Food Insecurity, Diet, and Clinical Outcomes in Pediatric Inflammatory Bowel Disease: The Role of Unprocessed Foods

Abstract

Background and Aims

Food insecurity (FI) is associated with lower diet quality and higher intake of ultra-processed foods, which contribute to inflammatory bowel disease (IBD). The role of FI and dietary patterns in pediatric IBD remains understudied. We aimed to assess FI prevalence, dietary intake, and clinical outcomes in newly diagnosed pediatric IBD patients.

Methods

This single-center cohort included patients aged 5–17 years. Caregivers completed the United States Department of Agriculture Household Food Security Survey Module. Dietary intake was classified according to the NOVA classification. Clinical outcomes, including remission, healthcare utilization, and mucosal healing, were assessed over 12 months. Similarity Network Fusion was used to integrate dietary, demographic, and clinical data to identify patient subtypes, which were then compared for baseline characteristics and clinical outcomes.

Results

Among 120 patients, 14% reported FI. Food-insecure families lived in areas of greater social deprivation (0.27 vs 0.25, P < .001) and had public insurance (65% vs 18%, P < .001). Similarity Network Fusion identified 2 patient clusters primarily distinguished by unprocessed food intake, age, race, and insurance. Cluster one (younger, more racially diverse, and more public insurance) presented with more moderate/severe disease but achieved comparable normalization of labs and greater improvement in body mass index z-score (0.54 vs 0.27, P = .05) after controlling for biologic use. No differences were found in ultra-processed foods intake, FI, biologic utilization, clinical remission, mucosal healing, or healthcare utilization at 12 months.

Conclusion

Greater unprocessed food intake was linked to improved outcomes, highlighting the potential role of diet quality in disease management in pediatric IBD.

Graphical Abstract

Introduction

Diet plays a critical role in the pathogenesis of inflammatory bowel disease (IBD) by modifying the gut microbiota composition.¹ A Western diet rich in fats, refined carbohydrates, simple sugars, and processed foods is strongly associated with the progression of IBD, inducing microbiota dysbiosis, intestinal barrier dysfunction, and immune system activation.¹ Additionally, ultra-processed foods (UPFs) have been linked to a higher risk of developing IBD, specifically Crohn’s disease (CD), and a greater risk for IBD-related surgery.2, 3, 4

Dietary behaviors have been linked to food insecurity (FI), defined as “limited or uncertain availability of nutritionally adequate and safe foods or limited or uncertain ability to acquire acceptable foods in socially acceptable ways.”^5,6 FI has been associated with higher intake of UPFs.^7,8 Consumption of sugar-sweetened beverages, various types of UPFs, and overall lower diet quality have been linked to FI in individuals with chronic liver disease.^9,10 Similarly, adult patients with IBD who are food insecure have been shown to consume more UPFs than those who are food secure.¹¹ Furthermore, UPFs have been associated with lower food costs, suggesting that FI may partially mediate the relationship between financial status and greater UPF consumption.^12,13 Compromised food affordability and higher UPF consumption have been thought to contribute to poorer clinical outcomes in type 2 diabetes and higher rates of obesity.^14,15

Despite the well-described negative impact of FI on health and nutrition, there remains a paucity of studies evaluating FI and dietary patterns in IBD, specifically among children. Our primary aim was to determine the prevalence of FI in a population of children seen at our institution. Our secondary aim was to characterize patterns in dietary consumption, patient demographics, and social determinants of health (SDOH) and examine their association with clinical outcomes and healthcare utilization. Our hypothesis was that higher consumption of UPFs would be associated with worse clinical outcomes including lower rates of clinical remission and greater healthcare utilization, in newly diagnosed pediatric IBD patients.

Material and Methods

Study Design and Participants

This single-center cohort study included newly diagnosed IBD patients between 5 and 17 years of age, diagnosed at our medical center. Patients were eligible if diagnosed with IBD, either CD or ulcerative colitis (UC), between August 2022 and December 2023. The date of diagnosis was defined as the date of initial endoscopy. Each patient’s diagnosis was confirmed after manual review of the initial endoscopic, histologic, and imaging findings. Comprehensive patient-level demographic and clinical data were obtained from the electronic medical record at diagnosis and again 6 and 12 months after diagnosis. Clinical data included disease phenotype, laboratory values, and medication use. Patients were excluded if at diagnosis they were younger than 5 years or older than 17 years of age or diagnosed prior to August 1, 2022. The study was approved by our institutional review board.

