Skip to main content

Thank you for visiting You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Nutrition in acute and chronic diseases

Food groups associated with immune-mediated inflammatory diseases: a Mendelian randomization and disease severity study



Immune-mediated inflammatory diseases (IMIDs) are prevalent diseases. There is, however, a lack of understanding of the link between diet and IMIDs, how much dietary patterns vary between them and if there are food groups associated with a worsening of the disease.


To answer these questions we analyzed a nation-wide cohort of n = 11,308 patients from six prevalent IMIDs and 2050 healthy controls. We compared their weekly intake of the major food categories, and used a Mendelian randomization approach to determine which dietary changes are caused by disease. Within each IMID, we analyzed the association between food frequency and disease severity.


After quality control, n = 11,230 recruited individuals were used in this study. We found that diet is profoundly altered in all IMIDs: at least three food categories are significantly altered in each disease (P < 0.05). Inflammatory bowel diseases showed the largest differences compared to controls (n ≥ 8 categories, P < 0.05). Mendelian randomization analysis supported that some of these dietary changes, like vegetable reduction in Crohn’s Disease (P = 2.5 × 10−10, OR(95% CI) = 0.73(0.65, 0.80)), are caused by the disease. Except for Psoriatic Arthritis and Systemic Lupus Erythematosus, we have found ≥2 food groups significantly associated with disease severity in the other IMIDs (P < 0.05).


This cross-disease study demonstrates that prevalent IMIDs are associated to a significant change in the normal dietary patterns. This variation is highly disease-specific and, in some cases, it is caused by the disease itself. Severity in IMIDs is also associated with specific food groups. The results of this study underscore the importance of studying diet in IMIDs.


Immune-mediated inflammatory diseases (IMIDs) are a group of pathologies characterized by the dysregulation of immune pathways leading to inflammation, organ damage and multiple comorbidities [1]. Rheumatoid arthritis (RA), psoriatic arthritis (PsA), psoriasis (PS), systemic lupus erythematosus (SLE), Crohn’s disease (CD) and ulcerative colitis (UC) are among the most prevalent IMIDs, collectively affecting ~4% of the population [2]. They are complex diseases caused by an interplay of genetic and environmental factors. In the last 15 years, genome-wide association studies (GWAS) have been highly successful at characterizing the genetic basis of IMIDs [3]. One of the key findings of GWAS is that IMIDs tend to share genetic risk factors [4]. Much less is known, however, on the environmental factors that influence IMIDs. Despite nutrition is a major environmental modifier, there is a lack of well-powered cross-sectional studies characterizing the dietary patterns of IMIDs and their association to disease severity.

There is increasing evidence that recent changes in dietary habits could be behind the increased prevalence of certain inflammatory diseases like type 1 diabetes or asthma [5]. Also, recent studies are demonstrating how nutrition can play an essential role in the regulation of the immune system activity, either through the direct provision of specific compounds [6] or by the modulation of the activity of the intestinal microbiome [7, 8]. For these reasons, there is an increasing interest to characterize the dietary patterns of inflammatory diseases [5]. To date however, there is a lack of large population-based studies evaluating the association of food with IMIDs.

Changes in dietary factors could either reflect habits that contributed to disease risk or changes induced by the disease [9]. Mendelian randomization (MR) analysis is an epidemiological tool that can be very useful to discern what is the most probable causal relationship in these cases [10]. Through the use of MR, we here provide evidence on which altered food habits are caused by the disease. The food frequency changes that cannot be explained as caused by IMIDs are of interest because they might be contributors to disease worsening. Identifying these dietary factors could be very helpful to design simple preventive and interventional strategies to improve the outcome of patients. To this end, we have also performed an association analysis between the different food types and disease severity, as defined for each IMID. Using this approach, we have found several foods that are correlated with an increase of disease activity levels. Together, the simultaneous analysis of six prevalent diseases using different analytical approaches shows that diet is important in IMIDs, and we provide associations that may be in future used to improve the health status of patients suffering from these chronic diseases.


The baseline epidemiological characteristics of the study population are shown in Table 1. Healthy individuals had a similar mean age to IMID patients (49.6 years, although with a narrower distribution, SD = 7.0), lower prevalence of underweight (0.2%), higher prevalence of overweight (47.3%), and elevated rates of high physical activity (46%) and high educational status (70% of individuals having completed tertiary education). The baseline characteristics between the six IMIDs were more related among themselves than against controls. Some features, however, are distinctive of each disease. For example, there was a significant female sex predominance in RA and SLE cohorts (78% and 93%, respectively). This is very close to the commonly reported 3:1 and 9:1 ratio, respectively. Other features that were found to be differential among IMIDs, are less well known or new. This is likely due to the lack of previous cross-IMID comparative studies. For example, PsA patients showed the highest rate of high physical activity (28%) and UC the lowest levels (15%) among the six IMIDs. The prevalence of obesity among PsA and PS patients was 26.5% and 25.2% respectively, the highest of all cohorts, including healthy individuals. PsA and, particularly, RA had the highest mean age (52.3 ± 13.2) and 60.7 ± 12.7) years, respectively), and CD patients were the youngest on average (41.9 ± 13.5). The rate of high educational status for RA patients was the lowest (33%).

Table 1 Baseline epidemiological characteristics of the study population stratified by disease type.

