Summary
Background
Food processing has been hypothesised to play a role in cancer development; however, data from large-scale epidemiological studies are scarce. This study investigated the association between dietary intake according to amount of food processing and risk of cancer at 25 anatomical sites using data from the European Prospective Investigation into Cancer and Nutrition (EPIC) study.
Methods
This study used data from the prospective EPIC cohort study, which recruited participants between March 18, 1991, and July 2, 2001, from 23 centres in ten European countries. Participant eligibility within each cohort was based on geographical or administrative boundaries. Participants were excluded if they had a cancer diagnosis before recruitment, had missing information for the NOVA food processing classification, or were within the top and bottom 1% for ratio of energy intake to energy requirement. Validated dietary questionnaires were used to obtain information on food and drink consumption. Participants with cancer were identified using cancer registries or during follow-up from a combination of sources, including cancer and pathology centres, health insurance records, and active follow-up of participants. We performed a substitution analysis to assess the effect of replacing 10% of processed foods and ultra-processed foods with 10% of minimally processed foods on cancer risk at 25 anatomical sites using Cox proportional hazard models.
Findings
521 324 participants were recruited into EPIC, and 450 111 were included in this analysis (318 686 [70·8%] participants were female individuals and 131 425 [29·2%] were male individuals). In a multivariate model adjusted for sex, smoking, education, physical activity, height, and diabetes, a substitution of 10% of processed foods with an equal amount of minimally processed foods was associated with reduced risk of overall cancer (hazard ratio 0·96, 95% CI 0·95–0·97), head and neck cancers (0·80, 0·75–0·85), oesophageal squamous cell carcinoma (0·57, 0·51–0·64), colon cancer (0·88, 0·85–0·92), rectal cancer (0·90, 0·85–0·94), hepatocellular carcinoma (0·77, 0·68–0·87), and postmenopausal breast cancer (0·93, 0·90–0·97). The substitution of 10% of ultra-processed foods with 10% of minimally processed foods was associated with a reduced risk of head and neck cancers (0·80, 0·74–0·88), colon cancer (0·93, 0·89–0·97), and hepatocellular carcinoma (0·73, 0·62–0·86). Most of these associations remained significant when models were additionally adjusted for BMI, alcohol and dietary intake, and quality.
Interpretation
This study suggests that the replacement of processed and ultra-processed foods and drinks with an equal amount of minimally processed foods might reduce the risk of various cancer types.
Funding
Cancer Research UK, l’Institut National du Cancer, and World Cancer Research Fund International.
Introduction
Estimates suggest changes in diet and lifestyle factors could prevent 30–50% of cancer cases.
Over the past decades, diets have shifted towards the consumption of ultra-processed foods, characterised by increased energy density and reduced nutritional quality.
,
,
According to the NOVA food processing classification system, ultra-processed foods are defined as industrial formulations of chemical compounds that are derived from food and drink but not used in culinary preparations, such as cosmetic additives.
Ultra-processed foods can contribute to up to 25–60% of the total daily energy intake in high-income and middle-income countries.
,
,
,
Accumulating evidence suggests intake of ultra-processed food is associated with obesity
,
,
,
,
and other adverse health outcomes, such as cardiovascular disease, cerebrovascular disease, depression, and all-cause mortality.
Evidence before this study
We searched Medline, Web of Science, and Google Scholar with the search terms “food processing”, “ultra-process*”, “NOVA”, and “cancer”, for publications published in English from database inception until June, 2022. We found that epidemiological evidence has suggested a positive association between consumption of ultra-processed food and breast cancer, colorectal cancer, and chronic lymphocytic leukaemia outcomes. However, some conflicting results have also been reported. The evidence regarding associations between dietary intakes of minimally processed food, as assessed by the NOVA classification, and cancer risk is scarce, with only a few studies reporting a positive association between consumption of processed food and prostate cancer risk and an inverse association between consumption of minimally processed food and breast cancer risk.
Added value of this study
To our knowledge we have conducted the largest and most comprehensive study to date investigating the association between dietary intake according to the degree of food processing and risk of cancer at 25 anatomical sites using data from the European Prospective Investigation into Cancer and Nutrition (EPIC) study and assessing whether replacing ultra-processed and processed foods by minimally processed foods might lower cancer risk.
