Intake of Ultra-processed Foods Is Associated with an Increased Risk of Crohn’s Disease: A Cross-sectional and Prospective Analysis of 187 154 Participants in the UK Biobank

Abstract Background and Aims Ultra-processed food [UPF] consumption has been linked to globally increasing incidence and prevalence of chronic diseases, including inflammatory bowel diseases [IBD]. We aimed to investigate the association between UPF consumption and IBD incidence, prevalence, and IBD-relevant outcomes. Methods We performed a cross-sectional and prospective cohort study in 187 854 individuals included in the national UK Biobank, using 24-h dietary recall questionnaires. Multivariable logistic regression and Cox proportional hazard regression were used to examine the association between UPFs and the prevalence and incidence risk of IBD, respectively. Results A total of 185 849 participants with a mean age of 56.2 were included, with a mean follow-up of 9.84 years. During follow-up, 841 developed IBD (251 Crohn’s disease [CD], and 590 ulcerative colitis [UC]). UPF intake in IBD patients was significantly higher: CD: odds ratio [OR] 1.94 (95% confidence interval [CI]: 1.52, 2.49, p <0.001); UC: OR 1.39 [95% CI: 1.17, 1.65, p <0.001]. Compared with low consumption, higher UPF consumption was significantly associated with incident CD: hazard ration [HR] 2.00 [95% CI: 1.32, 3.03, p = 0.001], but not UC. We also found a significant association between UPF intake and need of IBD-related surgery: HR 4.06 [95% CI: 1.52, 10.86, p = 0.005]. Conclusion Higher intake of UPFs was associated with higher incidence of CD, but not UC. In individuals with a pre-existing diagnosis of IBD, consumption of UPFs was significantly higher compared with controls, and was associated with an increased need for IBD-related surgery. Further studies are needed to address the impact of UPF intake on disease pathogenesis and outcomes.


Introduction
Ultra-processed foods [UPFs] make up more than half of the total dietary energy consumed in many high-income countries. 1,2 Although most foods are processed to some extent, UPFs are formulations of ingredients that result from a series of industrial processes, as defined by the NOVA classification system. 2,3 UPFs have been recognised as energy-dense products, high in sugar, unhealthy fats, and salt, and low in dietary fibre, protein, vitamins, and minerals. Cross-sectional and longitudinal studies have shown that increases in the dietary proportion of UPFs result in deterioration of the nutritional quality of the overall diet and increased obesity, hypertension, coronary and cerebrovascular diseases, dyslipidaemia, metabolic syndrome, cancer, and gastrointestinal disorders. 2 In addition, experimental studies indicate that UPFs can induce high glycaemic responses, have a low satiety potential, and create a pro-inflammatory gut environment. 4 Inflammatory bowel diseases [IBD] are chronic inflammatory conditions of the gastrointestinal tract and comprised two main entities, namely Crohn's disease [CD] and ulcerative colitis [UC]. Although historically these have been considered to be Western diseases, incidence and prevalence are increasing globally, particlarly in industrialised and industrialising regions of the world such as Asia, the Middle East and Latin America. 5,6 In order to explain these increasing incidence rates, the role of diet has been closely examined. High dietary intake of total fats, polyunsaturated fatty acids [PUFAs], omega-6 fatty acids, meat, and sugar-sweetened beverages have been associated with an increased risk of CD and UC in observational studies. 7,8 Other than macronutrients, the non-nutritional or 'organoleptic characteristic' components in our diet such as emulsifiers and colourants, have recently been implicated as playing a role in driving inflammation and metabolic derangement in a number of animal and in vitro studies. [9][10][11] A number of recent studies have assessed the association between UPFs and IBD, although to date the findings have been inconsistent. [12][13][14] The French NutriNet-Santé cohort did not find any significant association between UPF intake and IBD incidence. 13 Most recently, however, two larger studies have been completed. The global PURE cohort reported a positive association between UPF intake and risk for IBD, whereas in the American Nurses' Health Study cohort, the authors reported only on an association of UPF intake with CD. 12,14 These findings require to be further validated. In this study we aimed to investigate the association between UPF consumption and IBD incidence, prevalence, and IBDrelevant outcomes, in the UK Biobank.

Study design and participants
The current study was conducted in the UK Biobank, which is a large cohort study incorporating over 500 000 participants, aged 40-69 years, from 2006 to 2010 in the UK. Further details of the study have been described elsewhere. 15 In this study, 191 910 participants had at least one valid 24-h dietary recall questionnaire with credible energy records [>0 and <18 MJ for females, >0 and <20 MJ for males] and were included in the analysis [ Figure 1]. 16 Three separate sub-studies were constructed: a cross-sectional study with IBD patients at baseline [according to hospital diagnosis or general practice reports, thereby including all prevalent cases] and participants without IBD, a prospective cohort with participants without IBD at baseline to investigate IBD incidence, and another prospective cohort with IBD patients only, to investigate the influence of UPF intake on relevant disease outcomes such as colorectal neoplasia and need for IBD-related surgery. Participants were excluded when no dietary information was available, when the type of IBD diagnosis was left unspecified, or genetic information was unavailable.

