Mental and Somatic Conditions in Children With the Broad Avoidant Restrictive Food Intake Disorder Phenotype

Key Points

Question

What mental and somatic conditions are observed in children with the broad avoidant restrictive food intake disorder (ARFID) phenotype?

Findings

In this cohort study including 616 children with and 30 179 children without ARFID, children with ARFID had significantly increased risks of being diagnosed with neurodevelopmental, gastrointestinal, endocrine or metabolic, respiratory, neurological, and allergic conditions. They also had longer hospitalizations than children without ARFID.

Meaning

Risk increase in a broad range of coexisting conditions suggests complex patterns of health needs in children with ARFID, underscoring the critical importance of attention to ARFID across all pediatric specialties.

Abstract

Importance

Avoidant restrictive food intake disorder (ARFID) is a feeding and eating disorder characterized by limited variety and/or quantity of food intake impacting physical health and psychosocial functioning. Children with ARFID often present with diverse psychiatric and somatic symptoms and therefore consult various pediatric subspecialties. Large-scale studies mapping coexisting conditions are, however, lacking.

Objective

To characterize the health care needs of youth with ARFID.

Design, Setting, and Participants

This cohort study used the Child and Adolescent Twin Study in Sweden (CATSS), in combination with inpatient and specialized outpatient clinical diagnoses from the Swedish National Patient Register. Data were collected from July 2004 to April 2020, and data were analyzed from September 2022 to February 2024.

Exposure

Using a composite measure derived from parent or guardian reports and register data, children with the broad ARFID phenotype occurring between the ages of 6 to 12 years were identified, as well as children without ARFID.

Main Outcomes and Measures

From more than 1000 diagnostic International Classification of Diseases (ICD) codes, mental and somatic conditions within or across ICD chapters, the number of distinct per-person diagnoses, and inpatient treatment days between participants’ birth and 18th birthdays were specified (90 outcomes). Hazard ratios (HRs) and incidence rate ratios (IRRs) were calculated.

Results

Of 30 795 CATSS participants, a total of 616 children (2.0%) with the broad ARFID phenotype occurring between the ages of 6 to 12 years were identified, and 30 179 children without ARFID were identified. Of 616 children with ARFID, 241 children were female (39.1%). Relative risks of neurodevelopmental, gastrointestinal, endocrine or metabolic, respiratory, neurological, and allergic disorders were substantially increased in children with ARFID (eg, autism: HR, 9.7; 95% CI, 7.5-12.5; intellectual disability: HR, 10.3; 95% CI, 7.6-13.9; gastroesophageal reflux disease: HR, 6.7; 95% CI, 4.6-9.9; pituitary conditions: HR, 5.6; 95% CI, 2.7-11.3; chronic lower respiratory diseases: HR, 4.9; 95% CI, 2.4-10.1; and epilepsy: HR, 5.8; 95% CI, 4.1-8.2). ARFID was not associated with elevated risks of autoimmune illnesses and obsessive-compulsive disorder. Children with ARFID had significantly more distinct mental diagnoses (IRR, 4.7; 95% CI, 4.0-5.4) and longer hospital stays (IRR, 5.5; 95% CI, 1.7-17.6) compared with children without ARFID. Children with ARFID were diagnosed with a mental condition earlier than children without ARFID. No sex-specific differences emerged.

Conclusions and Relevance

This cohort study yields the broadest and most detailed evidence of coexisting mental and somatic conditions in the largest sample of children with ARFID to date. Findings suggest a complex pattern of health needs in youth with ARFID, underscoring the critical importance of attention to the illness across all pediatric specialties.

This cohort study among Swedish children with and without avoidant restrictive food intake disorder (ARFID) examines coexisting mental and somatic conditions among children with ARFID and characterizes their health care needs.