All participating families were provided with state-specific resources to facilitate access to Supplemental Nutrition Assistance Program (SNAP) benefits and local food banks, as needed. Patients who screened positive for FI were additionally referred to the IBD program social worker for a comprehensive social needs assessment. Based on this assessment, the social worker determined which additional resources were best suited to address specific family needs.

Independent Variables

Our primary independent variable of interest was food security status. At the time of enrollment, parents were screened for FI using the United States Department of Agriculture Household Food Security Survey Module.⁶ According to the United States Department of Agriculture survey, the number of affirmative answers was totaled, and households were classified into 4 ordinal groups: high food security, marginal food security, low food security, and very low food security. For this study, high food security was classified as food secure, while those with marginal, low, or very low food security were classified as food insecure. Additionally, parents were asked if they received SNAP benefits.

Our secondary independent variable of interest was dietary patterns at the time of FI screening. Dietary consumption was collected for the 24 hours preceding the questionnaire. Diet recall information was obtained from the child; however, parents had to assist them in recalling information. Each food item consumed was classified as either a whole food item or a miscellaneous item. Whole food items included fruits, vegetables, unprocessed meat, poultry (eg, chicken, turkey), fish and seafood, legumes (lentils, beans, soybeans, peas, chickpeas, peanuts), or eggs. Miscellaneous food items (eg, cereal, fast food, bacon, and pizza) were classified by the degree of processing according to the NOVA classification.¹⁶ The NOVA classification was developed in 2010 to describe foods according to their degree of processing.¹⁶ Foods classified as NOVA group 1 are unprocessed foods, while NOVA group 3 contains processed foods, and group 4 contains UPFs. NOVA group 2 contains processed culinary ingredients which were not obtained from diet recall; thus, this information was not included in the analysis. Beverages including water, oral hydration replacement drinks like sports drinks and juices were excluded. Milk was one of the beverages included in the analysis, as it is often consumed as part of a meal (eg, dry breakfast cereal with added milk). Diet data from patients following the CD Exclusion Diet (CDED), partial enteral nutrition therapy and CDED, or exclusive enteral nutrition at the time of data collection were excluded. This excluded 5 patients from the Similarity Network Fusion (SNF) analysis (4 for dietary reasons listed above, 1 declined diet information). Diet histories were analyzed by 2 members of the research team, including a registered dietitian (H.B., N.Z.).

To ascertain neighborhood level SDOH, we identified the Social Deprivation Index (SDI) for each participant. The social determinants included in this value are measures of poverty, education, housing, transportation, and unemployment. Each patient’s residential address was entered into a geomarker assessment software that geocodes and links the addresses to census tracts in ways that are compliant with the Health Insurance Portability and Accountability Act.^17,18 Each census tract is then joined to a corresponding SDI score, which ranges from 0 to 1, with higher scores indicating greater deprivation.^17,18 Other SDOH included in the analysis included insurance type and receipt of SNAP benefits, a potential proxy for socioeconomic status.

Outcome Assessment

Clinical outcomes and healthcare utilization measures were further evaluated by dietary patterns based on NOVA food classifications. We evaluated corticosteroid-free clinical remission at 1 year, defined as a Physician Global Assessment (PGA) of quiescent disease and no corticosteroids for ≥4 weeks prior to the 12-month visit. Sustained remission was defined as corticosteroid-free remission at both the 6-month and 12-month visits. We evaluated mucosal healing, defined as a MAYO endoscopic score of 0 for UC patients or a Simplified Endoscopic Mucosal Assessment of 0 or 1 for inactive or minimal mucosal inflammation for CD patients.^19,20 For those patients who did not undergo endoscopic evaluation, we used a fecal calprotectin of <150 μg/g at the 12-month time frame as a surrogate marker of healing. Healthcare utilization was also collected through the first year of diagnosis and included the number of IBD-related clinic visits, emergency room visits, and hospitalizations. Lastly, IBD-related surgeries were gathered through the study period.

Additionally, for each patient, we collected demographic information from the patient medical record including age at diagnosis, sex, race, ethnicity, and insurance. We collected anthropometrics including height, weight and body mass index (BMI) z-score at diagnosis, 3-month, 6-month, and 12-month follow-up visits. Lastly, we collected serum inflammation markers including C-reactive protein (CRP), erythrocyte sedimentation rate (ESR), albumin, platelet count, hemoglobin, and hematocrit levels at the same time points.

