Comorbid mental disorders during long‐term course in a nationwide cohort of patients with anorexia nervosa

Abstract Objective Comorbid mental disorders in anorexia nervosa during long‐term course require detailed studies. Method This matched cohort study was based on nationwide Danish register data of all patients born 1961–2008 with a first‐time ICD‐10 diagnosis of anorexia nervosa (AN) between 1994 and 2018 at age 8–32 and matched controls taken from all individuals without an eating disorder (ED). For nine categories of non‐eating mental disorders, time from date of first AN‐diagnosis (inclusion date) to time of first diagnosis, accounting for censoring, was studied by use of time‐stratified Cox models. Results A total of 9,985 patients with AN (93.5% females) and 49,351 matched controls were followed for a median (IQR) of 9.0 (4.4–15.7) years. For patients, there was about 25% and 55% risk of receiving any non‐ED disorder during the first 2 years and two decades after inclusion, respectively. A hazard ratio (HR) of seven for any non‐ED was found for the first 12 months after inclusion, a ratio that reduced to two at five or more years after inclusion. During the first years, large HRs ranging in 6–9 were found for affective, autism spectrum, personality, and obsessive–compulsive disorders with the latter displaying the highest continuous increased risk. The HR at 12 months after inclusion was highest for any non‐ED disorder and affective disorders in patients aged 8–13 at diagnosis. Discussion Comorbid mental disorders in AN are most frequently diagnosed in the first years after diagnosis of AN and on longer terms imply a double immediate risk.

A systematic review of all long-term follow-up studies of the clinical course of AN performed in the 20th century reported a variety of comorbid disorders including anxiety disorders and phobias (25.5%), affective disorders (24.1%), substance use disorders (14.4%), obsessive-compulsive disorders (12.2%), and various personality disorders (16.6-31.4%) (Steinhausen, 2002). More recent follow-up studies have revealed that comorbid depression at the start of treatment is a negative predictor of outcome for the restrictive type of AN (Eskild-Jensen, Stoving, Flindt, & Sjogren, 2020;Franko et al., 2018), and that OCD is both a frequent life-time and follow-up diagnosis (Mandelli, Draghetti, Albert, De Ronchi, & Atti, 2020). Both OCD (Mandelli et al., 2020) and autistic traits are also associated with a less favorable outcome (Wentz et al., 2009). At the 5-10-year follow-up, 28% of patients with a childhood onset of AN had a current comorbidity and 64% had a past comorbid diagnosis (Herpertz-Dahlmann et al., 2018).
Age of onset has been shown to have an effect on outcome. In particular, onset of AN during adolescence was associated with a more benign outcome in a sizeable number of outcome studies (Steinhausen, 2002;Steinhausen, Boyadjieva, Griogoroiu-Serbanescu, & Neumarker, 2003). None of these studies considered the association between AN and comorbid mental disorders concurrent or subsequent to the AN diagnosis as outcomes.
Further, previous research has not accounted for the risk in controls, implying that the reported results may to some degree reflect common challenges in adolescent life.
The present report is part of a multi-domain study of the longterm course of a nationwide Danish cohort of patients with AN compared to matched controls. In the entire set of analyses, a specific focus is on the effect of age at diagnosis as the best proxy of age at onset of AN. The aim of this study was to investigate whether concurrent and subsequent mental disorders during longterm follow-up are more frequent and differ across time in AN patients compared to controls, and whether age at diagnosis of AN modifies the association.