Introduction

Avoidant restrictive food intake disorder (ARFID) is a feeding and eating disorder with an estimated prevalence of 1% to 5%.^1,2 ARFID is characterized by severely limited food intake, impacting weight and growth, nutritional status, and psychosocial functioning.³ Unlike anorexia nervosa, ARFID is typically not driven by body shape or weight concerns, but rather by sensory aversion to smell, taste, or texture of food; low interest in food; and/or fear of aversive reactions to food intake like choking or vomiting. Commonly emerging in childhood,² ARFID is often persistent and disrupts daily life.⁴

Identification and treatment of ARFID are complicated by frequently co-occurring mental and somatic health issues, leading to complex clinical presentations, missed diagnosis, and delayed treatment.^5,6 Commonly reported among youth with ARFID are elevated depressive (7%-33%) and anxiety (9%-72%) symptoms^7,8; gastrointestinal disorders (19%-44%; eg, disorders of gut-brain interaction, esophagitis, acid reflux, celiac disease, and inflammatory bowel disease [IBD])^5,7,9,10,11; endocrine alterations (eg, thyroid and appetite-regulating hormone alterations)^9,12; and immunological conditions (eg, asthma and food or drug allergies).⁹ ARFID also frequently co-occurs with neurodevelopmental conditions (NDCs), particularly autism (8%-55% in ARFID) and attention-deficit/hyperactivity disorder (ADHD; 3%-39% in ARFID).^7,13

Research on ARFID-related comorbidities is lacking compared with other eating disorders.¹⁴ Most evidence comes from case reports and single-site studies conducted in relatively small and/or uncontrolled clinical samples.^5,6,8,9,15 Reported health problems stem from different sources of varying data quality, including parent-reported or self-reported medical history besides formal clinical diagnoses. Furthermore, the effects of age and sex on ARFID-related conditions, which could have important treatment implications, remain largely unknown.¹⁵ A deeper understanding of comorbidity in ARFID could elucidate the disorder’s etiology and inform diagnostic and therapeutic approaches.

In a previous study,¹ a composite measure was developed for the broad ARFID phenotype using parent-reported and register data in a sample of more than 30 000 Swedish twins. In this largest community sample of children with ARFID to date (n = 616 children; Figure 1), the current study aimed to systematically assess relative and absolute risks of the (so far) broadest range of mental and somatic diagnoses in children with ARFID from their birth to 18th birthday. Based on small-scale clinical evidence,^7,9,14 we hypothesized that children with ARFID would show increased risks compared with children without ARFID (henceforth referred to as controls) for most analyzed conditions, particularly anxiety, neurodevelopmental, and gastrointestinal disorders. For the first time, the effects of sex assigned at birth and age on these outcomes were also explored. Moreover, ARFID-related treatment needs were evaluated through number of per-person diagnoses and duration of inpatient stays, offering novel insights that underscore the pediatric clinical relevance of ARFID.

Figure 1. Avoidant Restrictive Food Intake Disorder (ARFID) Phenotype Definition.

ARFID phenotype definition is based on Diagnostic and Statistical Manual of Mental Disorders (Fifth Edition) (DSM-5) criteria. Available diagnoses, procedures, and prescribed drugs between twins aged 6 to 12 years were extracted from the National Patient Register (contains all diagnostic and procedure codes from inpatient care since 1987 and approximately 80% of codes from specialized outpatient care since 2001) and the Prescribed Drug Register. Elements designated as (P) were parent-reported at twin age 9 or 12 years and extracted from the Child and Adolescent Twin Study in Sweden. The goal of this study was to broadly map co-occurring conditions in ARFID. In epidemiological contexts, it is difficult to evaluate whether specific co-occurring conditions are cause, comorbidity, or consequence of avoidant or restrictive eating. Thus, we did not exclude ARFID based on co-occurring mental or somatic conditions that could potentially explain avoidant or restrictive eating (DSM-5 ARFID criterion D).³ Numbers of individuals who met the specified diagnostic criteria or elements at each step of the ARFID phenotype definition are stated, including overall prevalence of the ARFID phenotype out of all individuals in the study and proportion of girls among individuals with the ARFID phenotype. Please consult eFigure 1 in Supplement 1 for more details, as well as a detailed statistical analysis plan. BMI indicates body mass index; crit, DSM-5 criterion; Dx, diagnosis; EDNOS, eating disorder not otherwise specified; HR, hazard ratio, ICD-9/ICD-10, International Statistical Classification of Diseases and Related Health Problems Ninth or Tenth Revision; IRR, incidence rate ratio; Rx, prescription; Tx, treatment or procedure.

Methods

Study Design and Participants

The study was approved by the regional ethical review board in Stockholm, Sweden (Dnr 03-672, 2010/597-31/1, and 2010/322-31/2). Informed verbal consent was obtained from parents. Relevant data were retrieved from the Child and Adolescent Twin Study in Sweden (CATSS), which addressed all twins born in Sweden since July 1992.¹⁶ This cohort study included 30 795 individual twins born between 1992 and 2008 (eTable 1 in Supplement 1; response rate approximately 69%). This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.