Characterizing Inflammatory Bowel Disease Patient Subtypes

To characterize subtypes in our patient population based on a combination of diet, patient demographics, and SDOH, we undertook unsupervised clustering. To identify subtypes among our patients while integrating information from our differing data sources, we leveraged SNF, an unsupervised machine learning method that enables data from different sources to be integrated by fusing data-specific patient-patient similarity networks, using the SNFtool package v2.3.1.^21,22 From here, the eigengap heuristic was used to find the optimal number of clusters within the data by identifying the point at which the eigenvalues change the most from one eigenvalue to the next. Following identifications of the number of clusters, spectral clustering was applied to the SNF-integrated network to generate patient clusters. We defined a patient-patient similarity network for each input data set, here: diet, demographics, and SDOH. Diet information was based on NOVA classifications. Demographics included age of diagnosis, sex, race, and ethnicity. SDOH included the SDI, insurance, and SNAP benefits. We then fused the dataset-specific patient similarity networks together to create an integrated network, which was then used to discover clusters of patients, considering 2–5 clusters (Figure 1). Following derivation of the final patient clusters, association tests were performed to understand what variables were driving the cluster result. Then, while controlling for biologic use, the patient clusters were correlated with patient anthropometrics, FI, disease severity, medication utilization, and clinical outcomes to further characterize and understand the defined patient clusters.

Figure 1.

Similarity Network Fusion (SNF) input features. Baseline patient features used to form integrated clustering. Individual clusters are integrated into 2 distinct groups with a silhouette score of 0.33.

Regression Models

Normally distributed continuous variables were described as means with standard deviations, and non-normally distributed continuous variables were described using medians and interquartile ranges. Categorical variables were expressed as frequencies and proportions. We compared continuous variables with the independent sample t-test or Mann-Whitney U test. For noncontinuous measures, univariate X² or Fisher exact tests were undertaken to assess associations. Statistical significance was defined as a 2-tailed P value of <0.05. For healthcare utilization outcomes (count variables), we applied Poisson regression. Logistic regression was used to assess binary outcomes. Candidate covariates for regression modeling were selected based on clinical relevance and included NOVA group, age of diagnosis, sex, IBD type, baseline PGA, baseline CRP, and time on biologics. All analyses were performed using R version 4.3.2.

Results

Patient Characteristics

Our cohort consisted of 120 newly diagnosed patients with IBD, which included 86 (72%) patients diagnosed with CD and 34 (28%) with UC. The median age at diagnosis was 13.4 years (interquartile range (IQR): 10.8, 16.6), which included a cohort that was 57.5% male and 82.5% White. Overall, 17 (14%) families screened positive for FI.

Food Insecure Patients

Patients who were food insecure, compared to those classified as food secure, had a higher proportion of public insurance (65% vs 18%, P < .001) and SNAP benefits (41% vs 9%, respectively, P < .001) (Table 1). Families that were food insecure lived in areas of higher social deprivation (SDI 0.27 vs 0.25, respectively, P < .001, Supplementary Figure 1). There was no difference in clinical characteristics at diagnosis between the 2 groups, as defined by disease type, disease phenotype, PGA, and initial induction or maintenance medications. Despite these similarities, patients in the food-insecure group had higher median CRP (3.4 [IQR: 0.3, 8.6]; P = .047) and ESR (52 [36,87]; P = .007) and lower vitamin D levels at the time of diagnosis.

Table 1.