| Study samples
The study was based on the complete nationwide cohort of all individuals born 1961-2008 who were diagnosed with an eating disorder (ED) in 1994-2018 and at the time of the first diagnosis were at age 8-32 years. EDs were defined by the International Classification of Diseases and Related Health Problems, 10th edition (WHO, 1992), code F50 in the National Patient Register (NPR).
Patients with AN (F50.0, F50.1) were identified in the registers as individuals with at least one AN diagnosis at the first day of registration with a non-emergency incident ED diagnosis, regardless of whether the patient simultaneously had a diagnosis of bulimia nervosa. The age range of the sample was guided by the idea that the validity of the diagnosis of AN is uncertain outside this age range. Based on studies analyzing childhood AN (see Herpertz-Dahlmann et al., 2018), we decided to truncate the empirical distribution at the lower age of 8 years at first diagnosis. The upper age at first diagnosis limitation was guided by the following considerations. First, it is questionable whether late onset exists or is due to late diagnosis (Scholtz, Hill, & Lacey, 2010). Secondly, in a previous study on life-time incidence rates of first-time diagnosed AN up to age 65 based on Danish registry data, we found a sharp decline of incidences starting with the age-group 30-39 (Scholtz et al., 2010;Steinhausen & Jensen, 2015).
To avoid prevalent cases from being classified as incident cases, all patients with access to services before 1994 were excluded by study design. This procedure, which was used separately for each disorder, implies an exclusion of all patients and controls with the ICD-8 classification. In Denmark, the ICD-8 classification was used from 1969 to 1993, the ICD-9 was never adopted, and the ICD-10 was used from 1994.
Using a matched design with a 1:5 ratio, controls alive at the time of inclusion of matched patients were identified via the Danish Civil Registration System (DCRS), which covers the total general population. The date of the first AN-diagnosis defines the inclusion date for both the patients and the controls. The matching was done on sex, age, and area of residence at the time of the first eating disorder diagnosis in the patients. By design, controls did not have any ED diagnosis prior to or at any time after inclusion. All participants were alive, and resided in Denmark at the initiation of follow-up. 49,351 controls. The sample was divided into three subgroups by age in years at first diagnosis of anorexia nervosa, covering 8-13, 14-17, and 18-32 at the time of inclusion. The three subgroups were defined according to common developmental distinctions between childhood including preadolescence (age 8-13), adolescence (age 14-17), and adulthood (age 18 and older).

| Procedure
The study was based on the comprehensive Danish system of registries, which includes mandatory recording of patient visits to the public health service. The system covers various domains of data collection including extensive health information on the total national population since 1968 (Pedersen, 2011). In the present study, based on the personal identifica-

| Statistical analyses
Using Cox proportional hazards models with cluster robust standard errors, time from first AN-diagnosis to first diagnosis, accounting for censoring, was investigated for each of the nine mental disorders on the matched cohort data. Separate effects were estimated on periods for which time-invariance was assessed with Aalen's linear hazard models (Lee & Weissfeld, 1998). The Cox models were adjusted for previous mental diagnosis during the 2 years prior to inclusion, sex, study period at time of inclusion (1994-1999, 2000-2004, 2005-2009, 2010-2014, 2015-2018), family income (in tertiles), and parental previous mental disorder at the time of inclusion. The assumptions of proportional hazards were verified via Schoenfeld residuals tests. Tests for interaction between age group and anorexia nervosa (patients vs. controls) for each mental disorder were conducted. The method proposed by Holm (1979) was used to correct for multiple testing. The risk of each of the nine mental disorders over time in patients and controls were compared using cumulative incidences, with death as competing risk, for each agesubgroups as well as the combined sample. To guarantee anonymity of individuals, cumulative incidences are graphed after applying a multiplicative distortion of a Gaussian distributed random error with mean 1 and SD 0.05. All analyses were performed with Stata 16.1 (Statacorp, College Station, TX) on servers at Statistics Denmark.