Definition of ARFID and Coexisting Conditions

CATSS is linked to Swedish health registers, including the National Patient Register (NPR, available until December 2016, which used International Classification of Diseases, Ninth Revision [ICD-9] coding from 1987 to 1996 and ICD-10 coding since 1997)^17,18 and the Prescribed Drug Register (PDR, available until December 2017).¹⁹ In accordance with our previous study,¹ a broad ARFID phenotype (henceforth referred to as ARFID) was identified in all participants via a composite measure using CATSS parent reports at twin ages 9 or 12 years with NPR diagnostic or treatment codes and PDR codes between twin ages 6 and 12 years (Figure 1; eFigure 1 in Supplement 1). Sensory-based avoidance was an integral component of this study’s ARFID definition (criterion A0 in Figure 1).

NPR codes (composed of 1000 individual 3-digit or 4-digit ICD-9 or ICD-10 codes) between 1992 and 2016 covering inpatient and specialized outpatient diagnoses were used to define the mental and somatic conditions analyzed as study outcomes. ARFID phenotype–defining codes (eg, for feeding difficulties or loss of appetite) were not analyzed as outcomes (Figure 1; eFigure 1 in Supplement 1). Relevant conditions were identified via selection and grouping of available ICD codes according to clinical context and symptom overlap, previous ARFID literature, and prevalence in our sample. For a condition to be analyzed as an outcome, at least 5 affected individuals were required per group to maintain privacy. Outcome definition followed a 3-level approach: (1) ICD chapters (1-digit code level; eg, endocrine or metabolic disorders [chapter E]; 7 outcomes); (2) individual and grouped conditions within ICD chapters (3-digit or 4-digit code level; eg, diabetes and thyroid conditions [within chapter E]; 75 outcomes); and (3) grouped allergic and autoimmune conditions across ICD chapters (2 outcomes; see eTable 2 in Supplement 1 for ICD codes for all outcomes).

Further study outcomes included number of all distinct diagnoses (ie, unique ICD codes) per person (plus number of distinct mental and somatic diagnoses separately) and number of inpatient days per person due to any diagnosis (plus number of inpatient days due to any mental and any somatic diagnosis separately, 6 outcomes; eFigure 1 in Supplement 1). Only conditions occurring between birth and a participant’s 18th birthday, prior to potential death, emigration, or end of follow-up (in 2016) qualified as outcomes (mean [SD] participant age at end of follow-up, 15.35 [3.15] years).

Statistical Analysis

Statistical analyses were performed in R version 4.2.2 (R Foundation).²⁰ All tests were 2-tailed; statistical significance was indicated by the Benjamini-Hochberg false discovery rate (FDR)²¹–adjusted threshold of α_FDR = .0343 (correction across all 90 main outcomes). Correlations among twin pairs were statistically addressed by using cluster-robust standard errors and confidence intervals. All models accounted for exposure or follow-up time and potential censoring.

eFigure 1 in Supplement 1 provides an overview of all analyses. Lifetime prevalence of outcomes in ARFID and controls was computed as number of individuals diagnosed divided by total number of individuals, per sex (female and male), and per age group (birth to <6 years [early childhood]; 6 to <12 years [middle/late childhood]; and 12 to <18 years [adolescence]). For main analysis 1 (eFigure 1 in Supplement 1), time-to-event Cox proportional hazards regression models were performed to assess relative risks of the 84 analyzed conditions in ARFID vs controls via hazard ratios (HRs) and cluster-robust 95% confidence intervals. Consult Figure 2 for model specifications. Supplementary sex-stratified and age group–stratified Cox regression models were implemented to compare sex-specific and age-specific HRs in ARFID vs controls.

Figure 2. Prevalence and Relative Risks of Coexisting Conditions in Children and Adolescents With Avoidant Restrictive Food Intake Disorder (ARFID) vs Controls (Part 1).