Demographics

	Food secure (n = 103)	Food insecure (n = 17)	P value
Age (y)	13.5 (10.8, 16.7)	13.2 (9.9, 14.3)	.15
Male	58 (56%)	11 (65%)	.60
Race			.098
White	88 (85%)	11 (65%)
Black	6 (6%)	3 (18%)
Other/Asian/mixed race, etc.	9 (9%)	3 (18%)
Social Deprivation Index	0.25 (0.20, 0.34)	0.27 (0.21, 0.36)	<.001
Ethnicity			.22
Hispanic	3 (3%)	2 (12%)
Non-Hispanic	99 (96%)	15 (88%)
Other	1 (1%)	0
Insurance type			<.001
Private	84 (82%)	6 (35%)
Public	19 (18%)	11 (65%)
Receives SNAP benefits	9 (9%)	7 (41%)	<.001
Method used for screening			.23
In person	40 (39%)	4 (24%)
Phone	63 (61%)	13 (76%)
Distance to grocery store (miles)	3.2 (1.8, 6)	5 (3.3, 9.7)	.96
IBD type
Crohn’s disease	73 (71%)	13 (76%)	.64
Ulcerative colitis	30 (29%)	4 (24%)
Disease location for Crohn’s disease patients
L1 (ileal)	18 (25%)	2 (8%)	.29
L2 (colonic)	14 (19%)	5 (38%)
L3 (ileocolonic)	41 (56%)	6 (46%)
Any L4ab/L4b	19 (26%)	4 (31%)	.77
Inflammatory disease phenotype	59 (81%)	10 (77%)	.75
Perianal disease	7 (10%)	1 (8%)	.83
Disease extent for ulcerative colitis patients
E1/E2 (distal)	7 (23%)	1 (25%)	.94
E3/E4 (extensive)	23 (77%)	3 (75%)
PGA score (clinical disease activity)
Mild/moderate	76 (74%)	10 (59%)	.24
Severe	27 (26%)	7 (54%)
Anthropometrics
BMI Z-score	−0.22 ± 1.30	−0.38 ± 1.86	.66
Weight Z-score	−0.06 ± 1.09	−0.19 ± 1.33	.67
Height Z-score	0.16 ± 0.92	0.12 ± 0.88	.25
Labs at diagnosis
Vitamin D	28.8 (23.1, 35.7)	22.8 (16.2, 28.5)	.023
Albumin (g/dL)	3.5 (3,4)	3.2 (2.8, 3.9)	.20
Hemoglobin (g/dL)	11.4 (10.2, 12.8)	11.1 (10.3, 11.6)	.21
HCT (%)	36.3 (33.4, 39.0)	35.2 (33.6, 35.8)	.16
Platelets (x103/mcL)	385 (306.5, 478.5)	397 (338, 518)	.38
CRP (mg/L)	1 (0.3, 3.2)	3.4 (0.3, 8.6)	.047
ESR (mm/hr)	25.5 (13.3, 53.3)	52 (36, 87)	.007
Induction therapy
5-ASA	5 (5%)	0	.21
Corticosteroids	48 (47%)	7 (41%)
Budesonide	28 (27%)	9 (53%)
Anti-TNF	20 (19%)	1 (6%)
EEN therapy	2 (2%)	0
Initial maintenance therapya
5-ASA	30 (29%)	4 (24%)	.81
Biologics (anti-TNFb)	72 (71%)	13 (76%)

Values expressed as proportions, median (interquartile range, 25th centile, 75th centile), or mean ± standard deviation.

Values in bold represent statistical significance.

5-ASA, 5-aminosalicylic acid; EEN, exclusive enteral nutrition; HCT, hematocrit; TNF, tumor necrosis factor.

^aDenotes one patient excluded from food-secure group that was on CDED as primary treatment.

^bAnti-TNF therapy includes infliximab and adalimumab.

As a group, food-insecure patients consumed a greater proportion of processed meat and fast-food items compared to food-secure patients (Supplementary Figure 2). Additionally, food-insecure patients consumed more animal protein and eggs compared to food-secure patients (Supplementary Table 1). There was no difference in respect to the consumption of processed meat or fast-food items in the univariate analysis. Among food-insecure patients, 4 (24%) reported consuming at least one fruit or vegetable in their dietary recall, compared to 34 (35%) of food-secure patients; this difference was not statistically significant (P = .37).

Integration of Diet, Demographics, and Social Determinants of Health Using Similarity Network Fusion Reveals 2 Distinct Subgroups

We undertook an exploratory analysis to evaluate dietary consumption patterns using SNF. Considering our overall study goal, we aimed to determine distinct patient subtypes that incorporated dietary patterns based on NOVA classifications. To target our approach, we included 3 different data domains of input features for each patient, which included demographics, SDOH (SDI, insurance, and SNAP benefits), and food items based on NOVA classification (groups 1, 3, 4). The eigengap heuristic, where higher values indicate the optimal number of clusters, yielded values of 0.03, 0.007, 0.005, and 0.004 for 2, 3, 4, and 5 cluster solutions, respectively. Based on these results, spectral clustering was performed to generate 2 clusters, which yielded a silhouette score of 0.33 (Figure 1). Cluster 1 included 67 patients, and cluster 2 included 48 patients.

Based on these input features, we found that clustering was being driven by consumption of unprocessed foods (NOVA 1 group) (Figure 2, Table 2). Patients in cluster 1 consumed more unprocessed foods (NOVA 1 group) compared to cluster 2 (median: 3.0 [2.0, 5.0] vs 2.0 [1.0, 3.0], P < .001, respectively). The range of unprocessed foods (NOVA 1 group) consumed for cluster 1 patients was between 0 and 12 food items over a 24-hour period (Table 2). Specifically, cluster 1 consumed a higher proportion of vegetables and fruits when compared to cluster 2 (Supplementary Figure 3). Fifty-three (79%) of the patients in cluster 1 consumed at least one fruit or vegetable compared to 26 (54%) in cluster 2 (P = .004). The amount of UPFs, or NOVA 4 foods, was similar between the cluster subtypes. Although not statistically significant, we found that based on specific food items consumed, cluster 2 patients, as a whole, ate a higher proportion of packaged snacks, fast-food items, and sandwiches, all of which are UPFs (NOVA 4 group) (Supplementary Figure 3). We further evaluated dietary patterns stratified by sex and found no significant differences in overall dietary intake patterns among NOVA groups (Supplementary Table 2).