Table 2 shows the unadjusted dietary habits for each of the six IMID cohorts. A high heterogeneity was found across IMIDs, with some food items being consumed similarly between diseases and other food groups showing notable differences. Eggs were consumed rather similarly across all IMIDs, with differences smaller than 4%. On the other hand, fruit consumption showed differences as large as 31% (i.e., between CD and RA patients). All IMID patients systematically showed lower rates of consumption of stimulant and alcoholic beverages compared to healthy individuals. While only 28% of the healthy individuals reported to be abstinent, the lowest abstinence rate observed among IMIDs was 43% in PS patients. The highest rates of alcohol abstention were found in RA (72%) and SLE (78%). The lowest levels of consumption of tea/coffee were found in IBDs, with 38% (CD) and 34% (UC) of non-drinkers, compared to 15% of the control cohort.

Table 2 Unadjusted dietary habits of the six IMID cohorts and healthy controls.

Table 3 shows the association results between the six IMIDs and the 13 food categories adjusted by the two models (Model 1: adjusted for potential confounders, Model 2: adjusted for potential confounders and mediators). At the disease-level, all IMIDs showed a significant association to at least three of the 13 food items in the two regression models considered. The largest diet alteration was found in the two IBDs, with 9 and 8 food items consumed at different rates in CD and UC compared to healthy controls, respectively. RA patients showed changes in 4 food items, while SLE, PS and PSA showed dietary changes in 3 items. At the food-level, 11 categories were consumed differently in at least one IMID. Only eggs and fish showed no significant difference in any disease group against controls. In this case-control analysis, two disease-specific food associations were identified. Both occurred in CD and included a reduced ingestion of dairy products (P = 0.002) and a drastic decrease in vegetable consumption (P < 2 × 10−16). Three food items were found to be similarly changed between two IMIDs: legume consumption was reduced in UC and CD (P = 2 × 10−6 and P = 6 × 10−5, respectively) while rice/pasta was very significantly increased in these two IBDs (PCD = 2 × 10−11 and PUC = 6 × 10−9), and RA and UC patients showed a significant increase in fruit consumption levels (P = 0.01 and P = 3 × 10−6, respectively). UC, CD and PS shared a common increase in bread and/or grain weekly consumption compared to healthy controls (PUC = 5 × 10−10, PCD = 4 × 10−5 and PPS = 0.01). Meat consumption showed an opposite behavior in IBD compared to the two arthritis: while meat ingestion was significantly higher in the diseases involving the gut (PCD = 0.002 and PUC = 0.001), it was reduced in the diseases affecting the joints (PRA = 0.0003 and PPSA = 0.02). The more commonly shared dietary changes in IMIDs were the consumption of tea/coffee and alcoholic beverages. A major prevalence of alcohol abstinence was found in all six diseases (PUC = 4.6 × 10−12, PCD < 2 × 10−16, PPS = 8.5 × 10−13, PPSA = 2.8 × 10−14, PRA < 2 × 10−16, and PSLE = 3.3 × 10−16). Stimulant beverages are also avoided more frequently in IMIDs (PUC < 2 × 10−16, PCD < 2 × 10−16, PPS = 1.1 × 10−5, PRA = 2.6 × 10−5, and PSLE = 2.2 × 10−4) with the only exception of PsA, where there was no significant difference compared to controls.

Table 3 Comparison of dietary habits between the six IMIDs and healthy controls.

To help discern which food frequency patterns are caused by the presence of the disease we performed a Mendelian randomization (MR) analysis. Table 4 shows the MR results for the 13 food items. At the disease-level, all IMIDs but PSA showed significant associations. Both CD and UC showed the largest number of food group associations (n = 5), followed by PS (n = 4), and RA and SLE (n = 2). At the food-level, associations caused by the disease were found for all food groups except for legume consumption. Abstinence from alcohol and stimulating drink beverages was found to be associated in all IMIDs with the exception of PsA (PUC = 0.004, PCD = 9.7 × 10−8, PPS = 0.002, PRA = 2.2 × 10−7, and PSLE = 3.6 × 10−11). At the disease-specific level, a strong association with vegetable reduction was found in CD (P = 2.5 × 10−10). The development of either IBD was associated with an increase in sweets in the diet (PCD = 0.009 and PUC = 0.007). Meat consumption frequency disparity between UC and RA observed in the case-control analysis (i.e., high and low, respectively) was also found to be caused by each disease (PUC = 0.02 and PRA = 0.0001). Fruit consumption was also associated using MR analysis, revealing an association with the reduction in frequency in CD and PS (PCD = 0.0001 and PPS = 0.012).

Table 4 Mendelian randomization analysis for dietary habits on the six IMIDs.

The association between disease activity levels and food consumption within each IMID is shown in Table 5. The two alternative association models (M1: adjustment for confounders, and M2: adjustment for confounders and mediators) were also used to evaluate this association. At the disease-level, an increase in disease severity with food consumption was found in the two IBDs, RA and PS. Conversely, dietary variation was not associated with an increase in inflammatory status in PsA or SLE. Among the IMIDs showing an association between food frequency and activity, CD showed the larger number of food item associations (n = 5), followed by UC (n = 3), RA (n = 2) and PS (n = 2). At the food-level, 8 of the 13 food types were associated with disease activity in one or more IMIDs. Of these, associations were predominantly disease-specific. Active CD patients were found to consume less dairy and less vegetables than inactive patients (P = 0.03 and P = 1.7 × 10−5, respectively). Higher fish consumption was associated with high disease activity in UC patients (P = 0.048). RA patients with high disease activity showed a highly significant reduction in alcohol consumption (P = 6.7 × 10−5). Active PS patients were found to consume more rice/pasta than patients with low activity (P = 0.003). Shared dietary changes associated with disease activity were found between the two IBDs and between CD and RA. In the former, fruit consumption was markedly reduced in patients with an active disease (PCD = 0.00014 and PUC = 2.7 × 10−5), and in the latter, tea/coffee was found to be reduced in RA and CD with more severe features (PRA = 0.006 and PCD = 0.026). Finally, an opposite consumption pattern was found for legumes: while CD and UC patients with high disease activity reduce its consumption (PCD = 0.027 and PUC = 0.02), PS patients with a more severe disease consumed it more frequently (PCD = 0.009). Eggs, processed meat, bread/grain, meat and sweets were not associated to the presence of more severe symptoms in any of the six IMID cohorts.