Implications of all the available evidence
This study supports a positive association between the consumption of ultra-processed and processed foods and cancer risk, as found in previous studies (eg, NutriNet-Santé), although some conflicting results were also observed. Most importantly, this study provides robust evidence indicating that the replacement of processed and ultra-processed foods with an equal amount of minimally processed foods should be an important target of cancer prevention strategies in public health, although further research is needed to better understand the best way to achieve this kind of dietary transition.
,
Although epidemiological evidence has suggested a positive association between consumption of ultra-processed food and overall outcomes of cancer, breast cancer, colorectal cancer, and chronic lymphocytic leukaemia, some conflicting results have been reported.
,
,
,
Furthermore, evidence is scarce regarding associations between dietary intakes of foods exposed to lower levels of processing, as assessed by the NOVA classification, and cancer risk, with one study suggesting a positive association between consumption of processed food and prostate cancer risk
and another suggesting an inverse association between consumption of minimally processed food and breast cancer risk.
Therefore, we aimed to investigate the association between dietary intake according to degree of food processing and risk of cancer at 25 anatomical sites using data from the European Prospective Investigation into Cancer and Nutrition (EPIC) study.
Methods
Study design and participants
All study participants provided written informed consent.
,
were included in this study: head and neck cancers, oesophageal adenocarcinoma, oesophageal squamous cell carcinoma, gastric cardia cancer, gastric non-cardia cancer, colon cancer, rectal cancer, hepatocellular carcinoma, gallbladder cancer, pancreatic cancer, lung cancer, renal cell carcinoma, bladder cancer, glioma, thyroid cancer, multiple myeloma, non-Hodgkin lymphoma, leukaemia, melanoma, breast cancer (premenopausal and postmenopausal), cervical cancer, endometrial cancer, ovarian cancer, and prostate cancer. The codes of each cancer site can be found in the appendix (p 2).
Procedures
Gender data were collected via self-report questionnaires and options were male or female. Bodyweight and height were measured in all centres, except for Oxford (UK), France, and Norway where these data were self-reported. However, these self-reported anthropometric measures were shown to be valid for identifying associations in epidemiological studies.
,
Assessed weight and height measurements were used to calculate BMI.
Validated country-specific or centre-specific dietary questionnaires were used to obtain information on food consumption. In most centres, dietary questionnaires were self-administered, except for Ragusa (Italy), Naples (Italy), and Spain, where face-to-face interviews were performed by trained personnel. Extensive semiquantitative dietary questionnaires were used in northern Italy, the Netherlands, Germany, Spain, France, and Ragusa (Italy). Semiquantitative food-frequency questionnaires were used in Denmark, Norway, Naples (Italy), Umeå (Sweden), and the UK, whereas a food-frequency questionnaire was used with a 7-day record on hot meals in Malmö (Sweden). We obtained information on the Mediterranean diet score that was calculated by the EPIC cohort investigators.
Foods were classified as unprocessed or minimally processed (NOVA 1) if they were natural foods or natural foods altered by methods—eg, freezing, pasteurisation, and other processes that do not add additional salt, sugar, oils or fats, or other food substances. Examples of foods included in NOVA 1 are fresh, dry, or frozen fruits or vegetables; grains, flour, and pasta; fresh or frozen meat; milk; coffee; and beans. We classified processed culinary ingredients (ie, NOVA 2) as substances usually obtained directly from foods in NOVA 1 or from nature (eg, oils, fats, sugar, salt). Foods were classified as processed foods (NOVA 3) if they were industrial products made by foods in NOVA 1 and 2 using preservation methods, such as canning and bottling. Examples of foods included in NOVA 3 are breads, cheeses, beer, wine, and smoked fish. Foods in the ultra-processed group (NOVA 4) included those that were made from formulations of ingredients (ie, salt, sugar, fats, or other substances derived from foods), mostly of exclusive industrial use, and are products resulting from a series of industrial processes. Ultra-processed food usually contains many additives to make it palatable or appealing and is packed using synthetic materials. Examples of foods in this group are processed meats (eg, reconstituted meat products or sausage, ham, and other meat products), carbonated soft drinks, packaged breads and buns, sweet or savoury packaged snacks, chocolate, and ready-to-eat meals.