Exposure and outcome measurements
UPFs were defined according to the NOVA classification. 1,2 The Oxford WebQ questionnaire used by UK Biobank contained 206 food items and 32 alcohol and beverage items to assess dietary consumption over the past 24 h. The 24-h WebQ questionnaire was previously validated with good agreement with the food frequency questionnaire of the UK Biobank and the mean intake of multiple measurements further reduces bias. 17,18 Participants were asked to select how many portions they consumed for each item, with instructions specifying what one portion size represented, such as one sausage, one rasher of bacon, one slice of ham, or one 'serving' for some specific foods. When multiple rounds of dietary recalls were available for the same participant, the mean value was taken into account. The food intake weight in grams for each item was calculated by multiplying amounts of portion size by standard portion sizes in grams; then daily intakes of energy and nutrients were estimated by multiplying the food weight consumed by its nutrient composition. Portion size and nutrient and energy compositions for each food item used for UPF estimation were calculated according to the UK McCance and Widdowson's The Composition of Foods, 6th edition [2002] and its supplements as defined by the NOVA classification. 19,20 Food items included for the estimation of UPF intake are presented in Supplementary Table 1. Intake of each single UPF was calculated as the mean intake of each valid 24-h dietary recall questionnaire, and UPF consumption was calculated as the sum of all these dietary elements. Consumption was further divided into number of UPF servings, energy intake from UPFs, and proportion of energy percentage from UPFs.
Diagnostic information was obtained from both primary care and hospital inpatient records containing data on admissions, diagnoses, and operation procedures. The primary outcomes include the prevalence and incidence of CD

Covariate assessment
Information on covariates, including age, sex, ethnicity, education attainment, and Townsend Deprivation Index [TDI], was collected in the baseline questionnaire. Polygenic risk scores [PRS] were constructed to proxy the genetic propensity to CD and UC separately for each participant, by summing the number of risk-increasing alleles for genetic variants associated with CD or UC and weighted by their effect sizes, respectively, as reported by previous genome-wide association study of IBD. 21

Ethical statement
The

UPF consumption is associated with an increased risk of incidence of CD, but not UC, in individuals without a pre-existing IBD diagnosis
In the first prospective cohort study with a mean follow-up of 9   BMI, PRS and total energy [if the exposure was energy or energy proportion, total energy intake will not be adjusted. UPF, ultra-processed food; CD, Crohn's disease; UC, ulcerative colitis.SD, standard deviation; OR, odds ratio; CI, confidence interval; Q, quintile; TDI, Townsend Deprivation Index; BMI, body mass index; PRS, polygenic risk score.  Table  3]. When analysing UPF as energy intake and energy percentage intake, respectively, the incidence per SD of UPF intake (HR = 1.09 [95% CI: 1.02, 1.16], p = 0.009), with trend [p = 0.017], and incidence per SD of intake (HR 1.07 [95% CI: 1.00, 1.15], p = 0.048) remained significant after considering the fully adjusted model.
When only the risk of incident CD was considered, intake of UPFs measured through servings, energy intake, or energy intake percentage were all significantly associated at the highest quintile of intake with an HR as high as 2.00 [95% CI 1.32, 3.03, p = 0.001] for UPF intake as a proportion of energy percentage in the fully adjusted model. In this model, the results remained significant when considering intake per SD [p = 0.001] and for the trend [p = 0.001] [ Table 4].
For UC, however, we did not find any significant associations between UPF consumption and risk for developing UC [ Table 4].

Need for surgery in IBD
Regarding the association between UPF intake and risk of IBD-related operations in the fully adjusted model [which was further expanded with the additional high-risk clinical features of age at diagnosis, disease location and duration, medication use, stricturing and penetrating behaviour for CD, and baseline fever and weight loss for UC], 29 Table 6]. However, it needs to be noted that cases were few and the results on risk for colectomy in both CD and UC were less consistent [Supplementary Tables 5 and Table 7].

colorectal neoplasia in IBD
When assessing possible associations between UPF intake and risk of benign colorectal neoplasm in CD, we found a signal for the number of UPF servings (HR for the highest quintile of 3.21 [95% CI 1.15, 8.98, p = 0.026], and p = 0.010 for the trend in the fully adjusted model) [ Table 8]. Although a dose-response relationship was clear and results showed significance, only 65 cases were observed and curiously, results were only significant for the fully adjusted and not for the minimally adjusted model. No significant associations were found for UC or IBD and risk of benign colorectal neoplasm [Supplementary Table 6  and Table 8]. Similarly, no associations were found between UPF intake and risk of colorectal cancer in IBD patients, CD patients, or UC patients [Supplementary Tables 7 and Table  9].

