Eating Disorders, Autoimmune, and Autoinflammatory Disease

Since 1968, every person living in Denmark has been assigned a unique personal identification number in the Danish Civil Registration System,²⁸ enabling linkage across national registers. The Psychiatric Central Research Register (PCRR) and the National Patient Register (NPR) capture all inpatient psychiatric and general hospital admissions and, since 1995, all outpatient and emergency department visits.²⁹ The registers include dates of admission and discharge, reason for contact, and main and auxiliary diagnoses³⁰ following a Danish version of the International Classification of Diseases, Eighth Revision (ICD-8) from 1969 to 1993 and the International Classification of Diseases, 10th Revision (ICD-10) codes from 1994.³¹ This study was approved by the Danish Data Protection Agency. Because all data were anonymized, informed consent was not required.

All children born in Denmark to Danish-born parents between 1989 and 2006 who were alive and residing in Denmark on their sixth birthday (N = 966 141; 48.7% girls) were included. Because of the limitations of the ICD-8 diagnoses of eating disorders, we included only individuals who could have received an incident ICD-10 eating disorder diagnosis (starting in 1994). Individuals born as part of a multiparous birth were excluded to reduce nesting (N = 35 144; 49.1% girls). The study population included 930 977 children (48.7% girls). The study population and their relatives (parents, siblings, half siblings, and cousins) were followed until their date of death, emigration from Denmark, or December 31, 2012, whichever came first.

We used the ICD-10 diagnoses in the PCRR or NPR as follows: AN (F50.0 AN; F50.1 atypical AN), BN (F50.2 BN; F50.3 atypical BN), and EDNOS (F50.8 other eating disorders; F50.9 eating disorder, unspecified). Diagnoses were not mutually exclusive (eg, AN and BN could be included at different times). Onset was age at first inpatient or outpatient contact with a diagnosis in the PCRR or NPR after age 6.

Autoimmune and autoinflammatory diseases were defined by using the Danish ICD-8 and ICD-10 diagnoses in the NPR (Supplemental Table 5). Age of onset was age at first inpatient or outpatient contact with an NPR diagnosis. We identified 50 autoimmune diseases or 10 autoinflammatory diseases (Supplemental Table 5). Because associations between eating disorders and autoimmune diseases could be driven by gastrointestinal factors, the autoimmune disease category also included a subcategory of diseases with a suspected presence of gastrointestinal involvement with no minimum age of onset.

We used Stata version 13 (Stata Corp, College Station, TX) and adjusted for sex by means of separate underlying hazard functions using Cox proportional hazards regression for each of the following eating disorder outcomes: AN, BN, and EDNOS. All estimates were adjusted for calendar-time in the parametric part of the Cox model, and for age in the nonparametric part. Calendar year (categorized as 1995–1999, 2000–2004, 2005–2009, and 2010–2012) and autoimmune disorder status were treated as time-dependent variables, whereas all other variables were considered time-independent. Hazard of autoimmune and autoinflammatory disease were calculated in a similar manner, with eating disorder status as a time-dependent exposure variable. Sex-specific analyses were conducted when cell sizes were adequate. The combined follow-up time was 8.5 million person-years for all survival analyses. The association between eating disorders in the proband and autoimmune and autoinflammatory diseases in family members was evaluated by calculating odds ratios (ORs) by using logistic regressions adjusted for sex and birth order (first, second, third, or fourth–18th). The presence of an eating disorder in the proband and the presence of autoimmune and autoinflammatory disease in family members was defined as a diagnosis at any time before December 31, 2012. Family members were grouped as parents, full siblings, maternal half siblings, paternal half siblings, and cousins. Because the ICD-10 diagnoses were only used to define eating disorders starting in 1994, the reverse analyses (eg, association between autoimmune and autoinflammatory disease in the proband and eating disorders in family members) could not be completed.

