Co-existing mental and somatic conditions in Swedish children with the avoidant restrictive food intake disorder phenotype

Background: 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 a range of psychiatric and somatic symptoms, and therefore consult various pediatric subspecialties; large-scale studies mapping comorbidities are however lacking. To characterize health care needs of people with ARFID, we systematically investigated ARFID-related mental and somatic conditions in 616 children with ARFID and >30,000 children without ARFID. Methods: In a Swedish twin cohort, we identified the ARFID phenotype in 6–12-year-old children based on parent-reports and register data. From >1,000 diagnostic ICD-codes, we specified mental and somatic conditions within/across ICD-chapters, number of distinct per-person diagnoses, and inpatient treatment days between birth and 18th birthday (90 outcomes). Hazard ratios (HR) and incidence rate ratios (IRR) were calculated. Findings: Relative risks of neurodevelopmental, gastrointestinal, endocrine/metabolic, respiratory, neurological, and allergic disorders were substantially increased in ARFID (e.g., autism HR[CI95%]=9.7[7.5–12.5], intellectual disability 10.3[7.6–13.9], gastroesophageal reflux disease 6.7[4.6–9.9], pituitary conditions 5.6[2.7–11.3], chronic lower respiratory diseases 4.9[2.4–10.1], epilepsy 5.8[4.1–8.2]). ARFID was not associated with elevated risks of autoimmune illnesses and obsessive-compulsive disorder. Children with ARFID had a significantly higher number of distinct mental diagnoses (IRR[CI95%]=4.7[4.0–5.4]) and longer duration of hospitalizations (IRR[CI95%]=5.5[1.7–17.6]) compared with children without ARFID. Children with ARFID were diagnosed earlier with a mental condition than children without ARFID. No sex-specific differences emerged. Interpretation: This study yields the broadest and most detailed evidence of co-existing 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. Funding: Fredrik and Ingrid Thurings Foundation, Mental Health Foundation.


Definition of terms used in the Manuscript and Supplementary Material
ARFID, children and adolescents with the avoidant restrictive food intake disorder phenotype; controls, children and adolescents without the avoidant restrictive food intake disorder phenotype; sex, sex assigned at birth.

Short labels for outcomes
Collapsed labels to indicate analyzed conditions in a space-efficient manner (applicable to all figures and tables in the Manuscript and Supplementary Material): ChromosomalAbnorm, chromosomal abnormalities; Malform, malformation; NoninfectiveIBD, noninfective inflammatory bowel disease; ObsessCompulsDisorder, obsessive-compulsive disorder; OtherUpperGastrointest, other upper gastrointestinal conditions; Perinat, perinatal; PerinatCNSConditions, perinatal conditions of the central nervous system; PerinatRespiratCardiovasc, perinatal respiratory and cardiovascular conditions; ScholastDevelopment, scholastic development; VascArterial, vascular disorders of the arterial system; VascVenous, vascular disorders of the venous system.Absolute number (n) and percentage of individuals with ARFID and controls are stated overall and per sex, age at CATSS assessment, and birth year.Row-wise percentages were calculated (separately for 1., 2., 3., and 4.).Birth years were pooled in groups of five (first group 1992-1996) or four (all other groups) consecutive years.Pearson  2 tests were conducted to compare counts (highlighted with light-gray background) in ARFID vs.
controls.Abbreviations: ARFID, avoidant restrictive food intake disorder; CATSS, Child and Adolescent Twin Study in Sweden.415,416,417,420,421,422,423,424,425    Cumulative incidence (Cum.Inc., absolute risk) as the probability (proportion) of having any of the analyzed conditions before three consecutive time points (i.e., before 6, 12, and 18 years of age) is stated in ARFID and controls, including lower/upper limits of the cluster-robust 95% confidence interval (95% CI LL/UL).Perinatal conditions were excluded [n/a] given their occurrence during the perinatal period and lack of age-related dynamics).Cumulative incidence was estimated using the Kaplan-Meier method (Cum.Inc. = 1 -Kaplan-Meier survival estimate, accounting for censoring) after creating a modified analysis sample by matching n=10 unexposed individuals to each (n=1) individual twin with ARFID, stratified by sex and birth year.Conditions are sorted in accordance with Table S4, i.e., in descending order of base cox regression model hazard ratios within ICD-chapters, the group of further conditions, and grouped allergic and autoimmune conditions (highlighted with blue background).Abbreviations: ADHD, attention deficit hyperactivity disorder; ARFID, avoidant restrictive food intake disorder; CNS, central nervous system; GERD, gastro-esophageal reflux disorder; IBD, inflammatory bowel disease; IBS, irritable bowel syndrome.[1992-2008, factorized]; robust sandwich estimates given clustered twin data).Age group-specific HRs were compared (HRage group 1/HRage group 2 and HRage group 3/HRage group 2) and statistics of both comparisons are stated: lower/upper limits of the cluster-robust 95% confidence interval of HRage group 1/HRage group 2 and HRage group 3/HRage group 2, raw p-value, and significance according to the false discovery rate-adjusted threshold (FDR/Benjamini-

Hochberg procedure [BH]
).There was one significant age group difference (HRage group 3 vs.HRage group 2 for ChapterF:MentalConditions, marked by an asterisk [*] and orange background) in ARFID vs. controls at FDR=0.00014 (across 103 supplementary tests).All other nominally significant age group differences (p<0.050) are highlighted with light-orange background.Conditions are sorted in ascending order of HRage group 3/HRage group 2 for easier orientation.Abbreviations: ADHD, attention deficit hyperactivity disorder; ARFID, avoidant restrictive food intake disorder; IBD, inflammatory bowel disease.