Table 5 Association of dietary habits with disease severity in IMIDs.


Immune-mediated inflammatory diseases are a prevalent group of common disorders caused by the interplay of genetic and environmental factors. While a large fraction of the genetic component has been identified in the last years, very little is known on the relevance of a key environmental factor like the diet. Here we provide a simultaneous analysis of the food habits of six of the most prevalent IMIDs, comparing them to healthy individuals. Using a nation-wide cohort, we find that diet is profoundly altered in all IMIDs. With a Mendelian randomization analysis, we have identified which food habits are likely caused by the disease and differentiate them from those that could be contributing to the disease. To provide additional evidence of the latter, we performed an association between the diet and severity, and found evidence that the eating frequency of particular food items are associated with disease symptom aggravation.

Eleven out of the 13 food groups were consumed differently from controls in at least one IMID. Applying a Mendelian randomization analysis, several of these differences can be ascribed to the presence of the disease. Among these, the increase in the abstinence from alcoholic drinks is common to all six IMIDs. For some of the diseases, like RA [11], IBD [12] and SLE [13] there is previous evidence supporting this reduction in comparison to controls. This is the first time that the relationship between this dietary habit and IMIDs has been tested using a MR framework. Our results support that the reduction in alcohol consumption is largely due to the presence of the inflammatory disease. However, our results also show that this drastic reduction in alcohol intake in IMIDs is correlated with changes in disease activity. In fact, we found the evidence of the contrary in RA. In this IMID we found that patients with higher levels of disease activity consumed less alcohol than patients with lower inflammatory and pain markers. Although this result might seem contradictory given the associated toxicity, there is previous epidemiological evidence showing that alcohol consumption in inversely associated with disease severity in RA [11]. In line with our findings, recent experimental evidence demonstrates that ethanol consumption is able to mitigate autoimmune arthritis, and that is done by targeting key immunological processes associated with RA etiology [14].

Our MR analysis also provides evidence of disease-specific dietary changes. The increase of meat consumption in the two IBDs is a clear example. To our knowledge, it is the first time that this change in food frequency has been described for these two diseases. Red meat in particular, carries high amounts of tryptophan which has been associated with gut homeostasis [5]. UC patients with higher disease activity were also found to have a reduced ingestion of meat, further supporting this potential protective effect of meat in IBD. In RA, we found an opposite pattern to that of IBD, with patients eating lower quantities of meat. In this case, rather than being due to a specific nutritional aspect of this food, it could be explained by a specific functional feature of RA. In this IMID, inflammation tends to occur in the hands (in our RA cohort >98% of patients had joint erosions in hands), and is generally expressed as a symmetric arthritis that causes inability to perform more strenuous manual tasks like as cutting meat [15].

The MR analysis also showed that the development of CD is responsible for the drastic reduction in vegetable consumption observed in these patients. This highly significant dietary change, however, was not observed in UC, the other IMID targeting the gut. While both IBDs involve the colon mucosa, only in CD inflammation occurs in other sections of the gastrointestinal tract, which tends to be the ileal section. This dietary reduction could be heavily influenced by medical and dietary specialists. Many dietetic associations recommend patients from both IBDs to reduce their vegetable intake to facilitate digestion [16], particularly when the inflammatory symptoms worsen. Our data, however, indicates that this dietary change only occurs in CD patients and not in UC patients. Also, CD patients with a more active disease were found to consume lower levels of vegetables compared to CD patients with milder or no symptoms. The reason for this difference could be due to the digestive process of vegetables and how it negatively impacts inflammation at specific regions of the gut. In particular, this highly significant association in CD supports that vegetables in the diet would affect the inflammation in the ileum but not the colon. Supporting this hypothesis, we also found that CD patients with strictures -a narrowing of the wall of the small intestine that is induced by inflammation- consume less vegetables that patients without this obstruction (OR(CI) = 0.79 (0.64–0.98), P = 0.038). Together, our results suggest that dietary recommendations involving vegetable consumption should distinguish between CD and UC. While our data clearly supports that high vegetable consumption is detrimental in CD and that is should be reduced in patients with more active disease, it doesn’t support a negative impact on UC. Vegetables provide many essential nutrients like vitamins, and therefore, a reduction should be adequately justified by evidence which, in the case of UC, is missing.

Conversely, in UC we identified an increased consumption of fruits compared to healthy individuals, which was also significant in the MR analysis. This increase of fruit consumption in the diet was clearly not observed in CD, in neither the case-control nor the MR analyses. To our knowledge, there is no previous evidence reporting this specific dietary behavior in UC. Furthermore, this result is in apparent contradiction with what would be expected according to the dietary recommendations from different clinical nutrition societies, where fruit consumption reduction is advised [16, 17]. However, when analyzing the association of this food group with disease severity, both IBDs showed a significant decrease in fruit consumption when the disease is more active (Table 3). Therefore, this result suggests that, when inflammatory symptoms appear in both IBDs, patients tend to follow the suggested recommendations and reduce fruit consumption. However, this result does not explain the significant increase in fruit consumption of UC patients globally compared to controls. Compared to vegetables, fruit fibers tend to be more fermentable and have a greater impact reducing transit time [18]. An increase in fruit consumption therefore could have been acquired by UC as a strategy to minimize the impact of digested food on the inflamed mucosa.