Subgroup and sensitivity analyses
In a subgroup analysis on the association between UPF intake and CD risk according to disease location, we found a signal for a higher risk for ileocolonic or undefined disease, but not for ileal or colonic disease [Supplementary Table 8].
When stratifying by sex, females seemed to have a significantly higher risk for CD incidence when consuming higher amounts of UPFs, whereas this association was not significant for males. No significant effect modification by sex was observed in these associations [Supplementary .
In a sensitivity analysis, we obtained similar results for the risk of CD and UC incidence when excluding liquid

Discussion
In this large cross-sectional and prospective cohort study with 187 854 participants, we provide evidence that UPF intake is higher in individuals with a pre-existing diagnosis of IBD than in other individuals, followed in the UK biobank. Furthermore, we report a significant association between UPF consumption and an increased risk of incident CD, but not UC, in individuals without a pre-existing IBD diagnosis. We also report that increased intake of UPFs might contribute to an increased need for surgery and an increased risk of benign colorectal neoplasia in patients with an IBD diagnosis. Most importantly, we report a robust and significant association between higher UPF intake and an increased incidence of CD [HR of 2.00], but not UC, although more interventional studies are needed to explore any causal effect. A different methodology was used in our strategy to capture dietary intakes (namely 24-h dietary recall as opposed to food frequency questionnaires [FFQs]), but consistent with those from the Nurses' Health study from Lo et al. and the PURE cohort from Narula et al. in the association between UPF intake and CD risk [Supplementary Table 22]. 12,14 However, we were unable to replicate the signal for UC incidence reported in the unadjusted analysis by Narula et al. In fact, it is noteworthy that in their fully adjusted model, this signal failed to reach statistical significance. Nonetheless, possible demographic explanations for the apparent inconsistency between these datasets may be considered worthy of discussion. These include the effect of the slightly younger cohort and multiple ethnicities and regions that were explored in the PURE cohort, which we were unable to capture through UK Biobank which focuses on middle-aged adults. It could be possible that UPF intake exerts a differential effect in different age groups, or that the cumulative UPF intake in one's lifetime should be considered as well. In addition, we cannot exclude the possibility that UPF intake interacts with [epi]genetic predispositions that certainly vary between different ethnicities, which we were unable to correct for as this study was performed on UK data only.
Given the demographic of UK Biobank, we are unable to draw conclusions regarding the possible association of IBD incidence and UPF intake in the paediatric and younger adult population. Nonetheless, the highly significant association between UPF intake and risk of incident CD strengthens our conclusion that CD, but not UC, incidence is associated with UPF intake. This apparently exclusive association between UPF and CD might relate to a greater biological propensity of CD to react to luminal contents in the gut [such as faecal derivatives and nutrients], as evidenced by the efficacy of exclusive enteral nutrition in paediatric CD and the role of diversion of the faecal stream in controlling inflammation in CD. 32,33 Mechanistically, evidence from in vitro, animal, and human trials is emerging to understand how UPFs might drive gut inflammation. As an illustration, certain food additives that are frequently found in UPFs were reported to affect permeability of epithelial cell cultures, 34 induce intestinal inflammation in susceptible mice, and give rise to colonic ulcerations in guinea pigs, resembling those in humans, when administered through their drinking water. 9,35,36 Moreover, a study assessing the effect of another food additive [namely carboxymethylcellulose] in healthy volunteers found a perturbation in the faecal microbiota, with a reduced diversity and a decrease in short chain fatty acids. 37 The study performed by Vasseur et al. in France did not show any relationship between UPF intake and IBD incidence. 13 However, since this prospective cohort using 24-h dietary recall was only able to capture 75 cases over their follow-up period of 2.3 years, this study was probably underpowered to detect a significant effect.
In our study, 24-h dietary recall questionnaires were used as opposed to FFQs that more broadly capture dietary habits over the past month. The latter may lack the granularity of food logs and 24-h dietary recall questionnaires that is necessary to assess UPF intake, and we therefore decided not to      use FFQs in our analysis. Moreover, the FFQs available in UK Biobank are rather brief, which might differ in other studies or regions. Therefore, it is reassuring that even by using different strategies to measure dietary food intakes, consistent results on CD incidence were found. The finding that UPF intake in patients with IBD was almost twice as high as compared with non-IBD participants is novel. However, whether the nutritional habits of these patients could have led to a higher risk of IBD incidence and those habits were maintained after diagnosis, or if patients adapted their diet after diagnosis because of gastrointestinal symptoms, cannot be ascertained and would need to be addressed in a dedicated prospective trial.