Bar plots for prevalence of each analyzed condition (No. diagnosed/total No., %) in ARFID and controls (y-axis logarithmically scaled). Conditions are sorted by International Classification of Diseases (ICD) chapter. Within each ICD chapter, conditions are sorted in descending order of prevalence in controls. Hazard ratios (HRs) in ARFID vs controls for each analyzed condition are presented above respective bars. HRs were obtained with the base Cox regression model: twin age was the underlying time scale (birth to 18th birthday); first diagnosis of the analyzed condition constituted the event; ARFID phenotype (ARFID vs controls) served as the time-invariant predictor (ie, presence of ARFID was modeled as a consistent exposure from birth to 18th birthday); adjustment for sex assigned at birth (female or male) and birth year (range: 1992-2008, factorized) as categorical covariates; robust sandwich estimates given clustered twin data (see eFigure 2 in Supplement 1 for detailed statistics). Please note that some labels were collapsed to indicate conditions in a space-efficient manner (consult eTable 2 in Supplement 1 for details). ADHD indicates attention-deficit/hyperactivity disorder; GERD, gastroesophageal reflux disorder; IBD, inflammatory bowel disease; IBS, irritable bowel syndrome.

^aIndicates conditions for which HRs were significantly different (ie, increased) in ARFID compared with controls according to the base Cox regression model at the false discovery rate-adjusted threshold α_FDR = .0343.

As descriptive follow-up analysis, absolute risk was calculated via cumulative incidence of chapter-level (except perinatal) and grouped allergic or autoimmune conditions in ARFID vs controls from birth to participants’ 18th birthday (see eFigure 4 in Supplement 1 for individual conditions). Kaplan-Meier estimation was used after matching 10 unexposed children to each child with ARFID based on sex and birth year. Thus, we obtained the proportion of children diagnosed with any analyzed condition before a specific age while accounting for censoring and covariates.

To estimate treatment needs in ARFID, incidence rate ratios (IRRs) and cluster-robust 95% confidence intervals were computed for the count of all distinct, all distinct mental, and all distinct somatic ICD diagnoses per person in ARFID vs controls (main analysis 2, eFigure 1 in Supplement 1). Likewise, IRRs were computed for the count of inpatient days due to any, any primary mental, and any primary somatic ICD diagnoses in ARFID vs controls (main analysis 3). Poisson regression models were used, adjusted for sex and categorical birth year and including an offset term for exposure time. Supplementary analyses examined potential interaction effects of group (ARFID vs controls) and sex on number of distinct diagnoses and inpatient days.

Results

Sample Characteristics and Outcome Prevalence

Among children and adolescents meeting ARFID criteria (n = 616 [2.00% of the total sample]), 241 participants (31.9%) were female and 375 participants (60.88%) were male (eTable 1 in Supplement 1). Within the ARFID group, 87.18% of participants had 1 or more registered diagnoses compared with 73.03% within controls. In the ARFID group, the prevalence of any diagnosis within an ICD chapter was (in descending order) respiratory (53.08%), mental (35.88%), perinatal (34.42%), digestive (34.09%), congenital (22.40%), endocrine or metabolic (17.69%), and circulatory conditions (4.06%; Figures 2, 3; eTable 3 in Supplement 1). The most common mental conditions in the ARFID group were ADHD (17.5%), autism (13.8%), and intellectual disability (9.1%). The most common somatic conditions in the ARFID group were acute upper respiratory infections (30.36%), fetal growth retardation (28.73%), allergic conditions (27.06%), asthma (20.78%), acute lower respiratory infections (18.18%), chronic upper respiratory disorders (15.26%), neonatal jaundice (14.77%), and constipation (14.12%; Figures 2, 3).

Figure 3. Prevalence and Relative Risks of Coexisting Conditions in Children and Adolescents With Avoidant Restrictive Food Intake Disorder (ARFID) vs Controls (Part 2).

Bar plots for prevalence of each analyzed condition (No. diagnosed/total No., %) in ARFID and controls (y-axis logarithmically scaled). Conditions are sorted by International Classification of Diseases (ICD) chapter, the group of further conditions, and grouped allergic and autoimmune conditions. Within each ICD chapter, conditions are sorted in descending order of prevalence in controls. Allergic conditions: allergic rhinitis, asthma, allergic gastroenteritis, allergic contact dermatitis, urticaria; autoimmune conditions: autoimmune thyroiditis, diabetes type I, Crohn disease, ulcerous colitis, bullous skin conditions (including pemphigus or pemphigoid), atopic dermatitis, psoriasis, and juvenile arthritis (see eTable 2 in Supplement 1 for diagnostic codes). Hazard ratios (HRs) in ARFID vs controls for each analyzed condition are presented above respective bars. See Figure 2 caption for details regarding how HRs were obtained. CNS indicates central nervous system.

