Introduction

IBD, mainly including the subtypes Crohn’s disease (CD) and UC, is a globally rising immune-mediated disease characterised by relapsing inflammation in the GI tract. Although the reason for the increased incidence of IBD is unknown, changes in environmental factors, such as diet, may partially explain the recent increase in the incidence IBD.1

Studies in adult populations have suggested that a high intake of sugar,2 fat3 and red meat4 increases the risk of IBD. In contrast, high consumption of fruits,5 6 vegetables5 and fish,7 as well as high diet quality,8 are associated with reduced risk of IBD. Although diet in early life is critical for the development of the gut microbiome and gut immune tolerance,9 diet has primarily been assessed in adulthood and few studies have assessed childhood diet in IBD risk (online supplemental table 1). Retrospective data of adolescents’ diet suggest that high intake of vegetables and polyunsaturated fatty acids (PUFAs) may lower risk of IBD, whereas a high intake of sugary soft drinks increases the risk of IBD.10 The association between diet during the first 3 years of life and later IBD development has not been examined using prospective data, which may improve causal inference of results.

To our knowledge, this is the first study to prospectively investigate the association between early-life diet quality and intake frequency of specific food groups and later IBD risk by using two Scandinavian birth cohort studies.

Materials and methodsStudy population

We used data from the All Babies in Southeast Sweden (ABIS) Study and The Norwegian Mother, Father and Child Cohort Study (MoBa), which are parallel birth cohorts with large similarities in design and data characteristics (figure 1).11 12 Briefly, all 21 700 children born in Southeast Sweden from October 1997 to October 1999 were invited to participate in ABIS (79% participation rate). MoBa is a population-based pregnancy cohort conducted by the Norwegian Institute of Public Health. Pregnant women were recruited throughout Norway from 1999 to 2008 (41% participation rate). The cohort includes 114 500 children, 95 200 mothers and 75 200 fathers.13 We took advantage of ABIS questionnaires administered at birth, age 12 and 30–36 months and MoBa questionnaires administered during pregnancy, age 6, 18 and 36 months (online supplemental figure 1). The cohorts also contain individual-level data linked through personal identity numbers from the national health registers of Sweden14–16 and Norway.17 18 This study restricted participation to 81 280 and 65 692 children with any food data recorded at age 1 or 3 years (figure 1).

Figure 1

Flow chart of the study population in the All Babies in Southeast Sweden (ABIS) and The Norwegian Mother, Father and Child Cohort Study (MoBa). There were no available data on stillbirths/death <12 months of age in ABIS. PIN, personal identification number.

Early-life diet

Information about early-life diet was obtained from specific food questions included in comprehensive questionnaires administered at 12 and 30–36 months in ABIS and at 18 and 36 months in MoBa. The questionnaires contained information on the child’s upbringing and lifestyle habits, including the child’s food intake at the time of filling in the questionnaires. The food questions cover intake of meat, fish, fruits, vegetables, breast milk, porridge, baby foods, sweets and snacks, and beverages and have been used in several studies of diet-outcome associations.13 19–23 Due to lack of data, they have not been compared with any other dietary assessment tool. In both cohorts, parents reported the child’s current intake frequency of a standard portion of specified food items with four to seven response alternatives ranging from never to ≥4 times per day. Each questionnaire contained 40–50 food items, and we converted all data to weekly frequency intake. To reduce the risk of erroneously recorded data, the child’s food intake level was within each cohort modelled as a trichotomous exposure variable. A more detailed description is provided in the online supplemental file 1.

Because of the wording of the food questions and in line with previous studies,24 25 children with incomplete food data but information from at least one food group were categorised into the lowest frequency category. Among those 1 year old included in the analyses, 73% and 95% had complete data on food groups in ABIS and MoBa, respectively. Only 2% (ABIS) and <1% (MoBa) had data on less than half of the food groups. To reduce the influence from erroneously reported food intake frequencies, 1613 children (<0.1%) in ABIS and MoBa were excluded from at least one food group analysis (eg, intake of >88 portions of dairy per week was assumed to be implausibly high).

We examined two measures of the child’s diet at 1 and 3 years of age: diet quality and nine specific food groups.