For each NOVA group, the daily total absolute intake in grams and calories as well as their percentage contribution to the total daily intake in grams and calories were calculated. For NOVA groups 3 and 4, this classification process was repeated after removing alcoholic drinks.
Statistical analysis
The main analyses were performed using the middle-bound scenario for the NOVA classification. The daily percentage intake in grams was used because it also considers foods that do not provide energy (eg, artificially sweetened drinks) and non-nutritional factors associated with food processing (eg, neoformed contaminants). Baseline characteristics were examined for the total population and by sex-specific quartiles for the daily percentage intake in grams of each NOVA food group. Descriptive analyses were performed for each NOVA food group considering the absolute daily intake in calories and grams and the percentage intake. Individuals with missing data for the NOVA category were not included.
The associations between the percentage intake of each NOVA group in grams and the incidence of cancers were assessed using Cox proportional hazards regression models. The models were stratified by age at recruitment (in 1-year categories) and centre and adjusted for sex, smoking status and intensity, educational level, physical activity, height, and diabetes (model 1). To investigate the putative effect of food processing independent of the nutritional quality and energy content of foods (eg, due to processing contaminants), we also adjusted for the potential effect of body size, dietary intake and quality, and alcohol intake by adjusting the models further for BMI, Mediterranean diet, alcohol intake, total energy intake, and total fat, sodium, and carbohydrate intakes at recruitment (model 2). Colorectal cancers were further adjusted for fibre and calcium intake in model 2. Renal cell carcinoma was further adjusted for hypertension, and female-specific cancer sites were further adjusted for menopausal status, hormone therapy, oral contraceptive use, age at menarche, and age at first full-term pregnancy in models 1 and 2. In these models, time at entry was age at recruitment and exit time was age at cancer diagnosis, end of follow-up, loss to follow-up, or death, whichever came first. These analyses were repeated using the processed and ultra-processed food groups without alcoholic drinks. To test the proportional hazards assumption, we generated log–log (survival) versus log–time plots.
Since the percentage intake of the NOVA groups corresponds to compositional data, a substitution analysis was performed. To assess the effect of replacing 10% of processed foods and ultra-processed foods with minimally processed foods on cancer risk, we used Cox proportional hazards regression models. For each cancer site, we included the relative intakes corresponding to NOVA groups 1, 2, and 4 in the same model. As a result, NOVA 3 served as a reference, and the relative risk estimate for NOVA 1 represented the substitution of 10% of NOVA 3 by 10% of NOVA 1, while keeping the other NOVA groups constant. We repeated the same analyses using NOVA 4 as the reference. The models were stratified by age at recruitment (in 1-year categories) and centre and adjusted for the same covariates as in the associations analyses. Similarly, the analyses were also repeated using the processed and ultra-processed food groups without alcoholic drinks.
Sensitivity analyses were performed (1 SD increment only) by excluding individuals diagnosed with cancer during their first 2 years of follow-up. The adjustment for total water intake was tested in the models using the daily percentage intake of NOVA food in grams. The association between food processing and cancer risk was also tested using daily percentage calorie intake, as well as lower-bound and upper-bound scenarios for the NOVA classification. Statistical tests used in the analysis were all two-sided, and Bonferroni correction for 26 tests was applied for multiple testing. Statistical analyses were conducted using STATA (version 11.0), and graphs were created with R (version 3.6.3).
Role of the funding source
The funders of the study had no role in the study design, data collection, data analysis, data interpretation, or writing of the report.
Results
Table 1Baseline characteristics for all participants and sex-specific quartiles of percentage daily intake of ultra-processed foods in diet in grams
Data are mean (SD) or n (%). All differences in baseline characteristics between quartiles were significant (all p<0·001). Quartile 1 contains participants with the lowest consumption of ultra-processed foods, and quartile 4 contains those with the highest consumption of ultra-processed foods. NOVA 1=unprocessed or minimally processed foods. NOVA 2=processed culinary ingredients. NOVA 3=processed foods. NOVA 4=ultra-processed foods.
