO Cancer Screening Trial nor the new research. It was based on research carried out in 1997, before the PLCO trial began to collect diet information from the DHQ, and its aim was to investigate how closely DHQ results matched those from detailed 24-hour food recalls, which were considered more accurate. That study found that the correlation between food components measured by the DHQ and by the dietary recalls was only about 0.5, meaning that the DHQ results explained only roughly a quarter of the variability in the more accurate dietary recalls. However, I am no expert on dietary measurement and you’d need to check with dietitians to check this concern.
* ‘Origin and fate of acrolein in foods’, Jiang et al. https://www.mdpi.com/2304-8158/11/13/1976 , though my own expertise does not extend to knowing how dangerous (if at all) these amount of acrolein would be to health.
Prof Sam Hare, Consultant Chest Radiologist, Royal Free London NHS Trust, said:
“A quarter of lung cancer cases occur in non-smokers so we do need research exploring whether other factors are associated with lung cancer. We also know immunity is linked to cancer biology so it is a good idea to do research into factors like diet.
“However, further work is needed to establish direct causation between UPFs and lung cancer. Crucially, whilst the study does make some adjustments for smoking status, the amount of smoking is not factored in, which is known to be directly related to lung cancer development. Dietary habits also change considerably over the course of such long term studies. As such, it is difficult to directly conclude that lung cancer is related to the level of UPF consumption alone given it was only declared at the start of the study.
“That said, given the relative dearth of information on non-smoking related risk factors in lung cancer, it is important that the scientific community conducts more studies like this – we need genuine evidence-based advancement in the early diagnosis of lung cancer in non-smokers, but this study isn’t quite able to give us the answers yet.”
Rachel Richardson, Acting Head of Methods Support, The Cochrane Collaboration, said:
“This is a well-conducted observational study, but there are some important factors to bear in mind when considering any wider implications.
“Firstly, this study uses data from a series of trials of screening conducted amongst older adults living in the USA. It is very likely that regulations governing food (for example, permitted additives) are different from those in the UK, which means that American diets are likely to be different from ours in the UK. The fact that this study only included people aged 55-74 also limits the extent to which the findings may be relevant to younger people.
“Secondly, because this study uses data that were originally collected for a different purpose, the information they have is limited in terms of the research question they are trying to answer. A significant limitation is that information on food intake was only collected once which does not reflect the fact that diets change over time. The authors also do not seem to have had information on other factors that are related to lung cancer, for example, exposure to air pollution or second-hand smoke – these are likely to be confounding factors.”
Dr Adam Jacobs, Executive Director and Strategic Consultant, Biostatistics, Ergomed, said:
“Wang et al found a statistically significant association between ultra-processed food (UPF) consumption and lung cancer. This association was strong enough to be clinically relevant, with a 41% increased risk of lung cancer in the highest quartile of UPF consumption compared with the lowest – owever, the important question is whether this association is causal, in other words whether it is the UPF consumption itself that drives the risk of lung cancer or whether people with high UPF consumption are more likely to be at increased risk of lung cancer for other reasons.
“The most obvious confounding factor here is smoking, which is well known to cause a greatly increased risk of lung cancer. If people with high UPF consumption smoked more than people with low UPF consumption, then that difference in smoking could easily lead to the observed results. Although Wang et al attempted to adjust for smoking in their analysis, their adjustment was very crude: they categorised participants into only 2 categories: current or former smokers, and non-smokers (assessed by self report). If people in the high UPF consumption group were more likely to be current smokers rather than former smokers, or smoked more cigarettes per day, or were more likely to have claimed to be non-smokers when in fact they were smokers, then none of these factors would have been captured in the analysis. Wang et al correctly identify this limitation themselves in the paper.
“Without more detailed adjustment of their statistical model for smoking intensity, it seems unsafe to conclude that the observed association between UPF consumption and lung cancer risk was a causal effect of the UPF consumption, and confounding by smoking also seems a plausible explanation of the results.”
Prof Tom Sanders, Professor emeritus of Nutrition and Dietetics, King’s College London, said:
“This an analysis of data collected in the United States comparing diet and lifestyle and subsequent development of lung cancer. The authors claim to find about a 0.4 fold increase in risk of lung cancer associated with the highest ultraprocessed after adjusting for other risk factors. This could be entirely due to residual confounding due to poorly recorded exposure to tobacco as well as occupational exposure to inhaled carcinogens.
“Other studies show that smoking is an enormous cause of lung cancer, increasing risk 12 fold. The statistical analysis only includes current/previous smokers without detail in terms of number of pack years of smoking. It also does not ascertain properly current smokers because individuals are known to lie about smoking habit. A test such as plasma or urinary cotinine is needed to check for smoking status. Unhealthy diets often go hand in hand with smoking habit and low socioeconomic status. But there appears to be no plausible mechanism to explain why ultraprocessed food should affect risk of lung cancer.”
‘Association between ultra-processed food consumption and lung cancer risk: a population-based cohort study’ by Kanran Wang et al. was published in Thorax at 23:30 UK time on Tuesday 29 July 2025.
DOI: 10.1136/thorax-2024-222100
Declared interests
Prof Kevin McConway: “I don’t have any relevant interests to declare.”
Prof Sam Hare: “No conflicts of interest directly relevant to this work. I am a past National Specialty Adviser to NHS England for Imaging. I am CEO and co-founder of the HLH Imaging Group Limited.”
Rachel Richardson: “I have no interests to declare.”
Dr Adam Jacobs: “No competing interests to declare.
Ergomed is a contract research organization: https://ergomedcro.com/.”
Prof Tom Sanders: “I have received grant funding for research on vegans in the past. I have been retired for 10 years but during my career at King’s College London, I formerly acted as consultant for companies that made artificial sweeteners and sugar substitutes.
“I am a member of the Programme Advisory Committee of the Malaysia Palm Oil Board which involves the review of research projects proposed by the Malaysia government.
“I also used to be a member of the Scientific Advisory Committee of the Global Dairy Platform up until 2015.
“I did do some consultancy work on GRAS affirmation of high oleic palm oil for Archer Daniel Midland more than ten years ago.
“My research group received oils and fats free of charge from Unilever and Archer Daniel Midland for our Food Standards Agency Research.
“I was a member of the FAO/WHO Joint Expert Committee that recommended that trans fatty acids be removed from the human food chain.
“Member of the Science Committee British Nutrition Foundation. Honorary Nutritional Director HEART UK.
“Before my retirement from King’s College London in 2014, I acted as a consultant to many companies and organisations involved in the manufacture of what are now designated ultraprocessed foods.
“I used to be a consultant to the Breakfast Cereals Advisory Board of the Food and Drink Federation.
“I used to be a consultant for aspartame more than a decade ago.
“When I was doing research at King’ College London, the following applied: Tom does not hold any grants or have any consultancies with companies involved in the production or marketing of sugar-sweetened drinks. In reference to previous funding to Tom’s institution: £4.5 million was donated to King’s College London by Tate & Lyle in 2006; this funding finished in 2011. This money was given to the College and was in recognition of the discovery of the artificial sweetener sucralose by Prof Hough at the Queen Elizabeth College (QEC), which merged with King’s College London. The Tate & Lyle grant paid for the Clinical Research Centre at St Thomas’ that is run by the Guy’s & St Thomas’ Trust, it was not used to fund research on sugar. Tate & Lyle sold their sugar interests to American Sugar so the brand Tate & Lyle still exists but it is no longer linked to the company Tate & Lyle PLC, which gave the money to King’s College London in 2006.”