^aIndicates conditions for which HRs were significantly different (ie, increased) in ARFID compared with controls according to the base Cox regression model at the false discovery rate-adjusted threshold α_FDR = .0343.

Relative and Absolute Risks

As anticipated, relative risks for most conditions were significantly increased in ARFID compared with controls (Figures 2, 3; eTable 4, eFigure 2 in Supplement 1). Cox regressions yielded HRs greater than 1 in ARFID vs controls for all 7 analyzed ICD chapters, with highest relative risks for mental conditions (HR, 3.90; 95% CI, 3.36-4.52), endocrine or metabolic conditions (HR, 2.73; 95% CI, 2.21-3.36), and digestive conditions (HR, 2.03; 95% CI, 1.76-2.35). ARFID was associated with significantly elevated risk for 55 of 75 conditions (73.33%) within ICD chapters. The highest HRs emerged for mixed developmental disorders, nervous system malformations, dysphagia, intellectual disability, autism, cancer, motor development impairment, gastroesophageal reflux disease (GERD), cerebral palsy, epilepsy, upper gastrointestinal conditions, ADHD, and tic disorders. Conversely, irritable bowel syndrome (IBS), obsessive-compulsive disorder (OCD), diabetes, and obesity were not significantly associated with ARFID (Figures 2, 3; eFigure 2 in Supplement 1). While ARFID correlated with higher risk of allergic conditions, no significant risk increase was observed for autoimmune conditions (Figure 3; eFigure 2 in Supplement 1).

Inverted Kaplan-Meier curves in Figure 4 and eFigure 3 in Supplement 1 illustrate a steeper increase in cumulative incidence for ARFID vs controls for all ICD chapter–level conditions, particularly for mental, respiratory, endocrine, digestive, and grouped allergic conditions. Regarding within-chapter conditions, this absolute risk divergence was most notable for ADHD, autism, intellectual disability, asthma, GERD, and epilepsy (eFigure 4 in Supplement 1). However, Kaplan-Meier curves overlapped for grouped autoimmune conditions and some within–ICD chapter conditions like diabetes, OCD, IBS, and febrile seizures, indicating comparable absolute risks between the ARFID and control groups (Figure 4; eTable 5, eFigures 3-4 in Supplement 1).

Figure 4. Cumulative Incidence Plots for Coexisting Conditions in Children and Adolescents With Avoidant Restrictive Food Intake Disorder (ARFID) vs Controls.

Absolute risk via cumulative incidence (%) of having a specific condition (International Classification of Diseases [ICD] chapters and grouped allergic and autoimmune conditions) and lower and upper limits of its cluster-robust 95% confidence intervals are plotted over age (0 to <18 years) in ARFID vs controls (see eFigure 3 in Supplement 1 for absolute risk differences in ARFID vs controls and eFigure 4 in Supplement 1 for cumulative incidence plots of individual conditions within ICD chapters). Perinatal conditions were excluded given their occurrence during the perinatal period and lack of age-related dynamics. Cumulative incidence was estimated using the Kaplan-Meier method (1 − Kaplan-Meier survival estimate, accounting for censoring) after creating a modified analysis sample by matching 10 unexposed individuals to each individual with ARFID, stratified by sex and birth year. Descriptions for selected conditions with notable curve characteristics: mental conditions, steepest increase in ARFID vs controls between 7 and 11 years of age; respiratory conditions, steepest increase in ARFID vs controls from birth to 4 years of age with subsequent flattening; endocrine, digestive, and allergic conditions, more steadily diverging curves with increasing age in ARFID vs controls.

Number of Distinct Diagnoses and Inpatient Treatment Days

Children with ARFID had significantly more mean (SE) distinct diagnoses per person (6.27 [0.28]) compared with controls (2.78 [0.02]; eTable 6, eFigure 5A in Supplement 1). In particular, the IRR for distinct mental diagnoses per person was nearly 5-fold higher in the ARFID group than controls (IRR, 4.65; 95% CI, 3.99-5.42), while the IRR for distinct somatic diagnoses was nearly twice as high in the ARFID group (IRR, 1.98; 95% CI, 1.79-2.19; Figure 5A).