Table 2Percentage and absolute contributions of NOVA groups to the total daily diet by mass and energy, for the total cohort and by country
FigureForest plot for the association between daily percentage intake of NOVA groups in grams and cancer risk by quartiles in model 1 (A) and model 2 (B)
FigureForest plot for the association between daily percentage intake of NOVA groups in grams and cancer risk by quartiles in model 1 (A) and model 2 (B)
Table 3Associations between percentage daily intake of NOVA group foods by mass (g) and cancer risk
Data are hazard ratio (95% CI). NOVA 1=unprocessed and minimally processed foods. NOVA 2=processed culinary ingredients. NOVA 3=processed foods. NOVA 4=ultra-processed foods.
Table 4Substitution models replacing 10% of processed foods (NOVA 3) and ultra-processed foods (NOVA 4) with 10% of minimally processed foods (NOVA 1) and their effect on cancer risk
Data are hazard ratio (95% CI) and p value for the likelihood ratio test (the model fit for each analysis; substitution of NOVA 3 by NOVA 1 or NOVA 4 by NOVA 1). NOVA 1=unprocessed and minimally processed foods. NOVA 2=processed culinary ingredients. NOVA 3=processed foods. NOVA 4=ultra-processed foods.
Discussion
we found consistent results for the inverse association between minimally processed food consumption and risk of overall cancer and postmenopausal breast cancer. Some inconsistencies between our findings and previous findings
,
,
,
were observed for some cancer sites—eg, we found a positive association with consumption of processed food and risk of colorectal cancer and postmenopausal breast cancer, whereas other studies did not.
,
,
,
For breast cancer, a significant positive association had been reported with the consumption of ultra-processed foods in the NutriNet-Santé prospective cohort,
whereas our analysis in this study did not provide evidence for such association. Our study, using data from the large-scale EPIC cohort, is the largest study investigating these associations between food processing and cancer risk and therefore has greater power to detect differences in populations, potentially explaining why we found overall more significant results for different cancer sites than other cohorts.
,
,
,
,
and be associated with obesity,
,
,
,
an established risk factor for at least 13 cancer sites, including the head and neck.
,
Diets rich in processed foods tend to have an increased energy density,
as well as a high contribution of alcoholic drinks, which might have partly explained the association between processed foods and cancer risk in this study. When alcoholic drinks were removed from the NOVA classification, the associations between intake of processed food and rectal cancer, hepatocellular carcinoma, and postmenopausal breast cancer became non-significant, suggesting that drinking alcohol probably drove those associations.
and bisphenol-A (BPA).
,
Other non-nutritional compounds that might be implicated in cancer risk are specific food additives (eg, preservatives) widely used in ultra-processed and processed foods, and cosmetic additives (eg, flavours and emulsifiers) used only in the production of ultra-processed foods.
Sodium nitrate, for example, is used by manufacturers to preserve processed and ultra-processed meat and poultry. Some studies suggested that this compound might increase cancer risk due to formation of nitroso compounds that could yield carcinogenic nitrosamines.
Additionally, some emulsifiers have been postulated to promote inflammation in the gut,
a metabolic alteration also potentially associated with the cause of cancer.
Another concern is the possible effect of artificial sweeteners on cancer risk, which remains controversial.
In this study, intake of ultra-processed foods contributed to 32% of total daily energy intake, but nowadays it could represent 60% of total daily energy intake in European countries.