Figure S1. Cumulative incidence plots in ARFID vs. controls for individual and grouped conditions
Cumulative incidence of having a specific condition (%) and lower and upper limits of its cluster-robust 95% confidence interval are plotted over age (0 to <18 years) in ARFID vs. controls.The Kaplan-Meier-based estimation method is explained in the statistical analysis section of the main article and in the legend of Table S5.Conditions are sorted in accordance with Table S4, i.e., in descending order of base Cox regression model hazard ratios within ICD-chapters, the group of further conditions, and grouped allergic and autoimmune conditions (please consult Figure 4 in the main article for cumulative incidence plots for ICD-chapters and grouped allergic and autoimmune conditions; perinatal conditions excluded given their occurrence during the perinatal period and lack of age-related dynamics).Abbreviations: ADHD, attention deficit hyperactivity disorder; ARFID, avoidant restrictive food intake disorder; GERD, gastro-esophageal reflux disorder; IBD, inflammatory bowel disease; IBS, irritable bowel syndrome.

Figure S2. Age-varying hazard ratios in ARFID vs. controls for ICD-chapter F: mental conditions
Following-up on significant age group differences (HRage group 3 vs.HRage group 2) in ARFID vs. controls for ICD-chapter F: mental conditions (Table S11), time-varying hazard ratios (HRs) in ARFID vs. controls and lower and upper limits of their cluster-robust 95% confidence intervals are plotted over age (0 to <18 years).The curve was estimated using the base Cox regression model (predictor: group [ARFID vs. controls]; covariates: sex [female/male], birth year [1992-2008, factorized]; robust sandwich estimates given clustered twin data) and adding age-/time-related variance via cubic splines with pre-specified knots at 3, 6, 9, 12, and 15 years of age (boundary knots at 0 and <18 [≈17.9] years of age).Abbreviation: ARFID, avoidant restrictive food intake disorder.

Table S4 . Base Cox regression model estimates in ARFID vs. controls
Stated are for each condition: hazard ratio in ARFID vs. controls (HR), lower/upper limits of the cluster-robust 95% confidence interval (95% CI LL/UL), cluster-robust standard error (SE), z statistic, raw p-value, and significance according to the false discovery rate-adjusted threshold (FDR/Benjamini-Hochberg procedure[BH]) from the base Cox regression model (predictor: group [ARFID vs. controls]; covariates: sex [female/male], birth year[1992-  2008, factorized]; robust sandwich estimates given clustered twin data).Significantly different risks in ARFID vs. controls at FDR=0.0343 are marked by an asterisk (*).Conditions are sorted in descending order of HRs within ICD-chapters, the group of further conditions, and grouped allergic and

Table S6 . Number of distinct overall, all mental, and all somatic diagnoses in ARFID vs. controls
Mean and standard error of the mean (SEM) for number of all distinct diagnoses (i.e., unique diagnostic ICD-codes), number of distinct mental diagnoses, and number of distinct somatic diagnoses per individual twin are given in ARFID and controls.As test statistics, incidence rate ratio in ARFID vs. controls (IRR), lower/upper limits of the cluster-robust 95% confidence interval (95% CI LL/UL), cluster-robust standard error (SE), z statistic, raw p-value, and significance according to false discovery rate-adjusted threshold (FDR/Benjamini-Hochberg procedure[BH]) are stated for all distinct diagnoses, distinct mental diagnoses, and distinct somatic diagnoses.Statistics were obtained with Poisson regression models (predictor: group [ARFID vs. controls]; covariates: sex [female/male], birth year[1992-2008, factorized]; robust sandwich estimates given clustered twin data; offset term to account for follow-up/exposure time).Significantly different incidence rates in ARFID vs. controls at FDR=0.0343 are marked by an asterisk (*).Abbreviation: ARFID, avoidant restrictive food intake disorder.

Table S7 . Number of inpatient days due to any, any mental, and any somatic diagnosis in ARFID vs. controls
Mean and standard error of the mean (SEM) for number of inpatient days per individual twin due to any diagnosis (i.e., unique diagnostic ICD-code), any mental diagnosis (as main diagnosis/cause of hospitalization), and any somatic diagnosis (as main diagnosis/cause of hospitalization) are given in ARFID and controls.As test statistics, incidence rate ratio in ARFID vs. controls (IRR), lower/upper limits of the cluster-robust 95% confidence interval (95% CI LL/UL), cluster-robust standard error (SE), z statistic, raw p-value, and significance according to false discovery rate-adjusted threshold (FDR/Benjamini-Hochberg procedure [BH]) are stated for any diagnosis, any mental diagnosis, and any somatic diagnosis.Statistics were obtained with Poisson regression models (predictor: group [ARFID vs. controls]; covariates: sex [female/male], birth year[1992-2008, factorized]; robust sandwich estimates given clustered twin data; offset term to account for follow-up/exposure time).Significantly different incidence rates in ARFID vs. controls at FDR=0.0343 are marked by an asterisk (*).Abbreviation: ARFID, avoidant restrictive food intake disorder.