Measure 1. Diet quality was examined using a modified version of the Health Eating Index (HEI) developed to specifically measure the child’s diet.26 The modified HEI reflects the child’s overall dietary quality, rather than food quantity and energy intake. This index included the intake of seven food groups: ‘fruits and vegetables’, ‘dairy foods’, ‘meat’, ‘fish and eggs’, ‘soft drinks’, ‘salty snacks’ and ‘sweet snacks’ (online supplemental tables 2 and 3).26 The intake of each food group was categorised by ranking weekly intake frequency by quartiles with a score of 1–4. Based on WHO dietary recommendations for children,27 being in the lowest intake category for ‘healthy food groups’ (eg, fruits and vegetables and fish and eggs) was assigned 1 point, the highest intake category was assigned 4 points, and vice versa for unhealthy foods, such as salty snacks and sweet snacks. Finally, the total HEI score, ranging from 7 to 28, with a higher score indicating a higher dietary quality, was divided into thirds representing low, medium and high diet quality.

Measure 2. We assessed the intake frequencies of the following food groups previously examined in relation to IBD10 28: meat, fish, dairy, fruits, vegetables, grains, potatoes, sugar-dense and fat-dense food, and sugar-sweetened beverages (SSBs) (online supplemental tables 3 and 4). We refer to food intake as the weekly intake frequency of each food group. All intake of food groups was divided into thirds representing low, medium and high food intake, except for the intake of SSBs at 1 year of age, which was dichotomised into no or some intake (≥0.5 serving/week).

Inflammatory bowel disease

We defined IBD as a minimum of two International Classification of Disease 10th Revision (ICD-10) codes for IBD in the Swedish National Patient Register15 (ABIS) and the Norwegian Patient Registry17 (MoBa) (online supplemental table 5). These registers contain nationwide data on inpatient and hospital-based outpatient care.17 29 Data on IBD were captured until 31 December 2020 in ABIS and 31 December 2021 in MoBa. We used subtype-specific ICD codes to define CD and UC. Cases with a mix of ICD codes for CD and UC during the last 5 years of study follow-up were defined as IBD-unclassified (IBD-U). While included in the outcome of any IBD, we a priori decided not to assess IBD-U as a separate outcome because of the limited number of cases and risk of misclassification. In Sweden, this register-based definition of IBD has a positive predictive value of 93.0% on medical record review.30 31

Other data

We used ABIS parental-reported questionnaire data administered at the child’s birth up to 12 months of age, MoBa questionnaire data reported at pregnancy week 15 up to 6 months of age and register-based data to retrieve information on the child’s sex, parental origin, parental education level, parental IBD, maternal comorbidities, delivery mode, maternal smoking during pregnancy, maternal age at delivery, birth weight, gestational age and full breastfeeding duration (online supplemental table 6). Data were captured at one time point, except for maternal smoking which in MoBa was captured throughout the pregnancy and the first 6 months after birth (online supplemental table 6). Based on previous literature,32 maternal immune-mediated comorbidity was considered a potential confounder and included type 1 diabetes (T1D (insulin-treated diabetes before or during pregnancy (MoBa) or T1D/insulin-treated diabetes (ABIS)), autoimmune thyroid disease or rheumatoid arthritis. Parental origin was defined as the mother’s native language (MoBa) or the parent’s country of birth (ABIS). Data on age at weaning, antibiotic use and formula intake were reported at age 12 (ABIS) and 18 months (MoBa). We also captured data on household income, defined by annual gross income.

Statistical analyses

Cox regression was used to estimate HRs and 95% CIs for IBD, CD and UC. Subanalyses for the outcome CD ignored events of other IBD subtypes and vice versa in UC-specific analyses. The proportional hazard assumption was tested using Schoenfeld residuals33 by graphically assessing the data and exploring interactions with time. The assumption was valid for all IBD analyses. The heterogeneity between the two cohorts was examined by using the Cochran Q test and I2 test. Due to low/moderate heterogeneity for all IBD analyses, pooled HRs were calculated using a random-effects model.34 Follow-up started at the child’s age of 1 and 3 years and ended at the time of first IBD diagnosis or censoring at the end of data capture (31 December 2020 in ABIS and 31 December 2021 in MoBa).

Our main analyses were adjusted for the child’s sex, parental IBD, origin, education level and maternal comorbidities (model 1). Model 2 was additionally adjusted for delivery mode, maternal smoking during pregnancy, maternal age at delivery, birth weight, gestational age and full breastfeeding duration. Preplanned subanalyses considered the risk of childhood-onset IBD diagnosed <18 years of age. We also performed sensitivity analyses excluding children with incomplete dietary data. Finally, we reran our analyses for the child’s diet quality at 1 year of age after excluding children diagnosed with IBD <6 years of age, which often constitutes a highly genetically determined subtype of IBD.35 Statistical analyses were performed using SPSS (V.29) and R Statistical Software (V.4.1.3 and 4.2.2), including the R packages survival, survminer, meta and metafor. We did not adjust for multiple comparison as all analyses shared an underlying hypothesis.36

Post hoc analyses

To reduce the risk of residual confounding, we additionally adjusted for household income level, intake of formula at 1 year of age and antibiotics exposure. We assessed the potential interaction between diet quality at 1 year of age and breast feeding, parental IBD, child’s sex and maternal education. To test the robustness of UC and CD-specific analyses, we reran our analyses after changing the definition of IBD-U to only include children with a mix of codes in the last 2 years of follow-up.