The dietary changes in IMIDs that were found to be significantly different from controls but were not associated in the MR analysis could be indicative of food groups that contribute to the disease. In this group of dietary associations, we found that IBD patients consumed higher quantities of rice and pasta than healthy individuals. There is previous evidence that starchy foods could be a risk factor for IBD [19]. Analysis of fecal matter of CD patients has shown that they have a major reduction of microbiota involved in the fermentation of resistant starch [20]. In PS, we found a higher consumption of bread and grain compared to controls. Antibodies against gluten, anti-gliadin IgA antibodies, have been found to be augmented in PS patients compared to controls, even in the absence of celiac disease or non-celiac gluten sensitivity [21, 22]. In a previous study in the Scandinavian population, the prevalence of anti-gliadin antibodies was found to be 16% in PS patients, while the positivity in the general population was only 1% [23]. Both PS and celiac disease are inflammatory diseases that tend to co-occur, and GWAS have shown that there is genetic risk overlap between them [24]. Our results provide evidence in favor of a common etiological factor for these two diseases. Gut permeability and inflammation associated with celiac disease should therefore be contemplated as a potential causal mechanism in PS etiology.

We have also found several dietary habits in IMIDs that are not different from healthy controls but are significant when comparing patients with different levels of disease severity. This result shows the importance of incorporating the disease activity in the analysis of the diet, particularly in a group of diseases where patients fluctuate in their lifetime between periods of highly damaging flares and periods of much milder inflammation. In our study, PS patients with more severe disease (represented by a larger skin inflammatory involvement) showed an increase in the weekly consumption of legumes and rice and pasta. Based on this result, both food groups could be contributing to the increase in disease activity. As described previously, the high presence of anti-gliadin antibodies in PS patients promote a subclinical gut inflammatory process when pasta is increased in the diet. This activation of the immune system could have harmful effects that could extend beyond the gut and, in this case, increase the level of inflammatory activation in the skin. Legumes are highly abundant in lectins, a type of glycoproteins that are known to bind strongly to the surface of epidermal cells [25], including keratinocytes and subsets of epidermal Langerhans cells [26]. Experimental studies have shown that the epidermis from PS patients has a lower capacity to bind to lectins, and that this could be due to the previous occupancy in this cells with lectins from the diet [27]. According to these results, pasta and legumes are two types of food that promote worsening of symptoms in PS. Studying these associations in depth could help to reveal relevant pathogenic mechanisms for this IMID.

The observational nature of the present study involves limitations. Self-report of food intakes involves the generation of noisier statistical estimates [28]. However, the use of a large nation-wide cohort of >13,000 individuals is a strong measure to improve the statistical power of the results [29]. The simultaneous analysis of six IMIDs is a unique feature of this study, allowing to directly evaluate the reproducibility of dietary factor associations as well as demonstrating that diseases with a closer pathology (e.g., CD and UC, PS and PSA) have also similar dietary patterns. To further increase the robustness of our results, we performed an exhaustive search and adjustment for potential confounders. Dietary habits can be affected by many epidemiological features, and it is essential to reduce their impact in the study. Here, we used common epidemiological variables like age, sex and smoking status, but also we controlled for geographical location of the individuals, the season of the year, the educational level and, in the case of IMID patients, the number of years since diagnosis. All these variables can have an important impact in the dietary habits of individuals but are rarely incorporated jointly in dietary studies. Finally, in this study the MR framework could be used to determine the causality of IMIDs over dietary changes, but the contrary could not be directly analyzed. The lack of genetic variants -instrumental variables- associated with the frequency of the different food groups prevented to test this causal association. However, the identification of foods associated with a worsening on disease activity is a powerful alternative strategy. If these findings are validated in a controlled clinical trial, they could provide a simple and inexpensive way to reduce disease severity in IMIDs.

Our study provides a better understanding of the relationship between diet and immune-mediated inflammatory diseases. We show that this group of common diseases have dietary patterns that are different to that of healthy individuals. Changes in the frequency of food are not restricted to the diseases that affect the digestive system, CD and UC, but also occur in IMIDs that target other organs and tissues. Mendelian randomization allowed to identify which changes are caused by the disease and differentiate from those that could be involved in the disease pathology. The identification of food frequency habits that correlate with disease severity levels, provides powerful additional evidence in this direction. In the forthcoming years, the study of diet should be prioritized if we want to fully understand the complexity of IMIDs.


Study population

The patients and controls of this study were recruited on a multicenter collaborative project by the Immune-Mediated Inflammatory Diseases Consortium (IMIDC) [30]. The IMIDC is a nation-wide network of clinical, biology and epidemiological researchers in Spain focused on the study of IMIDs. In the recruitment of the IMID patients a total of 73 university hospital clinical departments participated, including rheumatology (for RA, PsA, and SLE), dermatology (for PS and PsA) and gastroenterology departments (for UC and CD). Simultaneously, a control group of healthy subjects form the same Spanish regions were also recruited.

All the patients included in this cross-sectional cohort fulfilled the consensus diagnosis criteria of each IMID (Supplemental Methods). The healthy group subjects were recruited from blood donors attending to the same university hospitals. Healthy subjects having first- or second-order relatives affected with an IMID were excluded from the study. All patients and controls were over 18 years old at the time of recruitment, were born in Spain and had also all four grandparents born in Spain. The data were collected from June 2007 to December 2012. The sample size at the end of the recruitment period consisted of 13,358 subjects, including 2282 RA, 2277 PS, 1481 PsA, 1070 SLE, 1723 UC and 2475 CD patients and 2050 healthy controls (Supplementary Fig. 1).