A total of 25 984 children and adolescents (54.2% girls) were diagnosed with an autoimmune or autoinflammatory disease. With the smoothed hazard curve, (Fig 1) we show that the risk for autoimmune but not autoinflammatory disease increases through childhood and adolescence. Of those diagnosed with autoimmune or autoinflammatory disease, 5617 (21.6%) had diseases with gastrointestinal involvement, 3456 individuals (13.3%) had diseases with autoinflammatory processes, and 157 (0.6%) individuals had a subsequent eating disorder diagnosis (Table 1).

In Cox models, hazard ratios (HRs) >1 indicate a higher probability of illness whereas ratios <1 indicate a lower probability of illness in comparison with unaffected individuals (Table 1). There was a 36% higher hazard for AN, 73% for BN, 72% for EDNOS, and 50% for any eating disorder for children and adolescents with autoimmune or autoinflammatory diseases. Autoimmune diseases with gastrointestinal involvement were significantly associated with higher hazards for AN (74% higher hazard) and for EDNOS (148% higher hazard). Autoinflammatory diseases were associated with significantly increased hazard for EDNOS (179% higher hazard). In sex-stratified analyses (Table 2), male children and adolescents with any autoimmune disease had a 368% higher hazard for BN and 198% for EDNOS. Male children and adolescents with any autoinflammatory disease had a 740% higher hazard for EDNOS. The hazards for male children have wide confidence intervals (CIs) and should be interpreted with caution.

A total of 3914 adolescents received at least 1 eating disorder diagnosis (91.9% girls) during the study period. With the smoothed hazard curve (Fig 1), we show that the risk for all eating disorders increases through age 14 and then remains high through the remainder of adolescence. Of those diagnosed with eating disorders, 2223 (56.3%) received an AN diagnosis, 727 (18.4%) a BN diagnosis, 1520 (38.5%) an EDNOS diagnosis, and 82 (2.1%) had a subsequent autoimmune or autoinflammatory diagnosis (Table 3). The percentages add to >100 because of overlapping categories.

AN was significantly associated with any subsequent autoimmune or autoinflammatory disease (64% higher hazard). EDNOS was significantly associated with any subsequent autoimmune or autoinflammatory disease (121% higher hazard) and autoimmune diseases with gastrointestinal involvement (97% higher hazard). In analyses stratified by sex, there were no male patients with eating disorders who had subsequent diagnoses of autoimmune or autoinflammatory disease.

Any family history of autoimmune or autoinflammatory disease was associated with significantly increased odds for AN, BN, and EDNOS (Table 4). Adolescents with parents who had autoimmune or autoinflammatory disease diagnoses had 13% higher odds of developing AN, 29% higher odds of developing BN, 27% higher odds of developing EDNOS, and 19% higher odds of developing any eating disorder. Individuals with full siblings who had autoimmune or autoinflammatory disease diagnoses had 48% higher odds of developing AN, 47% higher odds of developing EDNOS, and 37% higher odds of developing any eating disorder. Autoimmune or autoinflammatory disease in maternal or paternal half siblings were not found to be related to eating disorder status in probands. Individuals with cousins who had autoimmune or autoinflammatory disease also had 20% higher odds of developing EDNOS and 17% higher odds of developing any eating disorder.

Autoimmune and autoinflammatory diseases are associated with significantly increased hazard for subsequent eating disorders, and eating disorders are associated with a significantly increased hazard for subsequent autoimmune disease. These results confirm previous findings¹⁹; however, different measurement of risk, younger age of participants, and different follow-up times preclude direct comparison. We extend previous science by demonstrating that youth with EDNOS also have an increased hazard for autoimmune and autoinflammatory disease,¹⁹ and we present the first documented association between eating disorders and autoinflammatory disease and between familial autoimmune disease and eating disorders in offspring.