Patient and public involvement

No patients participated in the design of the study.

Results

We included a total of 81 280 children, 11 013 (48% girls) from ABIS and 70 267 (49% girls) from MoBa, with any dietary data at 1 year of age (figure 1). During 1 304 433 person-years of follow-up, 307 children were diagnosed with IBD (CD, n=131; UC, n=97; IBD-U, n=79), corresponding to an incidence rate of 32 per 100 000 person-years in ABIS and 22 per 100 000 person-years in MoBa (online supplemental table 7). Age at weaning was for most children at 4–6 months (table 1 and online supplemental table 8). The median follow-up time from 1 year of age was 21.3 years (ABIS) and 15.2 years (MoBa), and was related to the varying length of follow-up for these cohorts (online supplemental table 8). A total of 65 692 children remained with any food data recorded at 3 years of age. In MoBa, but not in ABIS, low maternal and paternal education levels were more common in children with low diet quality at 1 year of age (table 1). Maternal age, smoking status and incidence of IBD were similar in children with dietary data compared with all cohort participants (online supplemental table 9). The distribution of diet quality categories and food intake frequency partially changed between 1 and 3 years of age in particular with increased intake of SSBs and fruits observed in ABIS and MoBa, respectively (online supplemental table 10).

Table 1

Characteristics of children at 1 year of age in the ABIS and MoBa birth cohorts

Diet quality

Accounting for the child’s sex, parental IBD, origin, education level and maternal comorbidities (model 1), a high versus low diet quality at 1 year of age was associated with an adjusted HR (aHR) of 0.61 (95% CI=0.33 to 1.14) for IBD risk in ABIS and 0.79 (95% CI=0.57 to 1.10) in MoBa (online supplemental table 11). In pooled analyses, children 1 year old with medium or high diet quality (vs low diet quality) had a reduced risk of later IBD (pooled HR=0.74 (95% CI=0.57 to 0.97) and 0.73 (95% CI=0.55 to 0.97), respectively; figure 2). Estimates remained largely unchanged in model 1 (medium diet quality, pooled aHR=0.75 (95% CI=0.58 to 0.98); high diet quality, pooled aHR=0.75 (95% CI=0.56 to 1.00); figure 2) and further in model 2 adjusting for full breastfeeding duration and perinatal characteristics (online supplemental table 12). Cohort-specific estimates for CD and UC are reported in online supplemental table 13. The pooled analyses of high versus low diet quality at 1 year of age showed no association with later CD (figure 3) or UC (figure 4).

Figure 2

Pooled HRs of diet quality and food intake frequency at 1 year of age and risk of IBD. Adjusted HRs (aHRs) were adjusted for the child’s sex, parental IBD, origin, education level and maternal comorbidities.

Figure 3

Pooled HRs of diet quality and food intake frequency at 1 year of age and risk of Crohn’s disease (CD). Adjusted HRs (aHRs) were adjusted for the child’s sex, parental IBD, origin, education level and maternal comorbidities.

Figure 4

Pooled HRs of diet quality and food intake frequency at 1 year of age and risk of UC. Adjusted HRs (aHRs) were adjusted for the child’s sex, parental IBD, origin, education level and maternal comorbidities.

High versus low diet quality at age 3 years was not associated with later IBD in neither cohort-specific (online supplemental table 15) nor pooled analyses (pooled aHR=1.02 (95% CI=0.76 to 1.37); online supplemental figure 2). Likewise, diet quality at 3 years was not associated with CD or UC risk (online supplemental tables 16 and 17 (cohort-specific results); online supplemental figures 3 and 4) (pooled results)).

