Dietary Inflammatory Indices Are Not Associated With Inflammatory Bowel Disease Incidence and Progression

Introduction

Inflammatory bowel diseases (IBD), including Crohn’s disease (CD) and ulcerative colitis (UC), are thought to arise from inappropriate and maladaptive stimulation of the immune system. Emerging evidence demonstrates that environmental factors, including the diet, may play an important role in disease pathogenesis.¹ Patients are in need of guidance regarding which foods to eat and to avoid in order to prevent or control IBD. Dietary inflammation indices, including the empirical dietary inflammatory pattern (EDIP) based on food groups and dietary inflammatory index (DII) based on nutrients, were previously established to assess the overall inflammatory property of a diet.^2,3 The EDIP was based on the Nurses’ Health Study (NHS), with 18 selected food groups to represent dietary inflammatory potential.³ Briefly, 39 food groups were regressed with plasma inflammatory biomarker levels by reduced-rank regressions and stepwise linear regressions. The DII score, in contrast, was constructed by using data from peer-reviewed research publications through 2010 and leverages 45 dietary factors to predict concentrations of 6 inflammatory markers.² For both indices, higher scores indicate proinflammatory diets. Both scores were shown, using data from the NHS II and the Health Professionals Follow-Up Study (HPFS), to reliably predict concentrations of plasma inflammatory markers.⁴ Later, using the same 3 U.S. cohorts, higher EDIP scores were associated with an increased risk of CD.⁵ However, another study based on the PURE (Prospective Urban Rural Epidemiology) study cohort that included 7 countries (Argentina, Brazil, Canada, Chile, Poland, South Africa, and Sweden), observed a null association for both UC and CD.⁶ Last, the DII has been associated with an increased risk of UC in an Iranian case-control study of only 62 patients.⁷ All scores rely on food composition, with no account for the method of preparation nor degree of food processing.

In view of these discrepancies, we conducted a prospective cohort study using the UK Biobank cohort to validate these dietary inflammation indices and their relationship with IBD incidence and IBD-related clinical outcomes.

Methods

In this study, 121 472 participants from the UK Biobank with 2 to 5 valid 24-hour dietary recall questionnaires were included. Participants were asked about their food intake of the previous day, including 206 foods and 32 drinks. The nutrient calculation was based on the food composition table used in the UK Nutrient Databank.⁸ Consumption of each food and nutrient was the mean intake of all valid questionnaires and consumption of each food group was the sum of all included food. Sixteen components (8 nutrients and 8 foods) of the DII were unavailable in the UK Biobank, leaving 29 of the 45 dietary components for calculation of the score. The design of the UK Biobank has been detailed elsewhere.⁹

Participants free of IBD at baseline (n = 121 472) were followed up for IBD incidence, and participants with prevalent IBD (n = 1408) were followed up for IBD-related clinical outcomes (colorectal cancer, IBD-related surgery, and all-cause mortality). Covariates for adjustment included age, sex, ethnicity, Townsend’s deprivation index (TDI), education level, smoking status, drinking status, physical activity, body mass index, and total energy intake. The primary outcome was the incidence of IBD, CD, and UC, ascertained by health records linked to national hospital inpatient, primary care and death registries (International Classification of Diseases–Ninth Revision and –Tenth Revision codes). Secondary outcomes included the development of colorectal cancer, the need for IBD-related surgery, and all-cause mortality among IBD patients.

Person-years were calculated from the date of the first available 24-hour questionnaire to the date of IBD diagnosis (among general participants) or IBD-related clinical outcome (among IBD patients), death, loss, or the end of follow-up, whichever occurred first. Participants were grouped into quintiles of EDIP and DII scores. The lowest quintile was used as the reference group. Cox proportional hazards regression models were performed to examine the associations of EDIP and DII with the risk of IBD incidence and IBD-related clinical outcomes. A series of sensitivity analyses, including further adjusting for baseline comorbidities represented by Charlson Comorbidity Index, medication use (antibiotics, proton pump inhibitors and nonsteroidal anti-inflammatory drugs), and lag-1 year analysis were conducted to test the robustness of primary findings. Analyses were performed by R 4.2.1 (R Foundation for Statistical Computing). All tests were 2-sided, with a P value < .05 indicating statistical significance.