Figure 5. Number of Distinct International Classification of Diseases (ICD) Diagnoses in Avoidant Restrictive Food Intake Disorder (ARFID) vs Controls (A) and Duration of Hospitalization or Inpatient Treatment in ARFID vs Controls (B).

Poisson regression models (predictor: group [ARFID vs controls]; covariates: sex [female or male], birth year [1992-2008, factorized]; robust sandwich estimates given clustered twin data; offset term to account for exposure time) were applied to estimate incidence rate ratios (IRRs) in ARFID vs controls, their cluster-robust 95% confidence intervals, and P values. Please consult eFigure 5 in Supplement 1 for histograms. Dotted line indicates IRR of 1.

^aIndicates significantly different (ie, increased) incidence rates in ARFID compared with controls at the false discovery rate-adjusted threshold α_FDR = .0343.

ARFID was likewise associated with significantly more mean (SE) inpatient days (26.61 [2.73]) relative to controls (9.18 [0.13]; IRR, 2.88; 95% CI, 2.34-3.55; Figure 5B; eTable 7, eFigure 5B in Supplement 1). Although numerically most inpatient days were due to somatic diagnoses, the IRR in ARFID vs controls was substantially higher for mental health–related hospitalizations (IRR, 5.50; 95% CI, 1.72-17.60; Figure 5B).

Roles of Sex and Age

Sex-stratified Cox regressions revealed no significant sex differences in relative risk in the ARFID vs control groups for any analyzed condition after adjusting for multiple testing. Prior to FDR adjustment, however, female children and adolescents with ARFID tended to have higher relative risk than male children and adolescents with ARFID for short stature, motor development impairment, congenital conditions, and chronic upper respiratory conditions, whereas male children and adolescents with ARFID tended to have higher relative risk than female children and adolescents for anxiety disorders (eTable 8 in Supplement 1). No significant sex effects emerged for number of distinct diagnoses (eTable 9 in Supplement 1) or inpatient days (eTable 10 in Supplement 1) in ARFID vs controls.

Unlike in the control group, ARFID was linked to an earlier diagnosis of mental conditions at 6 to less than 12 years compared with 12 to less than 18 years (eTable 11, eFigure 6 in Supplement 1). Prior to FDR adjustment, autism, ADHD, and endocrine or metabolic conditions tended to be diagnosed earlier in ARFID than controls, whereas some dermatological conditions (eg, dermatitis, eczema) tended to occur later in ARFID than controls (eTable 11 in Supplement 1).

Discussion

In this cohort study of Swedish children and adolescents, an elevated risk was found of coexisting somatic and particularly mental conditions in individuals with vs without the broad ARFID phenotype. Relative risks of all ICD chapter–level conditions, grouped allergic conditions, and most individually analyzed conditions were significantly increased in ARFID vs controls. For instance, individuals with ARFID had a 10-fold higher risk of autism, intellectual disability, and dysphagia and a 6-fold higher risk of cerebral palsy and epilepsy. Furthermore, treatment needs, indicated by number of distinct per-person diagnoses and hospitalization duration, were substantially increased in ARFID vs controls, with mental disorders driving group differences.

Overall, approximately 36% of children with ARFID had a diagnosed mental condition (vs approximately 11% of controls). Whereas previous studies focused mostly on autism and ADHD,^7,8,22 we observed a particularly strong risk increase in ARFID vs controls for a variety of NDCs, including intellectual disability; mixed specific, other, or unspecified developmental disorders; tic disorders; and motor development disorder. Nonetheless, the predominant mental conditions in ARFID were ADHD (17.5%), autism (13.8%), and intellectual disability (9.1%) in this study. This supports findings from smaller-scale clinical studies where these NDCs occurred in children with ARFID with even higher prevalence.^13,22 The relatively lower NDC prevalence in the ARFID group here may be explained by the use of registered clinical diagnoses (as opposed to, for example, self-reports) and this study's general population sample with overall less symptom severity.