| RESULTS
The baseline characteristics of each group are shown in Table 1.
F I G U R E 1 Flowchart for identification of the study population The largest age group (45%) was 18-32 years old, while 38% were 14-17 and 17% were 8-13 years old. Median (IQR) follow-up time in years for the two samples was 9.0 (4.4-15.7). At the time of study inclusion, the AN sample consisted of 1,349 (13.5%) inpatients, 8, 595 (86.1%) outpatients and a negligible number of 41 patients in a dayclinic (0.4%). As controls were matched on age listed in years, the age at inclusion for controls exceeds the age range 8-32 (range 7.6-33.9).
A higher proportion of patients with AN than controls (12.7% vs. 3.6%) had a mental diagnosis in the 2 years prior to inclusion, but frequencies of the various other mental disorders in the two samples prior to inclusion were rather low. Among patients and controls, 0.7% respectively 0.2% were lost because of death.
For documentation, regardless of censored individuals, the frequencies of mental disorders other than AN for the entire observation period for the total samples and the various age groups are listed in Table S1. Cumulative incidence curves for each of the nine mental disorders for patients and controls are collected with Figure 2 for the total sample and with Figure 3 for the three age groups 8-13, 14-17, and 18-32. Table 2 lists cumulative incidences at selected time points for non-ED disorders in the total and age stratified samples of patients and controls. More than double cumulative incidence was observed at any time during follow-up for patients when compared to controls in each stratum. For example, at 4, 10, and 22 years the cumulative incidence (95% CI) for the total sample of patients was 32.1% (31.0%-33.3%), 44.3% (42.9%-45.6%), and 56.1% (54.2%-57.9%), while for controls it was 7.70% (7.46%-7.95%), 15.4 (15.0%-15.7%), and 25.0% (24.3%-25.6%).
For each of the nine categories of mental disorders, Table 3 shows HRs for patients compared to controls for the total sample and the three age groups 8-13, 14-17, and 18-32. For all mental groups and on each stratum, the HRs were larger when closer to inclusion while reduced to about two to three in the long term. For example, for any non-ED mental disorder during the first 12 months the hazard ratios were HR = 17.4 (95% CI, 14.0-21.7) and HR = 5.04 (95% CI, 4.42-5.75) for patients aged 8-13 and 18-32 at inclusion, respectively, while at five or more years the two ratios had reduced to HR = 2.17 (95% CI, 1.78-2.65) and HR = 2.02 (95% CI, 1.77-2.31), respectively. During the first 2 years, a HR larger than nine was found for affective, obsessive-compulsive, and personality disorders in the age group 8-13. In the combined sample, large HRs ranging in 6-9 during the first years were found for affective, obsessive-compulsive, any non-ED mental, autism spectrum, and personality disorders. There were no indications of non-proportional hazards (p > .70) for any disorder. Differences across age groups pertained to any non-ED mental, obsessive-compulsive, affective, and adjustment disorders at the first years after inclusion with higher HR for the age groups 8-13 and lower HR for the age group 18-32.