Results

Among 121 472 eligible participants (mean age 56.2 years, 55.8% female, 96.9% White), we documented 511 incident IBD cases (143 CD and 368 UC) during a mean follow-up of 10.3 years. We did not observe any significant associations between per SD increment of EDIP or DII and IBD incidence. When examining associations by quintiles, neither EDIP (hazard ratio [HR] in quintile 5 vs 1, 1.06; 95% confidence interval [CI], 0.80-1.40; P trend = .287) nor DII (HR in quintile 5 vs 1, 1.01; 95% CI, 0.74-1.36; P trend = .893) was associated with IBD risk in any model (Table 1). When considering CD and UC separately, a null association was observed for both CD (HR for EDIP in quintile 5 vs 1: 1.14; 95% CI, 0.65-1.99; P trend = .696; HR for DII in quintile 5 vs 1: 1.20; 95% CI, 0.70-2.06; P trend = .454) and UC (HR for EDIP in quintile 5 vs 1 = 1.03; 95% CI, 0.75-1.42; P trend = .315; HR for DII in quintile 5 vs 1 = 0.94; 95% CI, 0.65-1.35; P trend = .532) (Table 1). Similarly, neither EDIP nor DII was associated with the development of colorectal cancer, the need for IBD-related surgery, or all-cause mortality among IBD patients (Table 2). The null findings remained consistent in all sensitivity analyses.

Discussion

In this large and independent cohort study involving 121 472 participants, we computed 2 dietary inflammation indices based on food groups and nutrients to assess their associations with IBD risk. We found no associations between EDIP, DII, and the risk of IBD incidence and progression.

This contrasts the findings from the North American prospective study, showing a positive association between EDIP scores and CD incidence. The PURE cohort, on the other hand, found only a similar trend. Compared with the first study that was partly based on the same cohort used for the development of the score, the PURE study cohort included a global population with different ethnicities. Compared with these studies, we considered more confounding factors and additionally investigated the associations with the progression of IBD. Despite multiple sensitivity analyses, our results remained negative. Given our current results, the validity of the EDIP score as a tool for the assessment of the inflammatory potential of dietary patterns in IBD could be called into question. Of note, there are differences in incidence of IBD in our study (11.5 and 29.5 per 100 000 person-years for CD and UC) and the previous study (6.6 and 8.6 per 100 000 person-years for CD and UC in the NHS, NHS II, and HPFS).⁵ Similarly, start time and follow-up periods differ in cohort studies exploring for this topic. The period for the current analysis was 2006 to 2021, while it was 1984 to 2014, 1991 to 2015, and 1986 to 2012 for NHS, NHS II, and HPFS, respectively.⁵ Regrettably, the different measurements for follow-up time limited the normalization. In addition, populations in different regions, ethnicities (eg, United States and United Kingdom), age (eg, middle age in the NHS, NHS II, and HPFS and older age in the UK Biobank) were reported with the dietary discrepancy, thus resulting in different diet-related health outcomes.^10,11 Overall, confounding factors (including age, ethnicity, follow-up time, nutrient intake and response, genetics, family history, antibiotic use, etc.) differ between this and previous studies, limiting the comparison and extrapolation of results, although we have taken confounders into account as much as possible in the analytic models.

First, it needs to be considered that of the 18 selected food groups, fish, tomatoes, and “other vegetables” are said to have proinflammatory properties, whereas pizza, snacks, and fruit juice are mentioned to be anti-inflammatory.³ However, critically, cooking methods are not captured, and one might propose, for example, that fried fish could have a different inflammatory potential than steamed fish due to the formation of advanced glycation end products and contaminants such as trans fatty acids and acrylamide.¹