Food groups

The meta-analysis of estimates across the cohorts showed that children with high versus low fish intake at 1 year were at a reduced risk of later IBD (pooled HR=0.66 (95% CI=0.46 to 0.93); per category increase, pooled HR=0.82 (95% CI=0.70 to 0.97); figure 2). This estimate remained largely unchanged after adjustments in model 1 (pooled aHR=0.70 (0.49 to 1.00); figure 2) and model 2 (online supplemental table 12). Pooled analyses of high versus low fish intake at 1 year yielded an aHR of 0.67 (95% CI=0.39 to 1.17; figure 3) for CD and 0.46 for UC (95% CI=0.21 to 0.99; figure 4). Cohort-specific estimates of fish intake for children 3 years old and IBD, CD and UC risk are presented in online supplemental tables 15–17. The pooled aHR for IBD risk among children with high versus low fish intake at 3 years of age was 0.78 (95% CI=0.55 to 1.09; model 1; online supplemental figure 2). A high versus low fish intake at 3 years of age was associated with a reduced risk of UC (pooled aHR=0.46 (95% CI=0.24 to 0.90); model 1; online supplemental figure 4) but not CD (online supplemental figure 3).

The pooled analyses demonstrated that medium and high versus low vegetable intake at 1 year of age were associated with a reduced risk of IBD (medium, pooled HR 0.66 (95% CI=0.49 to 0.89); high, 0.72 (95% CI=0.55 to 0.95), respectively; figure 2), with similar estimates in model 1 (figure 2). Pooled analyses showed no association between vegetable intake at 3 years and later risk of IBD or its subtypes (online supplemental figures 2–4).

At age 1 year, 72% (n=58 730 of 81 280) reported intake of SSBs. Cohort-specific aHRs for IBD, CD and UC risk of some versus no intake of SSBs at 1 year of age are presented in online supplemental tables 11–13. The pooled aHRs showed that having some versus no intake of SSBs at 1 year of age was associated with an increased risk of later IBD (pooled aHR=1.42 (95% CI=1.05 to 1.90); model 1; figure 2). Pooled aHRs were 2.10 (95% CI=0.56 to 7.88; model 1) for CD and 0.92 (95% CI=0.25 to 3.26; model 1) for UC (figures 3 and 4). In contrast, cohort-specific and pooled analyses did not show any association between intake of SSBs at 3 years of age and later risk of IBD or its subtypes (online supplemental figures 2–4 and online supplemental tables 15–17).

Pooled analyses (figures 2–4 and online supplemental figures 2–4) and cohort-specific analyses (online supplemental tables 11, 13–17) showed no association between the other examined food groups, including meat, dairy, fruits, grains, potatoes and sugar-dense and fat-dense food, and risk of IBD, CD or UC.

Subanalyses

In line with our main results, childhood-onset IBD (<18 years) was inversely associated with diet quality at age 1 year, but not diet quality at age 3 years (online supplemental table 18).

Results were also largely unchanged in analyses excluding children with incomplete dietary data (online supplemental tables 19 and 20). Excluding children with very early-onset IBD diagnosis (<6 years, n=28) resulted in unchanged estimates (online supplemental table 21).

Post hoc analyses

Additionally, adjustment for (1) formula intake, (2) household income and (3) antibiotic use by age 1 year, all yielded essentially unchanged results (online supplemental tables 22–24). Also, there was no significant interaction between diet quality and IBD risk for any of the examined variables (online supplemental table 25). Changing the IBD-U definition from having a mix of codes in the last 5 years to the last 2 years of follow-up resulted in largely unchanged estimates (online supplemental table 26).

Discussion

In this pooled study of two Scandinavian birth cohorts, children with high diet quality at 1 year of age had a reduced risk of IBD compared with children with low diet quality. In addition, a high intake of fish and vegetables at age 1 year was associated with a reduced risk of IBD, whereas an intake of SSBs was associated with an increased risk of IBD. By contrast, at 3 years of age, only fish intake was associated with later IBD, particularly UC.

Although the gut microbiome is likely an important factor in the pathogenesis of IBD37 and early-life diet has a significant impact on gut microbiota composition,9 few studies have assessed diet in early life and risk of IBD. Our findings suggest that high diet quality, including a high intake of fish and vegetables at 1 year, may reduce the risk of later IBD development. In agreement with our results, a retrospective Italian study found that children aged 2–17 years with UC compared with healthy controls had significantly poorer adherence to a Mediterranean-style diet.28 This diet is characterised by a higher intake of fruits, vegetables and fish and a lower intake of sweets and fast food.28 Similarly, in a prospective cohort study of adults, high adherence to a modified Mediterranean diet was associated with a lower risk of CD, but not UC.38 A higher fish intake in children and adolescents has also been associated with a reduced risk of CD.39 The content of PUFAs or vitamin D in fish may be of special importance for IBD, as adult studies have observed an inverse association between a high intake of PUFAs3 and later UC and a high intake of vitamin D and CD.40

Vegetable consumption in childhood has been suggested to be associated with a reduced risk of IBD.10 It has been hypothesised that intake of vegetables and vegetable fibres may have programming effects on the immune system, reducing the risk of IBD.41 However, few studies have assessed the early-life intake of vegetables in the context of IBD, and because overall diet is multifactorial, there are several challenges to identifying the influence of a single diet component. Moreover, a type 2 error might be introduced for analyses of CD and UC, related to fewer events in those analyses and a higher risk of misclassification between subtypes.