Epidemiological variables considered in the study included age, sex, number of years since diagnosis, smoking status (at diagnosis and present), place of residence, season of the year, and educational level. Place of residence was categorized at the province level (50 total) and education was summarized as having completed university-level studies (yes/no). Biometric variables weight and height were also measured on the same day the diet was assessed, as well as the level of physical activity (i.e., doing regular exercise during leisure or at work, yes/no). Informed consent was obtained from all participants, and protocols were reviewed and approved by local institutional review boards. This study was conducted in accordance with the Helsinki Declaration of 1975 as revised in 1983.

Assessment of disease severity

Simultaneous to dietary assessment, disease severity of IMID patients was recorded at the day of visit to the hospital by the clinician. For each disease, established severity scores were measured: the Disease Activity Score for 28 joints (DAS28) for RA and PsA [31], the Psoriasis Area and Severity Index (PASI) for PS [32], the British Isles Lupus Assessment Group (BILAG) for SLE [33], the Harvey-Bradshaw Index for CD [34] and the Lichtiger Score for UC [35]. All the physicians participating in the study were trained to follow the same criteria.

Dietary assessment

The participants completed an FFQ during their visit to the hospital. The FFQ included 11 major food groups: fruits, vegetables (excluding legumes), meat, processed meat, fish, eggs, dairy products, bread and/or wholegrain, rice and/or pasta and/or potatoes, sweet products (e.g., pastry, marmalades) and legumes. For each food category, participants indicated their average frequency of consumption over the previous week by selecting between four different categories: “0 times a week”, “1 or 2 times a week”, “3–5 times a week” and “6 or 7 times a week”. Stimulating beverage drinking habits were measured by asking for the average number of cups of tea or coffee consumed per day. Finally, alcohol drinking habits were classified into three categories: “daily” when respondents reported drinking beer, wine or spirits during the week, “sometimes” when they only drank occasionally on weekends, and ‘abstinent’ otherwise.

Statistical analyses

Quality control analysis was performed on the dietary, epidemiological and clinical data. 1143 individuals showed high level of missingness (>50%) along the different questionnaires and were excluded from the study. Since the remaining missing proportions were less than 5% for each variable, an available-case analysis was performed. Quality control of dietary habits was performed by principal component analysis of the food item responses (excluding stimulating and alcoholic beverages). For this objective, frequency categories were numerically coded as 0, 1.5, 4 and 6.5 days/week. The first three components (i.e., eigenvalues >1) were evaluated for the presence of outliers or possible confounders. This analysis revealed a group of n = 354 consecutive participants from the same center that showed a significant deviation in the first principal component (Supplementary Fig. 2, P < 1e−127). In order to avoid the risk of confounding, this outlying group of participants was excluded from downstream analyses.

Diet categories were dichotomized into high- and low-consumption levels in order to perform the posterior analyses using logistic regression models. Dichotomization was done as follows: food categories where more than 50% of the sample had the maximum level of consumption (“6 or 7 times a week”, i.e., dairy, bread/grain and fruits), this level was coded as “1” (high) and the remaining consumption levels as “0” (low). The rest of food categories had “1 or 2 days per week” or “3–5 days per week” as their mode. For these variables, the four levels of consumption were dichotomized aggregating the two superior (high) and the two inferior (low) levels. Regarding stimulating beverage and alcohol consumption, variables were also binarized, dividing individuals into abstinent or drinkers. In the case of alcohol, week-end only drinkers were considered as abstinent.

Association testing with IMIDs was performed using logistic regression analyses with the binarized diet variables as the outcome. To examine the differences between IMID patients and the healthy group, an indicator variable for disease type was introduced as predictor, with the healthy group as the reference level. Relevant adjustment covariates were also included in the logistic model as predictors. Wald tests of the coefficients were used to test for significant differences against the control group on the log-odds scale. Multiple-test adjustment using Dunnett’s method and adjusted estimates of the rates of high consumption for each cohort were determined with the emmeans R package.

To test for association with disease severity, the previous regression models were modified by adding an interaction term between disease type and disease activity and by excluding the healthy control cohort. Main effects were considered for disease type but not for disease activity, so that disease activity effects were nested within disease type. Wald tests of the coefficients corresponding to the interactions were used to test for significant trends in the rate of high consumption when changing from low disease activity to high disease activity.

Adjustment for nondietary exposures

All the regressions for the food items were performed using two adjustment models. In Model 1 we adjusted for all potential confounders. As confounders, we considered those variables that were likely to influence and not be influenced by dietary habits and, simultaneously, were associated with at least one of the IMIDs. Age, sex and smoking status at diagnosis were included since they are known risk factors for IMIDs [36] and likely to be related with dietary habits. In this model we also included geographical region, season of the year and highest educational level as covariates. In Model 2, we adjusted for both for potential confounders as well as for potential mediators. We replaced smoking status at diagnosis for the present smoking status and additionally adjusted for the current physical activity habits. These covariates represent plausible mechanisms by which the disease might be indirectly affecting eating habits.

The same two-model approach was applied to study associations between disease activity and diet by including an interaction term between disease activity and the disease factor. In this case, the control cohort was excluded and the model was additionally adjusted for the number of years since diagnosis. In most IMIDs chronic exposure to the disease aggravates the severity of the disease and, consequently, it was considered a potential confounder.