Regarding somatic conditions, the highest risk increases in ARFID vs controls were found for gastrointestinal disorders (eg, GERD, upper gastrointestinal disorders, dysphagia); neurological disorders (eg, cerebral palsy, epilepsy); cancer; and nervous system malformations. The co-occurrence of ARFID and gastrointestinal disorders has repeatedly been highlighted.^11,23 Digestive symptoms, sensory aversion, fear of aversive gastrointestinal consequences, and food avoidance can reinforce and perpetuate each other, thereby contributing to diagnostic overshadowing.

Children with ARFID had higher risk of respiratory conditions. This is in line with a small study showing elevated prevalence of asthma⁹ and case reports of respiratory complications in ARFID (eg, pneumothorax, nontuberculosis mycobacteria infections),²⁴ as well as our clinical experience (eg, reduced immune function in ARFID). Data on respiratory conditions in ARFID are, however, extremely limited. Risk increase of combined allergic conditions was also observed in ARFID—for example, respiratory, gastrointestinal, and dermatological allergic conditions, potentially contributing to restrictive eating. The available level of detail of NPR diagnostic codes did not allow us to specify food allergies, which are common in ARFID.⁹

Against expectations, we did not find a significantly increased risk of OCD in children with ARFID, which might be attributed to underreporting and a portion of affected families being treated in primary care (ie, not visible in the NPR). In contrast, preliminary data in the same sample indicate highly increased risk of parent-reported OCD symptoms in ARFID.²⁵ Correspondingly, a Dutch study using similar procedures to identify ARFID in a community sample reported increased OCD symptoms.²⁶

In contrast to previous epidemiological evidence showing strong bidirectional associations between autoimmune conditions and eating disorders,²⁷ an association was not found between ARFID and grouped autoimmune conditions in this study. This may be due to our sample’s demographics. For example, this study’s upper age limit was 18 years, while the median age at diagnosis is 24 years for type I diabetes²⁸ and 20 to 30 years for psoriasis.²⁹ However, we did find a significantly increased risk for specific autoimmune conditions in ARFID, such as noninfective IBD and celiac disease. Importantly, given a prevalence of less than 5 individuals among children with ARFID (eTable 2 in Supplement 1), we could not examine associations between ARFID and autoimmune pediatric acute-onset neuropsychiatric syndrome following group-A-streptococcus infection (PANS/PANDAS), although evidence of the interplay of PANS/PANDAS, OCD, and ARFID exists.³⁰

As opposed to controls, children with ARFID were more likely to be diagnosed with a mental disorder earlier (ie, by late childhood [6 to <12 years]). In fact, cumulative incidence curves showed the steepest increase in mental diagnoses in ARFID between 7 and 11 years. This aligns with the peak age at first NDC diagnosis in the literature.³¹ Sex-specific relative risk differences between ARFID and controls did not survive multiple testing correction. To our knowledge, this is the first study on links between sex and comorbidity in ARFID, although sex differences in NDC comorbidity were found previously in children diagnosed with a feeding and eating disorder before age 3—a phenotype similar to ARFID.³²

Strengths and Limitations

This study has several strengths. We investigated ARFID-related mental and somatic comorbidities in the largest reported sample of children and adolescents with the broad ARFID phenotype to date. Previous evidence on ARFID-related comorbidities was limited to small-scale, often uncontrolled clinical studies in subpopulations. Our large-scale epidemiological data, including clinical diagnoses from high-quality sources and different clinical settings,^18,19 enhance knowledge and generalizability of ARFID’s complex clinical presentations. We covered a broad range of medical conditions (n = 84), including several not previously examined in ARFID, such as various NDCs and specific respiratory, perinatal, and congenital conditions. Additionally, our indices of treatment needs (number of distinct per-person diagnoses, hospitalization duration) provide novel insight into the complex health care needs associated with ARFID.