In this study, any versus no intake of SSBs in children at 1 year of age was associated with an increased risk of IBD. Another paediatric study found that soft drink consumption four or more times a week in children <15 years was associated with an increased risk of CD, independent of socioeconomic characteristics.10 This is also supported by a systematic review that found soft drink consumption to increase the risk of UC in adults.42 Animal data suggest that dietary sugar reduces the diversity of the gut microbiome43 and negatively affects the mucosal immune response and barrier.44 Also, artificial additives have been suggested to cause dysbiosis and to induce chronic inflammation and dysfunctional immune response associated with IBD.45 However, since other studies have not found an association between SSBs and later IBD,46 future prospective studies are warranted to investigate the relationship between early-life intake of SSBs and subsequent IBD risk.

Dietary habits at 3 years of age were mostly not associated with later IBD risk in our two cohorts, suggesting that the influence of diet on IBD risk may be age dependent. The early-life gut microbiome undergoes significant changes until it converges to a stabilised, more adult microbiome after age 2 and 3 years.47 Since the gut microbiome seems to develop very early in life,47 48 diet at 1 year rather than at 3 years may have a stronger impact on the microbiome. Regrettably, the lack of microbiome data prevented the study of whether changes in microbial composition mediated the IBD risk related to early-life diet.

Strengths and limitations

This study is unique to prospectively assess childhood diet, limited to the first 3 years of life, and later risk of IBD. The prospectively collected data reduced the risk of reverse causation. Our study population of >80 000 children, including 307 IBD cases, with long-term follow-up, allowed a more precise estimation of risks than examining association in one cohort only. We found an association between child’s diet quality at 1 year of age and any IBD diagnosis, but not CD or UC diagnosis specifically. However, the smaller number of CD and UC events, as well as the challenges to clinically differentiate these subtypes, may have prevented us from finding any true associations (type 2 error), particularly in the subtype-specific analyses. Our population-based approach minimised selection bias. Diagnoses of IBD were defined based on a register-based algorithm, which previously has shown a positive predictive value of 93.0% on medical record review in Sweden,30 31 with a similar validity in Norway.49 To reduce the risk of misclassification between CD and UC, we defined IBD-U as cases with no clear distinction between CD and UC during the last 5 years of follow-up. However, even when changing the definition of IBD-U to a mix of codes during the last 2 years of follow-up, it did not show any difference in the results.

Data retrieved from repeated questionnaires and national registers allowed adjustment of potential confounders, including breastfeeding duration, formula intake, sociodemographics and antibiotic exposure. Still, we cannot exclude the possibility that unmeasured or residual confounding from other health behaviours may have influenced our results. Also, we did not have sufficient data about ultra-processed foods or food additives, other than SSBs and other ultra-processed foods contributing to low diet quality.

The food questions in ABIS and Moba have been used in several studies to assess a child’s dietary patterns,13 19–23 and as recommended when assessing the overall diet–disease relationship,50 we captured overall dietary quality rather than quantity and intake of energy and micronutrients. However, the prospective nature of this study ensures that any misclassification of dietary data should be unrelated to the risk of IBD and therefore not cause erroneous associations.

While the participation rate in ABIS was 79%, it was 41% in MoBa. Mothers in MoBa were older51 and more educated than all Norwegian mothers.52 Importantly, however, this self-selection has not been shown to affect exposure–outcome associations in the MoBa cohort.51 Because the present data originate from Sweden and Norway, two high-income countries, our findings may not be generalisable to low-income or middle-income countries with other dietary habits. Reported incidence rates of childhood-onset IBD vary significantly across Nordic countries, ranging from 7.0 to 15.0 per 100 000 person-years,31 and these differences may be explained by discrepancies of diagnostic method and type of data. We believe the difference in incidence rates of UC across ABIS and MoBa is explained by the distinct follow-up time. When accounting for varying lengths of follow-up across the cohorts, we found similar incidence rates of UC. Finally, as this study focused on early-life diet, our findings may not relate to dietary patterns later in life.