Exploratory analysis of the different covariables showed non-linear relationships between age and BMI with the rates of high vs. low food consumptions (Supplementary Fig. 3). For this reason, age and BMI were converted into categorical variables. Age was divided into six age ranges according to quantiles (i.e., years 18–34, 35–42, 43–48, 49–56, 57–63 and 63–92) and BMI was categorized into ‘Underweight’, ‘Normal’, ‘Overweight’ and ‘Obese’, according to the WHO classification [37]. In the case of age, we also included the continuous predictor since both versions appeared to be independent significant factors for most of the outcomes.

In order to incorporate geography as a covariate, we used the most detailed residence information we had from the cohort individuals. Here it was represented by provinces; Spain is divided into 50 different provinces that range from 1090 km2 (Guipúzcoa) to 21,766 km2 (Badajoz). we approached the issue of having regions with relatively small sample size by aggregating individuals into larger regions. Sensitivity analyses were performed to see whether alternative region definitions had a relevant impact on the conclusions. We found that using an eight-region variable (resulting from aggregating neighboring provinces), the results were not qualitatively affected. For the analyses performed in present study we therefore used this more parsimonious version.

Bivariate analyses with the baseline variables were performed using ANOVA for age and chi-square tests for categorical variables. All the analyses were performed with R software, version 3.4.4 [38]. The level of significance was set at a two-sided p value of 0.05.

In order to test the causal hypothesis that IMIDs cause the observed changes in the diet, we performed a Mendelian randomization (MR) analysis. GWAS data was available for 7554 subjects (Supplementary Fig. 1). For this objective we used the genetic risk scores for each disease as the instrumental variables (IV). Genetic risk scores were calculated following recently described methodology (Supplementary Methods) [39]. The complete list of genetic variants included per each IMID are listed in Supplementary Table 1. The ratio method with bootstrap was then used to obtain empirical confidence intervals and variance of the ratio estimator [40]. The beta coefficients of the ratio method were obtained by logistic regression for both the IV-outcome and the IV-exposure. The 10 first principal components of genetic variation were calculated using the EIGENSTRAT method [41] and included in the models to account for potential population stratification. P-values were obtained by approximating the distribution of the ratio estimator by a normal distribution and applying a Wald-like test.


  1. 1.

    Bayry J, Radstake TR. Immune-mediated inflammatory diseases: progress in molecular pathogenesis and therapeutic strategies. Expert Rev Clin Immunol. 2013;9:297–299.

    CAS  Article  Google Scholar 

  2. 2.

    Davidson A, Diamond B. Autoimmune diseases. N. Engl J Med. 2001;345:340–50.

    CAS  Article  Google Scholar 

  3. 3.

    Gutierrez-Arcelus M, Rich SS, Raychaudhuri S. Autoimmune diseases: connecting risk alleles with molecular traits of the immune system. Nat Rev Genet. 2016;17:160–74.

    Article  Google Scholar 

  4. 4.

    Ellinghaus D, Jostins L, Spain SL, Cortes A, Bethune J, Han B, et al. Analysis of five chronic inflammatory diseases identifies 27 new associations and highlights disease-specific patterns at shared loci. Nat Genet. 2016;48:510.

    CAS  Article  Google Scholar 

  5. 5.

    Thorburn AN, Macia L, Mackay CR. Diet, metabolites, and “western-lifestyle” inflammatory diseases. Immunity. 2014;40:833–42.

    CAS  Article  Google Scholar 

  6. 6.

    Kleinewietfeld M, Manzel A, Titze J, Kvakan H, Yosef N, Linker RA, et al. Sodium chloride drives autoimmune disease by the induction of pathogenic T H 17 cells. Nature. 2013;496:518–22.

    CAS  Article  Google Scholar 

  7. 7.

    Maslowski KM, Mackay CR. Diet, gut microbiota and immune responses. Nat Immunol. 2011;12:5–9.

    CAS  Article  Google Scholar 

  8. 8.

    Marino E, Richards JL, McLeod KH, Stanley D, Yap YA, Knight J, et al. Gut microbial metabolites limit the frequency of autoimmune T cells and protect against type 1 diabetes. Nat Immunol. 2017;18:552.

    CAS  Article  Google Scholar 

  9. 9.

    Davey Smith G, Ebrahim S. ‘Mendelian randomization’: can genetic epidemiology contribute to understanding environmental determinants of disease? Int J Epidemiol. 2003;32:1–22.

    Article  Google Scholar 

  10. 10.

    Haycock PC, Burgess S, Wade KH, Bowden J, Relton C, Davey Smith G. Best (but oft-forgotten) practices: the design, analysis, and interpretation of Mendelian randomization studies. Am J Clin Nutr. 2016;103:965–78.

    CAS  Article  Google Scholar 

  11. 11.

    Maxwell JR, Gowers IR, Moore DJ, Wilson AG. Alcohol consumption is inversely associated with risk and severity of rheumatoid arthritis. Rheumatology. 2010;49:2140–2146.

    CAS  Article  Google Scholar 

  12. 12.

    Limdi JK, Aggarwal D, McLaughlin JT. Dietary practices and beliefs in patients with inflammatory bowel disease. Inflamm Bowel Dis 2016;22:164–70.

    Article  Google Scholar 

  13. 13.

    Hardy CJ, Palmer BP, Muir KR, Sutton AJ, Powell RJ. Smoking history, alcohol consumption, and systemic lupus erythematosus: a case-control study. Ann Rheum Dis 1998;57:451–455.

    CAS  Article  Google Scholar 

  14. 14.