Figure 2.

SNF input features that influence integrated clustering including demographic features like age at diagnosis and race. NOVA group 1 and insurance type are also influencing the development of the integrated clusters. Dots plotted above the red dashed line represent those with a significant P value (P < .05).

Table 2.

Input and Baseline Features Separated by Integrated Cluster Groups

	Cluster 1 (n = 67)	Cluster 2 (n = 48)	P value
Demographic features
Age at diagnosis	12.6 (9.5–15.1)	15.8 (12.5–17.0)	<.001
Sex: male	34 (51%)	32 (67%)	.13
Race			<.001
White	49 (73%)	48 (100%)
Black	8 (12%)	0
Other	10 (15%)	0
Ethnicity: Hispanic	4 (6%)	0	.14
SDOH features
SDI	0.22 (0.19–0.35)	0.29 (0.24–0.34)	.19
Insurance type: public	24 (36%)	4 (8%)	<.001
Receiving SNAP benefits	11 (16%)	3 (6%)	.15
Dietary features
NOVA group 1	3 (2–5) Range: 0–12	2 (1–3) Range: 0–7	<.001
NOVA group 3	1 (0–1) Range: 0–3	0 (0–1) Range: 0–3	.20
NOVA group 4	4 (2.5–5) Range: 0–11	4 (3–6) Range: 2–8	.14
Anthropometrics
BMI Z-score	−0.46 (1.4)	0.01 (1.4)	.07
Height Z-score	0.11 (0.79)	0.18 (1.1)	.71
Weight Z-score	−0.21 (1.07)	0.04 (1.23)	.27
Lab values at diagnosis
Albumin	3.3 (2.8–3.7)	3.9 (3.3–4.1)	.001
Hgb	11.1 (10.1–12.3)	12.0 (10.8–13.6)	.02
HCT	35.4 (32.1–37.4)	36.9 (33.6–41.1)	.03
Platelets	391 (326.5–509.5)	377 (289–454)	.14
CRP	1.9 (0.3–4.2)	0.78 (0.30–2.22)	.04
ESR	33 (18–71)	23.5 (11.75–47)	.13
Baseline disease characteristics
IBD type			.68
Crohn’s disease	49 (73%)	33 (69%)
Ulcerative colitis	18 (27%)	15 (31%)
Baseline PGA			.01
Mild	17 (25%)	24 (50%)
Moderate	31 (46%)	11 (23%)
Severe	19 (28%)	13 (27%)
Maintenance therapy			.04
5-ASA	14 (21%)	19 (40%)
Biologicsa	53 (79%)	29 (60%)

Values expressed as proportions, median (interquartile range), or mean ± standard deviation.

Values in bold represent statistical significance.

5-ASA, 5-aminosalicylic acid; HCT, hematocrit; Hgb, hemoglobin.

^aIf started on a biologic, all were on anti-TNF (anti-tumor necrosis factor) agents at diagnosis.

Cluster 1 Had Younger, More Racially Diverse Patients Than Cluster 2

Patients in cluster 1 were younger at diagnosis (12.6 [9.5, 15.1] vs 15.8 [12.5, 17], P = .0002) compared to those in cluster 2 (Table 2). Additionally, 12% of patients in cluster 1 self-identified as Black and 15% as either Asian, mixed race, or unspecified. By comparison, cluster 2 consisted entirely of patients who were identified as non-Hispanic, White. Patients in cluster 1 also had more public insurance compared to cluster 2 patients (36% vs 8%, P < .001, respectively). There were no other differences between patient demographics or SDOH between the 2 groups.

Cluster 1 Patients Had More Moderate-Severe Disease at Baseline

The cluster analysis observed differences in baseline IBD severity. At diagnosis, cluster 1 patients had lower albumin, lower hemoglobin, and higher CRP levels (Table 2, Figure 3, Supplementary Figures 4 and 5). A greater proportion of patients in cluster 1 had moderate-severe PGA at baseline compared to cluster 2 (75% vs 50%, respectively, P = .01). Additionally, patients in cluster 1 were more likely to have started on anti-tumor necrosis factor therapy for their maintenance medication compared to cluster 2 (79% vs 60%, P = .04). Rates of FI, baseline anthropometric measurements, and disease subtype demonstrated no difference across both groups (Table 2).