This discrepancy might explain the fewer significant associations observed between ultra-processed foods and cancer risk than in processed foods and cancer risk. However, the sensitivity analysis that classified food products on the basis of the modern food environment (upper-bound scenario) showed similar results (appendix pp 20–21). This similarity between middle-bound and upper-bound scenarios suggests that the population (generation) being studied, that grew up eating less ultra-processed foods than younger generations, might still not consume higher proportions of ultra-processed foods at present. If correct, the estimates in this study might offer a reasonable representation of long-term exposure to ultra-processed foods. However, the effect of the intake of ultra-processed food on cancer risk at present could be higher than the effect shown in this study. Although differences in dietary questionnaires between the EPIC centres could have affected the NOVA classification, models were stratified by centre to control for this issue and the NOVA classification considered differences in food intake between countries. Additionally, NOVA misclassification might have occurred since many assumptions had to be made while classifying the foods according to NOVA groups due to the absence of food processing information in the dietary questionnaires. However, data collected via 24-h dietary recalls in a subsample of individuals in all countries were used to assist assumption choices and minimise misclassification. Covariates included in the study might have also changed over time, such as physical activity, alcohol intake, and smoking, potentially causing residual confounding by these factors. Under-reporting of foods with a high energy density among people with obesity could lead to an underestimation of dietary consumption of ultra-processed foods. Additionally, analyses were performed by large cancer subgroups and in-depth analyses within each subgroup should be performed to further investigate these associations and potential pathways due to the causal heterogeneity within each subgroup. Finally, the exclusion of participants without substantial dietary data at baseline could have potentially led to selection bias.
This study provides additional evidence of the effect of food processing on cancer risk and suggests that substituting processed and ultra-processed food products with minimally processed food might reduce risk for overall cancer, head and neck cancers, oesophageal squamous cell carcinoma, colon cancer, rectal cancer, hepatocellular carcinoma, and postmenopausal breast cancer. Therefore, recommendations to encourage increased consumption of fresh and minimally processed foods, while reducing the consumption of processed and ultra-processed foods, could be integrated into public health cancer prevention strategies. Future research is needed to replicate these analyses in cohorts with dietary data collected more recently and to explore the mechanistic basis of the observed associations.
FR, RBL, IH, GN, and CC generated the food processing indicators. NK, with assistance from VV, MJG, and IH, did the analyses. NK, VV, MJG, IH, CAM, CM, FR, RBL, EPV, RC, and HF wrote the Article considering the comments and suggestions of the other coauthors. All authors had the opportunity to comment on the analysis and interpretation of the findings and approved the final version for publication. All authors were permitted full access to the data in the study, and NK and IH accessed and verified the data in the study. All authors accept responsibility to submit the manuscript for publication.
Acknowledgments
This work was supported by Cancer Research UK (C33493/A29678), World Cancer Research Fund International (IIG_FULL_2020_033), and the Institut National du Cancer (INCa number 2021–138). The coordination of EPIC is financially supported by the IARC and the Department of Epidemiology and Biostatistics, School of Public Health, Imperial College London, which has additional infrastructure support provided by the UK National Institute for Health and Care Research Imperial Biomedical Research Centre. The national cohorts are supported by the Danish Cancer Society (Denmark); Ligue Contre le Cancer, Institut Gustave Roussy, Mutuelle Générale de l’Education Nationale, Institut National de la Santé et de la Recherche Médicale (INSERM; France); German Cancer Aid, German Cancer Research Center (DKFZ), German Institute of Human Nutrition Potsdam-Rehbruecke (DIfE), Federal Ministry of Education and Research (BMBF; Germany); Associazione Italiana per la Ricerca sul Cancro-AIRC–Italy, Compagnia di SanPaolo and National Research Council (Italy); Dutch Ministry of Public Health, Welfare and Sports (VWS), Netherlands Cancer Registry (NKR), LK Research Funds, Dutch Prevention Funds, Dutch ZON (Zorg Onderzoek Nederland), World Cancer Research Fund, Statistics Netherlands (Netherlands); Health Research Fund (FIS)—Instituto de Salud Carlos III (ISCIII), Regional Governments of Andalucía, Asturias, Basque Country, Murcia and Navarra, and the Catalan Institute of Oncology (ICO; Spain); Swedish Cancer Society, Swedish Research Council and County Councils of Skåne and Västerbotten (Sweden); and Cancer Research UK (14136 to EPIC–Norfolk; C8221/A29017 to EPIC–Oxford) and Medical Research Council (1000143 to EPIC–Norfolk; MR/M012190/1 to EPIC–Oxford; UK). Where authors are identified as personnel of the International Agency for Research on Cancer or WHO, they are responsible for the views expressed in this Article and they do not necessarily represent the decisions, policy, or views of the International Agency for Research on Cancer or WHO.