    Azizov V, Dietel K, Steffen F, Dürholz K, Meidenbauer J, Lucas S, et al. Ethanol consumption inhibits T FH cell responses and the development of autoimmune arthritis. Nat Commun. 2020;11:1–14.

    Article  Google Scholar 

  15. 15.

    Kirwan J, REEBACK JS. Stanford Health Assessment Questionnaire modified to assess disability in British patients with rheumatoid arthritis. Rheumatology. 1986;25:206–209.

    CAS  Article  Google Scholar 

  16. 16.

    MedlinePlus. Crohn Disease, (2020).

  17. 17.

    Brown AC, Rampertab SD, Mullin GE. Existing dietary guidelines for Crohn’s disease and ulcerative colitis. Expert Rev Gastroenterol Hepatol. 2011;5:411–25.

    Article  Google Scholar 

  18. 18.

    De Vries J, Birkett A, Hulshof T, Verbeke K, Gibes K. Effects of cereal, fruit and vegetable fibers on human fecal weight and transit time: a comprehensive review of intervention trials. Nutrients. 2016;8:130.

    Article  Google Scholar 

  19. 19.

    Tragnone A, Valpiani D, Miglio F, Elmi G, Bazzocchi G, Pipitone E, et al. Dietary habits as risk factors for inflammatory bowel disease. Eur J Gastroenterol Hepatol. 1995;7:47–51.

    CAS  PubMed  Google Scholar 

  20. 20.

    Kang S, Denman SE, Morrison M, Yu Z, Dore J, Leclerc M, et al. Dysbiosis of fecal microbiota in Crohn’s disease patients as revealed by a custom phylogenetic microarray. Inflamm Bowel Dis. 2010;16:2034–42.

    Article  Google Scholar 

  21. 21.

    Kolchak NA, Tetarnikova MK, Theodoropoulou MS, Michalopoulou AP, Theodoropoulos DS. Prevalence of antigliadin IgA antibodies in psoriasis vulgaris and response of seropositive patients to a gluten-free diet. J Multidiscip Healthc. 2018;11:13.

    Article  Google Scholar 

  22. 22.

    Nagui N, El Nabarawy E, Mahgoub D, Mashaly H, Saad N, El‐Deeb D. Estimation of (IgA) anti‐gliadin, anti‐endomysium and tissue transglutaminase in the serum of patients with psoriasis. Clin Exp Dermatol: Exp Dermatol. 2011;36:302–304.

    CAS  Article  Google Scholar 

  23. 23.

    Ludvigsson JF, Card TR, Kaukinen K, Bai J, Zingone F, Sanders DS, et al. Screening for celiac disease in the general population and in high-risk groups. U Eur Gastroenterol J. 2015;3:106–20.

    Article  Google Scholar 

  24. 24.

    Lundin KE, Wijmenga C. Coeliac disease and autoimmune disease—genetic overlap and screening. Nat Rev Gastroenterol Hepatol. 2015;12:507.

    CAS  Article  Google Scholar 

  25. 25.

    Reano A, Faure M, Jacques Y, Reichert U, Schaefer H, Thivolet J. Lectins as markers of human epidermal cell differentiation. Differentiation. 1982;22:205–10.

    CAS  Article  Google Scholar 

  26. 26.

    Schuler G, Romani N, Linert J, Shevach EM, Stingl G. Subsets of epidermal Langerhans cells as defined by lectin binding profiles. J Invest Dermatol. 1983;81:397–402.

    CAS  Article  Google Scholar 

  27. 27.

    Freed DL. Lectins in food: their importance in health and disease. J Nutritional Med. 1991;2:45–64.

    Article  Google Scholar 

  28. 28.

    Neuhouser ML. Red and processed meat: more with less? Am J Clin Nutr. 2020;111:252–255.

    Article  Google Scholar 

  29. 29.

    Prentice RL, Huang Y. Nutritional epidemiology methods and related statistical challenges and opportunities. Stat theory Relat Fields. 2018;2:2–10.

    Article  Google Scholar 

  30. 30.

    Alonso A, Julià A, Vinaixa M, Domènech E, Fernández-Nebro A, Cañete JD, et al. Urine metabolome profiling of immune-mediated inflammatory diseases. BMC Med. 2016;14:133.

    Article  Google Scholar 

  31. 31.

    Prevoo M, Van’t Hof M, Kuper H, Van Leeuwen M, Van de Putte L, Van Riel P. Modified disease activity scores that include twenty‐eight‐joint counts development and validation in a prospective longitudinal study of patients with rheumatoid arthritis. Arthritis Rheum. 1995;38:44–48.

    CAS  Article  Google Scholar 

  32. 32.

    Mattei P, Corey K, Kimball A. Psoriasis Area Severity Index (PASI) and the Dermatology Life Quality Index (DLQI): the correlation between disease severity and psychological burden in patients treated with biological therapies. J Eur Acad Dermatol Venereol. 2014;28:333–337.

    CAS  Article  Google Scholar 

  33. 33.

    Hay E, Bacon P, Gordon C, Isenberg D, Maddison P, Snaith M, et al. The BILAG index: a reliable and valid instrument for measuring clinical disease activity in systemic lupus erythematosus. QJM: Int J Med. 1993;86:447–58.

    CAS  Google Scholar 

  34. 34.

    Harvey R, Bradshaw J. A simple index of Crohn’s-disease activity. Lancet. 1980;315:514.

    Article  Google Scholar 

  35. 35.

    Lichtiger S, Present DH, Kornbluth A, Gelernt I, Bauer J, Galler G, et al. Cyclosporine in severe ulcerative colitis refractory to steroid therapy. N Engl J Med. 1994;330:1841–1845.