What mediates the association between autoimmune or autoinflammatory disease and eating disorders remains unknown. Eating disorders are commonly associated with reward dysfunction, and neural inflammation can lead to changes in reward functioning including heightened reactivity to threat, increased sensitivity to punishment, and decreased connectivity in reward pathways.^32,–35 Why autoimmune illness is so strongly associated with increased hazard for eating disorders in male youth is unclear; although our results are preliminary, this finding has been documented previously.¹⁹ The development of eating disorders in boys may be more dependent on biological vulnerability than for girls.³⁶ 

With our evidence, we suggest that eating disorders are associated with autoimmune disease with gastrointestinal involvement. EDNOS predicted a higher hazard of a subsequent autoimmune disease with gastrointestinal involvement. Moreover, autoimmune disease with gastrointestinal involvement predicted a higher hazard of subsequent AN and EDNOS. This association may be a diagnostic artifact caused by the differential diagnosis between eating disorders and gastrointestinal autoimmune diseases. AN and EDNOS may be misdiagnosed in individuals whose weight loss is due to malabsorption from gastrointestinal autoimmune illness such as Crohn disease or celiac disease. Gastrointestinal autoimmune diseases may also be misdiagnosed in patients with AN and EDNOS because of a misinterpretation of the underlying causes of food reactivity and delayed gastric emptying that are often a consequence of fasting behaviors.^37,38 

Alternatively, proinflammatory cytokines produced by autoimmune diseases such as Crohn disease (eg, tumor necrosis factor α) lead to appetite suppression and weight loss, which may trigger eating disorder symptoms in vulnerable individuals.^39,40 In a case study,³⁹ administration of an antitumor necrosis factor α treatment of a woman comorbid for Crohn disease and AN led to weight gain and decreased eating disorder symptomatology. Behaviorally, pain in response to eating and avoidance of foods that cause flare-ups for gastrointestinal autoimmune disease could elicit eating disorder behavior or exacerbate symptoms in individuals with eating disorders. The associations between autoimmune disease and eating disorders are also not limited to gastrointestinal disorders, making misdiagnosis unlikely to be the sole explanation for the documented associations.

To our knowledge, this is the first study in which researchers document an association between autoinflammatory diseases and eating disorders. This fairly new classification of diseases is marked by diseases of the innate immune system and high levels of clinical inflammation. Genetic factors have been implicated in many autoinflammatory diseases,^41,42 and researchers who conducted animal research suggest that the interplay between nutritional composition and intestinal microbiota may contribute to their onset. In these studies,⁴³ mice genetically predisposed to autoinflammatory disease and exposed to a low-fat diet experienced changes in their intestinal bacteria (increases in Prevotella), which led to the onset of autoinflammatory disease. Although little is known about the intestinal microbiota in patients with EDNOS, we⁴⁴ have found that the intestinal microbiota of patients with acute AN is characterized by significantly lower taxa diversity than controls. Future research in this area should also include measures of nutritional composition and intestinal microbiota.

Autoimmune and autoinflammatory disease in parents is also associated with significantly increased odds for all eating disorders (AN, BN, EDNOS). Autoimmune and autoinflammatory disease in full siblings is also associated with significantly increased odds for eating disorders (AN and EDNOS in particular) and autoimmune and autoinflammatory disease in cousins is also associated with significantly increased odds for eating disorders (EDNOS in particular). Although the pattern of the association across varying degrees of relatedness can assist in making inferences about whether genetic or environmental factors explain the association,²⁷ we found no clear patterns to make this distinction, potentially caused by low power because of the small number of half siblings. That autoimmune disease in cousins who share fewer environmental exposures was associated with eating disorders may, however, point to shared genetic risk. Supporting this interpretation is the discovery of the first genome-wide significant locus for AN on chromosome 12 (rs4622308) in a region harboring a previously reported type 1 diabetes and autoimmune disease locus.⁴⁵ In addition, linkage disequilibrium score genetic correlations revealed intriguing negative genetic correlations (albeit nonsignificant) between AN and some autoimmune diseases (rheumatoid arthritis, ulcerative colitis, and Crohn disease), suggesting shared genetic effects with opposite directions of effect.⁴⁶ As sample sizes increase, we expect greater clarity in the association of eating disorders and autoimmune diseases on the genomic level. Moreover, genome-wide association data are not yet available for BN and EDNOS.