Figure 3.

Lab values through the study period, including the absolute change in labs. Alb, albumin; HCT, hematocrit; Hgb, hemoglobin; Plt, platelets.

Greater Improvement in Labs and Body Mass Index in Cluster 1 Subtype at 12 Months

During the 12-month follow-up period, we found no difference in clinical remission, sustained remission, mucosal healing, and healthcare utilization between the clusters (Table 3, Supplementary Figure 6). At 12 months, there was no difference in medications utilized for maintenance therapy. More specifically, there were no differences in rates of anti-tumor necrosis factor therapy used between the 2 groups by 12 months (Supplementary Table 3). Additionally, the number of patients exposed to a biologic was similar across clusters 1 and 2 (91% vs 79%, respectively, P = .07). Moreover, the time to first biologic use was comparable (cluster 1: 5.29 weeks [1.64, 12.2] vs cluster 2: 5.57 weeks [2.61, 12.5], respectively, P = .94). Despite more severe disease and the initial lab values at diagnosis, there was no difference between the proportion of patients with normal labs between the 2 clusters at 12 months (Supplementary Table 4). This included 89% of patients in cluster 1 had normal albumin compared to 96% in cluster 1 at the conclusion of the study (P = .21). Additionally, inflammatory markers were comparable between the 2 clusters, CRP (81% vs 74%, P 0.36) and ESR (57% vs 43%, P = .17) (Table 3, Supplementary Table 4).

Table 3.

Outcomes through 12 Months

	Cluster 1 (n = 67)	Cluster 2 (n = 48)	P value	P valuea
Change in anthropometrics
BMI	0.81 (0.01)	0.41 (0.74)	.030	.05
Height	−0.06 (0.4)	−0.20 (0.57)	.43	.14
Weight	0.57 (0.65)	0.36 (0.65)	.10	.19
Labs at 12 mo
Albumin	3.9 (3.6–4.1)	4.1 (3.8–4.3)	.018	.06
Hgb	12.6 (12.3–13.2)	13.3 (12.2–14.9)	.028	.001
HCT	38.1 (36.3–40.4)	39.5 (37–43.4)	.018	.002
Platelets	310 (262–339)	306.5 (260–359)	.65	.64
CRP	0.3 (0.20–0.40)	0.3 (0.3–0.55)	.16	.25
ESR	7 (3–20.5)	11.5 (4–25.5)	.38	.96
Healthcare utilization
Surgeries	0 (0–0)	0 (0–0)	.33	.08
Hospitalizations	0 (0–1)	0 (0–1)	.97	.45
ER visits	0 (0–1)	0 (0–1)	.67	.87
IBD clinic visits	4 (4–5)	4 (4–5)	.97	.74
Additional steroidsb	0 (0–1)	1 (0–1)	.054	.58
Clinical outcomes
Steroid-free remissionc	49 (74%)	28 (60%)	.11	.23
Sustained remission	23 (35%)	11 (23%)	.22	.48
Mucosal healingd	33 (57%)	27 (66%)	.41	.26
Biologicse
Time to first biologic (wk)	5.29 (1.64, 12.2)	5.57 (2.61, 12.5)	.94	-
Biologic exposure	61 (91%)	38 (79%)	.07	-

Values expressed as proportions, median (interquartile range), or mean ± standard deviation.

Values in bold represent statistical significance.

ER, emergency room; HCT, hematocrit; Hgb, hemoglobin.

^aIndicates analysis was replicated to control for biologic therapy.

^bNoted that this includes both corticosteroids and budesonide courses in addition to those used during the induction period.

^cDenotes some patients are missing due to loss to follow-up.

^dDenotes that some patients were not evaluate at 12 mo for mucosal healing at the time of data collection.

^eIncludes all biologics (infliximab, adalimumab, vedolizumab, ustekinumb) used during the study period.

At 12 months, patients in cluster 1 continued to have lower hemoglobin (12.6 [IQR: 12.3–13.2] vs 13.3 [IQR: 12.2–14.9], P = .28) and hematocrit values (38.1 [IQR: 36.3–40.4] vs 39.5 [IQR: 37–43.3], P = .02). After controlling for biologic use, these differences remained significant (Table 3). However, when evaluating the proportion of patients with hemoglobin and hematocrit within the normal range, these findings were comparable, hemoglobin (88% vs 91%, P = .52) and hematocrit (88% vs 89%, P = .79), respectively.

Lastly, patients in cluster 1 had greater improvement in BMI z-score from diagnosis to 12 months (Table 3, Supplementary Figure 7). After controlling for biologic use, change in BMI z-score continued to show a difference in further suggesting cluster 1 patients had greater improvement over the study period, outside of medication therapy alone (Table 3).