    CAS  Article  Google Scholar 

  36. 36.

    Hutchinson D, Shepstone L, Moots R, Lear J, Lynch M. Heavy cigarette smoking is strongly associated with rheumatoid arthritis (RA), particularly in patients without a family history of RA. Ann Rheum Dis. 2001;60:223–227.

    CAS  Article  Google Scholar 

  37. 37.

    Organization WH. The use and interpretation of anthropometry: report of a WHO expert committee. World Health Organ Tech Rep. Ser. 1995;854:312–409.

    Google Scholar 

  38. 38.

    R Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.

  39. 39.

    Torkamani A, Wineinger NE, Topol EJ. The personal and clinical utility of polygenic risk scores. Nat Rev Genet. 2018;19:581.

    CAS  Article  Google Scholar 

  40. 40.

    Burgess S, Small DS, Thompson SG. A review of instrumental variable estimators for Mendelian randomization. Stat Methods Med Res. 2017;26:2333–55.

    Article  Google Scholar 

  41. 41.

    Price AL, Patterson NJ, Plenge RM, Weinblatt ME, Shadick NA, Reich D. Principal components analysis corrects for stratification in genome-wide association studies. Nat Genet. 2006;38:904–909.

    CAS  Article  Google Scholar 

Download references


The IMID Consortium includes the following: Eduardo Fonseca, Jesús Rodríguez, Patricia Carreira, Valle García, José A. Pinto-Tasende, Lluís Puig, Elena Ricart, Francisco Blanco, Jordi Gratacós, Ricardo Blanco, Víctor Martínez Taboada, Emilia Fernández, Isidoro González, Fernando Gomollón García, Raimon Sanmartí, Ana Gutiérrez, Àlex Olivé, José Luís López Estebaranz, Esther García-Planella, Juan Carlos Torre-Alonso, José Luis Andreu, David Moreno Ramírez, Benjamín Fernández, Mª Ángeles Aguirre Zamorano, Pablo de la Cueva, Pilar Nos Mateu, Paloma Vela, Francisco Vanaclocha, Héctor Corominas, Santiago Muñoz, Joan Miquel Nolla, Enrique Herrera, Carlos González, José Luis Marenco de la Fuente, Maribel Vera, Alba Erra, Daniel Roig, Antonio Zea, María Esteve, Carlos Tomás, Pedro Zarco, José María Pego, Cristina Saro, Antonio González, Mercedes Freire, Alicia García, Elvira Díez, Georgina Salvador, César Díaz-Torne, Simón Sánchez, Alfredo Willisch Domínguez, José Antonio Mosquera, Julio Ramírez, Esther Rodríguez Almaraz, Núria Palau, Raül Tortosa, Mireia López, Andrea Pluma, Adrià Aterido. We would like to thank Dr Eduard Cabré for stimulating discussions.

IMID Consortium

Eduardo Fonseca17, Jesús Rodríguez18, Patricia Carreira19, Valle García20, José A. Pinto-Tasende21, Lluís Puig22, Elena Ricart23, Francisco Blanco24, Jordi Gratacós25, Ricardo Blanco26, Víctor Martínez Taboada26, Emilia Fernández27, Pablo Unamuno27, Isidoro González28, Fernando Gomollón García29, Raimon Sanmartí30, Ana Gutiérrez31, Àlex Olivé32, José Luís López Estebaranz33, Esther García-Planella34, Juan Carlos Torre-Alonso35, José Luis Andreu36, David Moreno Ramírez37, Benjamín Fernández38, Mª Ángeles Aguirre Zamorano39, Pablo de la Cueva40, Pilar Nos Mateu41, Paloma Vela42, Francisco Vanaclocha43, Héctor Coromines44, Santiago Muñoz45, Joan Miquel Nolla46, Enrique Herrera47, Carlos González48, José Luis Marenco de la Fuente49, Maribel Vera50, Alba Erra51, Daniel Roig52, Antonio Zea53, María Esteve Comas54, Carles Tomàs55, Pedro Zarco56, José María Pego57, Cristina Saro58, Antonio González59, Mercedes Freire60, Alicia García61, Elvira Díez62, Georgina Salvador63, César Díaz64, Simón Sánchez65, Alfredo Willisch Dominguez66, José Antonio Mosquera67, Julio Ramírez68, Esther Rodríguez Almaraz69, Núria Palau51, Raül Tortosa51, Mireia López51, Andrea Pluma51, Adrià Aterido51


This work was supported by the Spanish Ministry of Economy and Competitiveness grants (IPT-010000-2010-36, PSE-010000-2006-6, and PI12/01362).

Author information





AJ designed the study, conceived, designed and analyzed data and wrote the manuscript; SH performed data curation and statistical analysis; ED, JDC, CF, JT, JPG, AFN, ED, MBA, CP, RQ, FJLL, JLSC, JLM, MA, CM, JJPV, FM, SC and MLL contributed to patient recruitment, clinical data collection and analysis and manuscript revision; AA contributed to genetic data analysis; SM designed the study, coordinated clinical data collection and analysis, and co-wrote the manuscript.

Corresponding authors

Correspondence to Antonio Julià or Sara Marsal.

Ethics declarations

Conflict of interest

The authors declare no competing interests.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Members of the IMID Consortium are listed below Acknowledgements.

Supplementary information

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Julià, A., Martínez-Mateu, S.H., Domènech, E. et al. Food groups associated with immune-mediated inflammatory diseases: a Mendelian randomization and disease severity study. Eur J Clin Nutr 75, 1368–1382 (2021).

Download citation


Quick links