It is especially noteworthy that we recently reported a positive association between the intake of ultra-processed foods (UPFs) and CD incidence using data from the same UK Biobank cohort, although no associations were observed in UC. The association between CD incidence (not UC) and intake of UPF intake has now been confirmed by a recent meta-analysis including 4 other studies.¹² Thus, in addition to cooking methods, the degree of processing before products are bought for consumption may also be of great importance. To illustrate this point, when we consider the same food group “tomatoes,” this category included fresh tomatoes, tomato juice, and tomato sauce. Using the NOVA classification to assess the processing of these products, the first would be considered NOVA, whereas juice and sauce are at least NOVA,² or processed, if not ultra-processed.¹³ If we presume that the average American consumer is not preparing tomato juice and sauce from scratch, this would make these products NOVA, or ultra-processed. As well as variable degrees of processing and cooking, UPFs also typically have lower nutritional values, contain food additives and other industrial components, and contain contaminants of packaging.¹⁴ These elements might additionally impact on development of disease on top of the raw ingredients. Unfortunately, this type of granularity cannot be assessed using food frequency questionnaires, which potentially limits the validity of the EDIP score. Taken together, we propose that total UPF intake is a better-validated tool to assess healthier dietary patterns in IBD than the proinflammatory scores.

Regarding the DII, we acknowledge that we were only able to capture 29 of the 45 dietary components, which might negatively affect our results. However, the previously mentioned Iranian case-control study was only able to capture 27 items.⁷ In addition, for a Belgian prospective study in healthy volunteers, the yield was as low as 17 dietary components, which calls into question the practicality and usability of this scoring system.¹⁵ Finally, residual bias and confounding cannot be fully avoided in observational studies as well as in our analysis.

Finally, we would like to point out that interindividual differences in response to nutrients and food handling might exist, as was previously illustrated by Armstrong et al,¹⁶ who showed that dietary fibers might have a counterintuitive proinflammatory effect in individuals with active IBD who lack fermentative microbial enzymatic activities. Both human and microbial handling of foods are relevant. A better understanding of these interindividual differences may be key to the development of personalized dietary strategies.¹⁷

Conclusions

In summary, we examined the role of both food group and nutrient-derived inflammatory indices in the development and progression of IBD in the UK Biobank cohort and found null associations. Given the increasing awareness of the importance of diet on intestinal inflammation, this highlights the complexity and variability of dietary patterns, and the emerging need is for well-validated dietary scoring systems that take into account the degree of food processing and cooking techniques as well as the raw ingredients.

Acknowledgments

This research was conducted using the UK Biobank study under Application Number 66354. We want to thank all UK Biobank participants and the management team for their participation and assistance.

Author Contributions

J.C.: Conceptualization (equal), Methodology (equal), Formal analysis (lead), Writing-original draft (lead); T.F.: Methodology (equal), Formal analysis (lead), Writing-original draft (equal); J.W.: Methodology (supporting), Formal analysis (equal), Writing-review and editing (lead), Writing-original draft (lead); Y.Z.: Methodology (supporting), Formal analysis (equal), Writing-original draft (supporting); K.R.: Methodology (supporting), Formal analysis (supporting), Writing-review and editing (supporting); J.S.: Conceptualization (lead), Formal analysis (supporting), Methodology (equal), Project administration (lead), Writing-review and editing (equal); E.T.: Conceptualization (lead), Formal analysis (supporting), Methodology (equal), Project administration (lead), Writing-review and editing (equal); X.L.: Conceptualization (lead), Formal analysis (supporting), Methodology (equal), Project administration (lead), Writing-review and editing (lead). All authors contributed to data acquisition and interpretation, approved the final version of the manuscript.

Funding

This work was supported by the Natural Science Fund for Distinguished Young Scholars of Zhejiang Province (LR22H260001; X.L.), the National Nature Science Foundation of China (No. 82204019; X.L.), a CRUK Career Development Fellowship (C31250/A22804; E.T.); and by a PhD Fellowship Strategic Basic Research grant from the Research Foundation Flanders, Belgium (1S06023N; J.W.).

Conflicts of Interest

There were no financial conflicts of interest among any of the authors.

Data Availability

Data can be requested from the UK Biobank (www.ukbiobank.ac.uk/).

References

Author notes