Interestingly, when analysing the disease course over time, we found a novel association between UPF intake and risk of benign colorectal neoplasia in CD, and an association between UPF intake and need for IBD-related operations in UC, raising the possibility that UPF intake might impact on IBD course and contribute to adverse events in this patient population. Intriguingly, although cases for IBD-related surgery and [benign] colorectal neoplasia were few, we did find a clear dose-response relationship. This is in line with a recently published prospective cohort study in 1133 IBD patients who were followed up for 3 years, which found a significant association between the consumption of sugar-sweetened beverages [a substantial component in UPF diets] and a decreased time to hospitalisation. 38 Furthermore, a higher intake of these beverages was also associated with disease severity biomarkers and inflammation. Notably-and relevant to our findings-a higher intake of sugar-sweetened beverages in adulthood and adolescence was associated with a higher risk of early-onset colorectal cancer among women. 39 Of course, the same biological explanation as to how UPFs might drive intestinal inflammation discussed above, might also contribute to an unfavourable disease course and complications.
Although evidence from cohort studies and laboratory work supported the current findings of UPF intake and IBDrelated adverse outcomes, the UPF components such as food additives contributing to the IBD-related outcomes remain elusive. 40,41 A study comparing the effect of Mediterranean Diet and the Specific Carbohydrate Diet on CD did not observe any inflammation remission when consuming an elimination diet of food additives. 40 In addition, another study that investigated the exclusive enteral nutrition formulas used for the management of CD, reported that food additives are common ingredients in these nutrition feeds. 41 Thus, the role UPFs in managing IBD is ought to be explored in detail by interventional studies and laboratory experiments.
Our study has several strengths. To our knowledge, this is the first study to investigate the associations between IBD and UPF intake, using different measurements in a cross-sectional fashion, and to study the influence of UPFs on IBD-relevant outcomes. We were also able to correct for possibly important confounding variables such as social deprivation, BMI, comorbidities, and genetic risk. In particular, genetic factors are thought to contribute significantly to the development of IBD, 21 and we made efforts in this study to minimise the influence of genetic susceptibility by adjusting for polygenic risk score. Furthermore, this is a very large cohort study with a similar number of incident IBD cases [841 cases] as compared with the 857 cases of the Nurses' Health study cohort which remains the study with the largest incident cases today looking at UPFs. 14 We acknowledge, however, certain limitations of our study. First, the 24-h dietary recall was only captured once for several participants, making it impossible to account for possible changes in dietary habits of the participants over time. Next, we cannot exclude the possibility that participants adjusted their diet due to gastrointestinal symptoms that later turned out to be caused by new-onset IBD. However a sensitivity analysis, excluding all cases with diagnoses within 1 year after recruitment, yielded similar results, suggesting that this potential phenomenon did not influence our results. Last, it is important to note that UPF refers to the method of [industrial] food processing and not a specific food item per se. Consequently, these types of food products are also typically high-energydense products, high in sugar, unhealthy fats, and salt, and low in dietary fibre, protein, vitamins, and minerals. 2 In addition, colours, flavours, emulsifiers, and other additives are frequently added to make the final product palatable or hyper-palatable. 3 In this study, we adjusted the models for total energy intake, nutrient intake [total fat, carbohydrate, protein intake], and urine sodium, but we were unable to tease out the potential role of food additives separately. The cut-off for misreported energy was set empirically without considering the basal metabolic rate [BMR] and the proportion of individuals whose energy intake below 1.1 × BMR-500 kcal [calculate BMR using the Henry equation] 42 was 5% in the current study. Also, as the UK McCance and Widdowson's food compositions reference from 2002 was used to calculate nutritional values, it is probable that in the past 20 years these estimates have become less accurate. Further laboratory and clinical research aimed at these compounds specifically will be critical to determine their role in driving IBD risk and outcomes. In conclusion, in this nationwide cross-sectional and prospective cohort study of over 180 000 participants, we report an association between UPF consumption and incidence of CD, but not of UC. Furthermore, we found that UPF intake is higher in IBD patients than in non-IBD controls and that this might impact on disease outcomes. Taken together, we provide further evidence to implicate UPFs in the development and disease course of IBD, which might represent a promising strategy in tackling its globally increasing incidence. Further mechanistic and epidemiological research will be needed to further understand the biological basis for these findings and the impact of UPF intake in the developed and undeveloped worlds. Last, the influence of UPF on IBD incidence in all age groups will need further consideration, most notably in areas where IBD incidence in the paediatric population is increasing rapidly.
Researchers can request the data we used and approval from the UK Biobank at: [www.ukbiobank.ac.uk/].