Some limitations should be considered. First, diagnostic codes for ARFID, which were introduced into the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5)/ICD-11, were not available in this study. A composite measure was therefore created to identify ARFID using parent-reported and register data, aligning closely with DSM-5 criteria and consistent with our previous work.¹ Similar measures are used in other epidemiological ARFID studies.²⁶ Second, given the nature of this epidemiological study, it could not be determined whether certain conditions drove avoidant or restrictive eating or impaired nutrition and weight development. Some children identified with ARFID might have fit the broader pediatric feeding disorder phenotype.³³ For instance, severe gastrointestinal and cardiac disorders or surgery and malignancies might promote avoidant or restrictive eating. The increased cancer risk in ARFID requires cautious interpretation, as adverse effects of cancer treatment on appetite and weight may have mimicked ARFID symptoms. Nonetheless, sensitivity analyses excluding cancer diagnoses when analyzing number of distinct diagnoses and inpatient days in ARFID vs controls validated our main findings (eTables 6-7 in Supplement 1). Third, our phenotype definition primarily reflects ARFID cases with a sensory-based avoidance component, potentially introducing bias toward higher co-occurrence with NDCs.³⁴ Although lack of interest and fear profiles often co-occur with the sensory-based profile,³⁵ they may be underrepresented in our data. Fourth, we relied on NPR for diagnostic information, which does not include primary care diagnoses. Pediatric specialist referrals are recommended and thus common in Sweden.³⁶ However, we may have missed a portion of cases only seen in primary care and therefore not visible to us in the NPR, such as constipation, anxiety, and OCD. Since all children were affected by the underrepresentation of primary care diagnoses, relative risk assessments are unlikely to be significantly biased (main analyses 1-3; eFigure 1 in Supplement 1). Lastly, despite a sample size of more than 30 000 individuals, some sex-specific or age group–specific analyses were underpowered, highlighting the need for even larger datasets and longer follow-up periods.

Conclusions

In conclusion, this study provides the broadest and most detailed systematic evidence of coexisting mental and somatic conditions in the largest sample of children with and without ARFID to date, evaluating several previously unstudied conditions. We demonstrate that the broad ARFID phenotype is associated with higher prevalence and increased risk of multiple comorbidities, including a variety of neurodevelopmental, gastrointestinal, endocrine or metabolic, and neurological disorders. Importantly, we show that children with ARFID have greater treatment needs than control participants, characterized by multicomorbidity and prolonged hospitalizations. Our findings suggest heterogenous and complex clinical presentations of ARFID in childhood and adolescence, underscoring the critical importance of raising awareness and competence among health care professionals across specialties to improve ARFID detection and treatment. Future research should focus on longitudinal associations between ARFID and its comorbidities into adulthood as well as underlying genetic or environmental factors to better understand the etiology of ARFID.

Supplement 1.

eAppendix. Definition of Terms Used in the Manuscript and Supplementary Material

eTable 1. Descriptive Statistics of the Sample

eTable 2. Diagnostic Classification of Examined Mental and Somatic Conditions in ARFID and Controls

eTable 3. Prevalence of Mental and Somatic Conditions in ARFID and Controls Overall, per Sex, and per Age Group

eTable 4. Base Cox Regression Model Estimates in ARFID vs Controls

eTable 5. Cumulative Incidences of Mental and Somatic Conditions Before Ages 6, 12, and 18 Years in ARFID and Controls

eTable 6. Number of Distinct Overall, All Mental, and All Somatic Diagnoses in ARFID vs Controls

eTable 7. Number of Inpatient Days Due to Any, Any Mental, and Any Somatic Diagnosis in ARFID vs Controls

eTable 8. Sex-Stratified Cox Regression Model Estimates in ARFID vs Controls

eTable 9. Sex Effects on Number of Distinct Overall, Mental, and Somatic Diagnoses in ARFID vs Controls

eTable 10. Sex Effects on Number of Inpatient Days Due to Any, Any Mental, and Any Somatic Diagnosis in ARFID vs Controls

eTable 11. Age Group-Stratified Cox Regression Model Estimates in ARFID vs Controls

eFigure 1. Statistical Analysis Plan

eFigure 2. Hazard Ratio Plots for Co-Existing Conditions in Children and Adolescents With ARFID vs Controls

eFigure 3. Cumulative Incidence Plots in ARFID vs Controls for ICD-Chapter-Level, Grouped Allergic and Autoimmune Conditions

eFigure 4. Cumulative Incidence Plots in ARFID vs Controls for Individual and Grouped Conditions Within ICD-Chapters

eFigure 5. Histograms for Number of All Distinct, Distinct Mental, and Distinct Somatic ICD-Diagnoses in ARFID vs Controls (A) and for Duration of Hospitalization/Inpatient Treatment Due to Any, Any Mental, and Any Somatic ICD-Diagnosis in ARFID vs Controls (B)

eFigure 6. Age-Varying Hazard Ratios in ARFID vs Controls for ICD-Chapter F: Mental Conditions

Supplement 2.

Data Sharing Statement