| DISCUSSION
The present study used a large nationwide sample of patients with AN followed-up with a median duration of 9.0 years. The findings contribute to the more limited findings based on previous observations made in follow-up studies of rather small clinical samples with varying follow-up periods, and mostly without any controls, that never considered analyses in a time-to-event setting.
In the present study, there was about a 25% and a 55% risk of receiving any non-ED disorder for the patients during the first 2 years and two decades after inclusion, respectively. A HR of seven for any non-ED was found for the first 12 months after inclusion, a ratio that reduced to two at five or more years after inclusion. During the first years after inclusion, large HRs ranging in 6-9 were found for affective, autism spectrum, personality, and obsessive-compulsive disorders with the latter displaying the highest continuous increased risk.
The HR at 12 months after inclusion was highest for any non-ED disorder and affective disorders in patients aged 8-13 at diagnosis.
The findings of this study reflect a pronounced overall diagnosed psychiatric comorbidity in AN across time. For patients, one out of three was assigned at least one mental disorder in the 4 years after receiving their first AN diagnosis, while it was only eight out of 100 for the controls. Cumulative incidences of eight major categories of mental disorders among patients showed notable increased risks over time for patients compared to controls. The immediate risks were always at double or more when compared to controls and for some disorders with much larger HR when close to inclusion, particularly for the age group 8-13. For example, for any non-ED the HR was at seven during the first 12 months after inclusion and reduced to two at 5 years after inclusion. Close to inclusion, controls were at lower risk of any non-ED at young age than at older age, while a subdued but reversed pattern was found for the patients, leading to particular high HR close to inclusion for the age group 8-13.
This marked burden imposed on individuals with AN by additional mental disorders at the time of, and soon after, the first AN diagnosis is reflected in Figure 2  In addition to these general patterns of cumulative incidences, age at diagnosis proved to have a relevant impact on the occurrence of comorbid disorders for patients with AN. It was particularly the youngest age group with a diagnosis of AN as early as age 8-13 that had the highest hazards when compared to controls for any non-ED, obsessive-compulsive disorders, affective disorders, and adjustment disorders while the ratios were lower for those who were diagnosed only later at age 18-32. While it cannot be ruled out that different traditions and concepts of training in assessment as performed by child and adolescent psychiatrists versus adult psychiatrists may have contributed to this age effect, these findings point to the high vulnerability of the youngest group and also represent a more precise definition of the age at risk than described before in the literature on outcome of AN (Steinhausen, 2002). Obviously, the common distinction of age at onset in adolescence, that is, <18 years of age, versus adult age is not precise enough. However, it has to be admitted that the common reference to age at onset with its retrospective nature of definition may not have been used precisely enough in many of the older studies so that the term of age at diagnosis used in the present study is definitely more valid.
Furthermore, our study found that the hazard of being assigned a comorbid psychiatric diagnosis was declining in patients with AN com-   (Steinhausen, 2002).
In summary, the present findings may be regarded as coming from the most representative presentation of various comorbid disorders of AN in a single study. They extend the findings of a series of previous studies that focused predominantly on the prevalence rather than the incidence of single disorders in relatively small samples either at clinical intake or in uncontrolled follow-ups (Carrot et al., 2017;Fouladi et al., 2015;Franko et al., 2018;Herpertz-Dahlmann et al., 2018;Koch et al., 2015;Martinussen et al., 2017;Meier et al., 2015;Nazar et al., 2016;Root et al., 2010;von Lojewski et al., 2013;Wentz et al., 2009;Westwood & Tchanturia, 2017;Yilmaz et al., 2020) so that direct comparisons with the results of these studies are hampered.  While we studied the entire range of mental disorders, the absence of hyperkinetic disorders is noteworthy. The awareness of this comorbidity may still have been insufficient during our study period, particularly in adult patients (Svedlund, Norring, Ginsberg, & von Hausswolff-Juhlin, 2017) as the frequency of hyperkinetic disorders were <5 in our data, meaning that the data could not be included in the analyses due to legislation on anonymization of data from the Danish registers.
Other concerns related to studies based on registries are less limiting. There was no independent verification of the accuracy of diagnoses entered in the DCPRR in the present study, although prior quality checks on the DCPRR suggest that diagnostic validity is high across various disorders (Kessing, 1998;Lauritsen et al., 2010;Loffler et al., 1994;Mohr-Jensen et al., 2016). However, validity studies using structured diagnostic interviews with the patients are missing. It should be kept in mind that the AN group in the data set was determined by treatment seeking and referral to hospital-based services.
Given that not all patients with ED seek treatment, it can be assumed that a number of patients with AN may have contributed to a selection bias of unclear amount with an over-representation of comorbid disorders in the patient sample. Furthermore, all patients were treated in the public health system. There is no obligation to register data on patients treated in the rather small sector of privately practising physicians. Finally, there was an unknown but likely small number of individuals that were lost during follow-up due to emigration. Future studies may build from the present findings and explore and disentangle concurrent and sequential comorbidity, also focusing on the temporal stability and predictive validity of a comorbidity-based approach to ED classification.
In conclusion, the findings of the present study document the excessive burden of comorbid mental disorders imposed on patients suffering from AN. Clinically, both assessment and interventions of comorbid mental disorders should be integrated into a highly complex plan that goes beyond the mere consideration of AN itself. This multidimensionality of care is mandatory for preventing life-damaging courses of the disease in the afflicted patients.

ACKNOWLEDGMENTS
The study was funded by the Jascha Fund Denmark, A.P. Møller Fund Denmark, Psychiatric Research Fund of the Region of Southern Denmark. The sponsors had no role in the design, analysis, interpretation, or publication of this study.

CONFLICT OF INTEREST
In the past 3 years, HCS has worked as a speaker for Medice and has received book royalties from Cambridge University Press, Elsevier, Hogrefe, Klett, and Kohlhammer publishers. No other disclosures were reported.

DATA AVAILABILITY STATEMENT
Data were provided by Statistics Denmark in an anonymized fashion by request of the authors.