Unprocessed Foods (NOVA 1 Group) Were Not Associated With Clinical Outcomes at 12 Months

Based on findings from initial SNF and associations with NOVA 1 foods, multivariable regression was used to evaluate for associations with clinical outcomes. There was no association between FI and unprocessed foods (NOVA 1 group). In addition, we found no association between unprocessed foods (NOVA 1 group) and steroid-free remission, sustained remission, or healthcare utilization when controlling for age at diagnosis, sex, IBD type, baseline PGA, CRP at diagnosis, and time on biologic (Table 4).

Table 4.

Multivariable Regression Results for Impact of NOVA 1 Foods on Food Insecurity and Clinical Outcomes

Dependent variable	Marginal coefficient	Marginal logged estimate	Univariate marginal up-value	Multivariable coefficient	Multivariable logged estimate	Multivariable P value
Food insecurity	−0.002 (−0.23, 0.23)	1 (0.79, 1.26)	0.99	0.01 (−0.23, 0.25)	1.00 (0.8, 1.3)	.92
Steroid-free remission	−0.01 (−0.19, 0.16)	0.99 (0.83, 1.2)	0.89	0.03 (−0.16, 0.21)	1.03 (0.85, 1.20)	.79
Sustained remission	0.003 (−0.18, 0.18)	1 (0.84, 1.2)	0.97	0.09 (−0.13, 0.31)	1.1 (0.88, 1.4)	.42
ER visits	−0.04 (−0.17, 0.08)	0.96 (0.84, 1.09)	0.49	−0.03 (−0.16, 0.10)	0.97 (0.85, 1.1)	.61
Hospitalizations	−0.02 (−0.13, 0.09)	0.98 (0.88, 1.1)	0.73	0.02 (−0.1, 0.14)	1.02 (0.9, 1.1)	.76
Clinic visits	0.004 (−0.09, 0.10)	1 (0.91, 1.11)	0.94	0.03 (−0.07, 0.13)	1.03 (0.93, 1.1)	.61

Independent variable: NOVA 1 foods.

Adjusted covariates: age of diagnosis, sex, IBD type, baseline PGA, CRP at diagnosis, time on biologic.

ER, emergency room.

Discussion

This is the first study to report the prevalence of FI in a pediatric IBD cohort, with 14% of families screening positive. Using unsupervised clustering, we identified 2 patient subtypes, cluster 1 consisted of patients who consumed more unprocessed foods (NOVA 1 group), were younger in age, greater racial diversity and higher rates of public insurance. Despite presenting with more severe disease at diagnosis, and after adjusting for biologic use, this group demonstrated comparable normalization of inflammatory markers and greater improvement in BMI z-scores at 12 months compared to patients in cluster 2. No significant differences in clinical outcomes were observed based on food security status or unprocessed food consumption.

Prevalence of FI has varied across chronic gastrointestinal conditions, ranging from 5% up to 48%.^11,23 Few studies have reported the prevalence of FI among an IBD cohort.^11,23, 24, 25 Additionally, studies have demonstrated that FI was associated with lower diet quality and poor nutrition among patients with chronic gastrointestinal conditions.9, 10, 11 We observed that food-insecure patients were less likely to consume at least one fruit or vegetable—a finding that has been consistently reported in the literature.26, 27, 28, 29, 30, 31 Dietary intake is complex, often influenced by sociodemographic and cultural factors. For example, it has been shown Hispanic families who are food-insecure consume more fruits and vegetables over food insecure non-Hispanic White families.³² In the clustering, there was no association between FI and those patients that consumed more unprocessed foods. However, this group did have more racial diversity and public insurance, a potential surrogate marker for socioeconomic status and other SDOH. FI can affect dietary patterns differently by race and ethnicity.³² In our cohort, the cluster that consumed more unprocessed foods (NOVA 1) was younger. Older children consumed more fast food and packaged snacks, both NOVA group 4 foods, more often than younger children—a similar trend seen in children as they transition into secondary schools.33, 34, 35 Collectively, these findings highlight the multifaceted impact of sociodemographic and cultural factors on dietary intake.