Table S8 . Sex-stratified Cox regression model estimates in ARFID vs. controls Condition Sex: female Sex: male Comparison female vs. male
There were no statistically significant sex differences (HRfemale vs. HRmale) in ARFID vs. controls at FDR=0.00014 (across 103 supplementary tests).Nominally significant sex differences (p<0.050) are highlighted with light-orange background.Conditions are sorted in descending order of HRfemale/HRmale for easier orientation.Abbreviations: ADHD, attention deficit hyperactivity disorder; ARFID, avoidant restrictive food intake disorder; CNS, central nervous system; GERD, gastro-esophageal reflux disorder; IBD, inflammatory bowel disease.

Table S9 . Sex effects on number of distinct overall, mental, and somatic diagnoses in ARFID vs. controls
Mean and standard error of the mean (SEM) for number of all distinct diagnoses (i.e., unique diagnostic ICD-codes), number of distinct mental diagnoses, and number of distinct somatic diagnoses per individual twin are given in female and male individuals with ARFID and female and male controls.As test statistics, incidence rate ratio of the group (ARFID vs. controls)-x-sex interaction effect (IRR), lower/upper limits of the cluster-robust 95% confidence interval (95% CI LL/UL), cluster-robust standard error (SE), z statistic, raw p-value, and significance according to supplementary false discovery rate-adjusted threshold (FDR/Benjamini-Hochberg procedure [BH]) are stated for all distinct diagnoses, distinct mental diagnoses, and distinct somatic diagnoses.Statistics were obtained with Poisson regression models including a group-x-sex interaction term (other predictors/main effects: group [ARFID vs. controls], sex [female/male]; covariate: birth year[1992-2008, factorized]; robust sandwich estimates given clustered twin data; offset term to account for follow-up/exposure time).There were no significant group-x-sex interaction effects at FDR=0.00014 (across 103 supplementary tests).a IRRs>1 indicate greater incidence rates in females with ARFID than males with ARFID.IRRs<1 indicate greater incidence rates in males with ARFID than females with ARFID (regarding directionality [not statistical significance]).Abbreviation: ARFID, avoidant restrictive food intake disorder.

Table S10 . Sex effects on number of inpatient days due to any, any mental, and any somatic diagnosis in ARFID vs. controls
Mean and standard error of the mean (SEM) for number of inpatient days per individual twin due to any diagnosis (i.e., unique diagnostic ICD-code), any mental diagnosis (as main diagnosis/cause of hospitalization), and any somatic diagnosis (as main diagnosis/cause of hospitalization) are given in female and male individuals with ARFID and female and male controls.As test statistics, incidence rate ratio of the group (ARFID vs. controls)- x-sex interaction effect (IRR), lower/upper limits of the cluster-robust 95% confidence interval (95% CI LL/UL), cluster-robust standard error (SE), z statistic, raw p-value, and significance according to supplementary false discovery rate-adjusted threshold (FDR/Benjamini-Hochberg procedure[BH]) are stated for any diagnosis, any mental diagnosis, and any somatic diagnosis.Statistics were obtained with Poisson regression models including a group-x-sex interaction term (other predictors/main effects: group [ARFID vs. controls], sex [female/male]; covariate: birth year[1992-2008, factorized]; robust sandwich estimates given clustered twin data; offset term to account for follow-up/exposure time).There were no significant group-x-sex interaction effects at FDR=0.00014 (across 103 supplementary tests).a IRRs>1 indicate greater incidence rates in females with ARFID than males with ARFID.IRRs<1 indicate greater incidence rates in males with ARFID than females with ARFID (regarding directionality [not statistical significance]).Abbreviation: ARFID, avoidant restrictive food intake disorder.

Table S11 . Age group-stratified Cox regression model estimates in ARFID vs. controls Condition Age group 1: 0 to <6 years Age group 2: 6 to <12 years Age group 3: 12 to <18 years Comparison age groups 1 vs. 2 Comparison age groups 3 vs. 2
Age group-specific hazard ratios in 0 to <6-year-old individuals with ARFID vs. controls of the same age group (HRage group 1), 6 to <12-year-old individuals with ARFID vs. controls (HRage group 2), and 12 to <18-year-old individuals with ARFID vs. controls (HR age group 3 ), lower/upper limits of the cluster-robust 95% confidence interval (95% CI LL/UL), and p-values are stated for each analyzed condition (conditions with n<5 individuals per age group in ARFID or controls were not analyzed) from the age group-stratified Cox regression model (predictor: group [ARFID vs. controls]; age/time group strata [age groups 1-3]; covariates: sex [female/male], birth year