PRINCIPAL FINDINGS
In the present case-control study, consumption of ultra-processed foods and drinks was associated with increased odds of colorectal cancer. Overall, no association was observed between consumption of ultra-processed foods and drinks and breast cancer after adjusting for confounding factors; however, some associations emerged in some sub-groups of women, i.e. former and current smokers. No association was observed with prostate cancer.
COMPARISON WITH EXISTING LITERATURE
Since the development of the NOVA classification of foods and drinks according to the degree of processing [3], numerous epidemiological studies have evaluated the association between ultra-processed food and drink consumption and adverse health outcomes [5], such as cardiovascular disease [16] and mortality [4, 17–19]. In 2018, based on the French NutriNet-Santé prospective cohort of approximately 105,000 participants of median age 42.8 years, the first and, as far as we know, the only study on ultra-processed food and drink consumption and cancer risk was published. In that study, a 10% increase in the consumption of ultra-processed foods and drinks, was significantly associated with an increased risk of total cancer (Number of cases 2228, hazard ratio (HR) 1.12, 95% CI 1.06 to 1.18) and breast cancer (Number of cases 739, HR 1.11, 95% CI 1.02 to 1.22). The HR for colorectal cancer (Number of cases 153) was 1.13 (95% CI 0.92 to 1.38), not reaching the standard threshold for statistical significance, maybe due to the low number of incident cases. The HR for prostate cancer was closer to 1.
In the MCC study, ultra-processed food and drink consumption was significantly associated with colorectal and the OR (per 10% increase in ultra-processed food and drink consumption OR 1.11, 95% CI 1.04 to 1.18) was of similar magnitude to the HR observed in the NutriNet-Santé cohort, but statistically significant, maybe due to the larger number of cases (1842). The association was observed for both colon and rectal cancer. Further adjustment for nutritional characteristics of diets rich in ultra-processed foods and drinks, i.e. daily energy density, total saturated fat or simple carbohydrate intake, did not attenuate the association, indicating that the association may be driven by factors beyond the diet quality of such foods and drinks, such as food additives [7]. On the other hand, when fibre intake, or fruit and vegetable consumption were included in the model, the association was attenuated. This could indicate that the association between ultra-processed foods and drinks and colorectal cancer may be partly explained by the low intake of fibre, fruit and vegetables in high consumers of ultra-processed foods; nevertheless, when analyses were stratified by low versus high consumption of fruit and vegetables, the association between ultra-processed foods and drinks and colorectal cancer was only significant in the group of high consumers of fruit and vegetables. This possible interaction between fruit and vegetable consumption and ultra-processed foods and drinks on colorectal cancer, deserves further investigation, but may indicate that, in low fruit & vegetable consumers, other factors such as low fibre or folate intake, might be more relevant for the development of colorectal cancer than other characteristics of the diet related to food processing[30, 31].
For breast cancer, results of our study differ from those in the French cohort as we did not find evidence for an association between ultra-processed food and drink consumption and breast cancer, in the overall sample. Reasons for such discrepancies in results are difficult to elucidate and could be explained by differences in study design or study population. For instance, participants in the NutriNet-Santé cohort were younger on average than participants in the MCC-Spain study, and in our study there was some evidence that the association was stronger in younger population sub-groups (i.e. premenopausal women). Of note, in minimally adjusted models, the association between ultra-processed food and drink consumption and breast cancer was statistically significant; further adjustment by total energy intake and/or ethanol intake resulted in an attenuation of the association and loss of statistical significance. This could indicate that the effect of such foods on breast cancer risk, if any, would be mediated through alterations in the energy balance [32], or its contribution to ethanol intake, well known risk factors for breast cancer [33]. Lastly, in the subgroup of former and current smokers, the association between ultra-processed food and drink consumption and breast cancer was statistically significant. It is known that smoking and some dietary factors might have some synergetic effects on the development of cancer [34], as it might be the case with the consumption of ultra-processed foods and drinks and smoking on breast cancer; however, this finding needs confirmation.
In studies, ultra-processed food and drink consumption was not associated with prostate cancer. This is not surprising given that the evidence linking dietary factors to prostate cancer risk is indicative of no association [35].
STRENGTHS AND WEAKNESSES OF THE STUDY
Advantages of the study include the substantial sample size of histologically-confirmed incident cancer cases. Foods and drinks in the validated FFQ were carefully classified using the NOVA system, according to the degree of processing, by a panel of nutritionists. We performed several sensitivity analyses to test the robustness of our results. Main limitations are inherent to the case-control design of the study, i.e. recall bias and selection bias. Regarding recall bias, the dietary data collected at recruitment referred to the preceding year, and was collected early after cancer diagnosis. Thus, if recall bias exists, it would probably be non-differential, thus implying underestimation of the effects studied. Regarding selection bias, the MCC-study was designed with the goal of minimizing selection biases by recruiting population-based controls, and all cases with a first diagnosis of cancer in the selected health areas, ensuring few incident cases were missed in the study. Another limitation is related to the use of the NOVA classification to assign FFQ food items to different NOVA groups: for some food items, the FFQ does not provide enough information of food processing to determine if the food items belongs to one food group or another, which may have resulted in some degree of misclassification; nevertheless, we discussed each food item between a team of nutritionist and used information on food composition and food system in Spain to classify all foods items. Also, the NOVA methodology/classification has limitations that have been criticized by some [36], but it is the most used method for classifying ultra-processed foods and drinks today [37]. Dietary data might be also subject to measurement error; nevertheless, we used a previously validated FFQ for Spanish population. When interpreting results of analyses carried out in certain sub-group, we need to bear in mind the potential lack of statistical power due to small sample sizes. These associations should be interpreted in the context of multiple comparisons and possibility of chance findings. Finally, although we adjusted for a range of potential confounders, residual confounding cannot be totally ruled out.