Dietary modification has been associated with improved clinical outcomes in IBD. In a study by El Amrousy et al³⁶, children and adolescents with mild to moderately active IBD demonstrated improvements in clinical disease activity scores and inflammatory markers following 12 weeks on a Mediterranean diet, which is rich in unprocessed foods (NOVA 1). Similarly, both the CDED and Specific Carbohydrate Diet—diets that emphasize unprocessed foods—have been linked to improvement in clinical outcomes.^37,38 In our cohort, although patients’ intake of UPFs (NOVA 4 group) was comparable across groups, patients in cluster 1 with higher consumption of unprocessed foods (NOVA 1 group), which may have mitigated the negative effects of UPFs. Meyer et al² found a higher consumption of unprocessed NOVA 1 foods was associated with a lower risk of developing CD, suggesting a potential protective benefit. Moreover, higher intake of fiber-rich foods, such as fruits and vegetables, has been shown to modulate immune responses within the gastrointestinal tract, providing a plausible mechanistic link to the clinical benefits including improvement in serum inflammatory markers and BMI that was observed in our study.39, 40, 41 Emerging evidence further suggests that diets rich in minimally processed, fiber-containing foods may support intestinal barrier integrity and influence gastrointestinal motility, both of which are central to IBD pathophysiology.42, 43, 44 Alterations in epithelial permeability and dysregulated motility have been implicated in immune activation and disease progression.^43,44 Although we did not directly measure intestinal permeability or motility parameters in this cohort, these pathways represent plausible biologic mechanisms through which diet quality may influence inflammatory outcomes and warrant further investigation.

After adjusting for biologic exposure, we observed no difference in remission rates, mucosal healing, or healthcare utilization between patient subtypes. This may reflect the limited follow-up period or the complex interaction between diet and biologics. Data on combining biologics with whole food-based diets remain limited.^36,45,46 Similar to our findings, Keshteli et al⁴⁶ reported no difference in clinical symptoms among UC patients following an anti-inflammatory diet, though improvements in diet quality and fecal calprotectin levels were observed. Given biologics substantially improve clinical and endoscopic outcomes, subtle dietary effects may be more evident in subclinical markers and long-term outcomes.⁴⁷

From a clinical standpoint, these findings support routine assessment of dietary intake among pediatric IBD patients at diagnosis and throughout the disease course. IBD. Improvements in BMI trajectory alongside normalization of inflammatory markers may reflect meaningful nutritional rehabilitation, even without differences in remission rates, as restoration of growth remains a critical therapeutic goal.⁴⁸ While biologic therapy is the cornerstone of treatment, adjunctive dietary strategies may provide additional benefits. Incorporating structured dietary assessment and screening for FI and other SDOH into clinical care may help identify modifiable barriers and strengthen comprehensive, multidisciplinary IBD management.

We recognize there are several limitations to our study. It was conducted at a single center, which may limit the generalizability given regional variations. The screening was completed by interview, which may underestimate its prevalence compared to that of an anonymous survey. This lower prevalence may have limited statistical power to detect differences in clinical outcomes based on food security status. FI is dynamic and may have fluctuated during the study period, which we could not fully capture. The goal of this study was to determine relative dietary patterns; therefore, a single diet history was obtained, which may be seen as a limitation, and future studies should obtain more comprehensive dietary information. Additionally, physical activity and other lifestyle behaviors were not systematically collected and may represent unmeasured confounders influencing inflammatory markers and BMI trajectories. Finally, we did not assess direct measures of gastrointestinal motility or intestinal permeability, limiting our ability to evaluate mechanistic pathways linking dietary intake to gut barrier function and inflammation.

Future studies should leverage multicenter cohorts to improve generalizability and enhance representation of racially, ethnically, and socioeconomically diverse populations. Longitudinal dietary assessments throughout the disease course, particularly during the first year after diagnosis, would provide a more comprehensive understanding of intake patterns. Prospective interventional studies targeting diet quality are needed to evaluate how dietary strategies may complement biologic therapy. Additionally, incorporating standardized measures of physical activity and other health behaviors, as well as objective assessments of intestinal permeability, motility, and microbiome composition, will be critical to clarifying the mechanisms through which diet influences pediatric IBD outcomes.

Conclusion

After controlling for biologic use, greater consumption of unprocessed foods was associated with normalization of inflammatory markers, suggesting that diet may play a mechanistic role in modulating intestinal inflammation. Clinically, our findings suggest that emphasizing unprocessed food intake early in the disease course may support nutritional recovery and inflammatory control and should be considered as part of comprehensive, multidisciplinary pediatric IBD management. Interestingly, patients from more racially diverse backgrounds, despite not experiencing FI, were more likely to have public insurance, a proxy for broader SDOH. Contrary to expectations from the literature, this subgroup demonstrated higher diet quality, underscoring the complexity of dietary behaviors and influences. Further research is needed to delineate how these factors interact to shape dietary patterns and to evaluate how targeted nutritional interventions can complement medical therapy to improve outcomes in pediatric IBD.