Importantly, unmeasured environmental mediators could explain the observed associations. These could be specific, shared environmental exposures (eg, nutrition, smoking exposure) that place family members at risk for both classes of illness or general, such as increases in familial stress because of illness or greater illness detection because of increased healthcare contact. Adequately-powered molecular genetic and twin designs are necessary to explicate these genetic and environmental pathways.

No private psychiatric hospitals exist in Denmark; hospital-based treatment is free of charge, and the nationwide registration of psychiatric disorders is almost complete. However, psychiatrists and psychologists do work in private practice and treat up to 20% of individuals with eating disorders.³⁰ These individuals may not be captured in the current study. The use of medical surveillance may have led to increased documented relative prevalence of AN in this sample in comparison with BN and EDNOS. Of those with eating disorders in the current study, 2233 (57.1%) received an AN diagnosis, 727 (18.6%) received a BN diagnosis, and 1520 (38.8%) received an EDNOS diagnosis, which differs from population-based surveys. It is likely that individuals with BN and EDNOS were under-detected, which could have affected our conclusions in 2 ways. First, patients with eating disorders who had inpatient and outpatient hospital-based contacts may be more medically compromised and more likely to also suffer from an autoimmune disease, thereby elevating the observed association. Second, undetected individuals with BN and EDNOS could have increased our statistical power to examine this association and thus, the observed associations are attenuated.

In addition, we only examined eating disorders in children and adolescents who were <24 years using the Danish registers because of the limitations of the ICD-8 diagnoses, which may have attenuated HRs for both the antecedent and subsequent autoimmune and autoinflammatory diseases. The peak age of incidence for eating disorders occurs during adolescence,²⁴ but the period of incident risk for eating disorders extends across the lifespan. Although some autoimmune and autoinflammatory diseases onset in childhood and adolescence (eg, type 1 diabetes, Crohn disease) and would be captured in these analyses,⁴⁷ others typically do not onset until midlife (eg, rheumatoid arthritis).⁴⁸ As this study population ages, the HRs may intensify or attenuate commensurably. Like eating disorders, some autoimmune and autoinflammatory diseases may be under-detected in primary care, which may also have attenuated these associations.

Because of the limitations of the ICD-8 diagnoses, we were unable to examine potential associations between parental eating disorders and autoimmune and autoinflammatory diseases in offspring. Although we could have included parental diagnoses of eating disorders after the PCRR deployed the ICD-10 in 1994, we would have missed a large number of parents with an eating disorder history before 1994.

The biological associations between autoimmune and autoinflammatory disease and eating disorders are becoming increasingly clear. Our findings support compelling lines of evidence from researchers suggesting that immune system disturbance is both comorbid with psychiatric disorders and can increase risk for illness.^3,–5,19 Research into immune system dysfunction represents a novel lens to examine risk for eating disorders and could inform new treatments. Novel ideas are desperately needed in eating disorders; only 10% of adolescents receive treatment, the average duration of most eating disorders is >7 years, ∼25% of adolescents develop chronic illness, and there are no medications specifically for AN.⁴⁹ Supported by our findings, we suggest further examination into the complex interplay between immune system disturbance, nutritional, and neuropsychiatric processes. Understanding the role of the immune system could explicate the etiology of eating disorders, aid early detection, personalize intervention, and ultimately improve outcomes.

 
• AN

  anorexia nervosa

 
• BN

  bulimia nervosa

 
• CI

  confidence interval

 
• EDNOS

  eating disorder not otherwise specified

 
• HR

  hazard ratio

 
• ICD-8

  International Classification of Diseases, Eighth Revision

 
• ICD-10

  International Classification of Diseases, 10th Revision

 
• NPR

  National Patient Register

 
• OR

  odds ratio

 
• PCRR

  Psychiatric Central Research Register

Thanks to Prof Kerstin von Plessen for additional assistance with the Danish ICD diagnostic codes.