Introduction
Anorexia nervosa (AN) is a severe psychiatric disorder characterized by low body weight, severe food restriction, and cognitive distortions related to eating, weight, and shape. Despite some advances in treatment, the prognosis for many individuals with AN remains poor. End-of-treatment remission rates vary between 23 and 33% for adolescents and 0–25% for adults [Reference Murray, Loeb and Le Grange1]. Approximately 40% of individuals with AN achieve recovery or remission, but relapse is also common [Reference Miskovic-Wheatley, Bryant, Ong, Vatter and Le2, Reference Fichter, Quadflieg, Crosby and Koch3].The impact of AN extends beyond the immediate health consequences, as individuals with AN experience high levels of disability, face unemployment, and frequently receive public assistance [Reference Streatfeild, Hickson, Austin, Hutcheson, Kandel and Lampert4]. AN is associated with significantly increased mortality [Reference van Eeden and van Hoeken5], with a recent Danish study reported a threefold increase in overall mortality rates [Reference Larsen, Yilmaz, Bulik, Albiñana, Vilhjálmsson and Mortensen6]. The most common causes of death are medical complications related to AN and suicide [Reference Larsen, Yilmaz, Bulik, Albiñana, Vilhjálmsson and Mortensen6, Reference Arcelus, Mitchell, Wales and Nielsen7].
Although severity is a critical issue, there is no consensus definition of severe and enduring AN (SE-AN) [Reference Wonderlich, Bulik, Schmidt, Steiger and Hoek8]. The current severity criteria in DSM-5 focus solely on body mass index (BMI) [9]. However, BMI has not been found to be a strong predictor of poor outcome or involuntary treatment for AN [Reference Arcelus, Mitchell, Wales and Nielsen7, Reference Clausen and Jones10]. A systematic review found that illness duration and the number of previously failed treatment attempts are the most prevalent criteria across proposed SE-AN definitions [Reference Broomfield, Stedal, Touyz and Rhodes11]. Classification guidelines by Hay and Touyz define SE-AN by unrelenting symptoms including dysfunctional behaviors and cognitions associated with AN leading to persistent low body weight, an illness duration of a minimum of 3 years, and exposure to at least two evidence-based treatments for AN [Reference Hay and Touyz12]. Another review found a lack of consistency in the identification of patients with SE-AN, but one of the most commonly used criteria was an illness duration of at least 7 years [Reference Ramsay, Allison, Temples, Boccuto and Sarasua13]. However, the duration criterion is debated due to the lack of consensus on defining chronic AN [Reference Tierney and Fox14] and limited evidence for staging the illness solely by duration [Reference Fernández-Aranda, Treasure, Paslakis, Agüera, Giménez and Granero15–Reference Maguire, Touyz, Surgenor, Crosby, Engel and Lacey18]. Moreover, greater levels of illness severity and duration may not necessarily be associated with poorer treatment outcomes [Reference Raykos, Erceg-Hurn, McEvoy, Fursland and Waller19].
Denmark is uniquely positioned to study the epidemiology of AN thanks to its population registers. The use of population based samples minimizes selection bias and accurately represents cases as detected throughout the national healthcare system. This allows for robust longitudinal studies, which are essential for understanding the progression and outcomes of AN [Reference Larsen, Yilmaz, Bulik, Albiñana, Vilhjálmsson and Mortensen6, Reference Chatwin, Holde, Yilmaz, Larsen, Albinana and Vilhjalmsson20, Reference Momen, Plana-Ripoll, Yilmaz, Thornton, McGrath and Bulik21]. Despite the advantages of these registers, detailed clinical records are currently not readily available and have only recently become electronic. Given the high morbidity and mortality associated with AN and the lack of a consensus definition of SE-AN, a practical and standardized method to capture AN severity using register-based data could prove valuable. In addition to a binary classification system, access to rich health data from Danish registers provides a unique opportunity to develop a more comprehensive and nuanced continuous measure of severity while incorporating some of the key aspects of the classification guidelines.
The aim of this study was to construct a register-based severity index for AN, which could facilitate the identification of severe AN cases in Denmark and countries with similar data sources and contribute to a better understanding of the etiology and course of AN. The secondary aim of this study was to evaluate the associations between the severity index and overall and cause-specific mortality in individuals with AN.
Methods
This study utilizes nationwide registers to identify all Danish individuals diagnosed with AN (ICD-8: 306.50; ICD-10: F50.0, F50.1) between January 1, 1969 and December 31, 2013. The Supplementary Material provides detailed descriptions of the data sources and study population, along with comprehensive explanations of the methods, results, and discussion pertaining to sensitivity analyses.
Anorexia nervosa register-based severity index
AN severity was measured using an anorexia nervosa register-based severity index (AN-RSI), calculated as the sum of points assigned to each of the included severity-related variables available in the Danish registers. These six variables (described next) were identified by previous studies and are operationalizable using existing register data. The points and weights were assigned with input from experienced clinicians. The AN-RSI score was calculated 5 years after first diagnosis. Given the lack of consensus in the literature concerning duration, with studies often using follow-up periods of 3 or 7 years [Reference Hay and Touyz12, Reference Ramsay, Allison, Temples, Boccuto and Sarasua13], we chose the intermediate period of 5 years to balance the need for sufficient data to capture illness progression and practical considerations of duration of follow-up and potential attrition. Overall, the aim was to provide a measure for AN severity that is robust, practical, and clinically relevant.
The six severity-related variables included in the AN-RSI were:
-
1. Early onset was defined as age below 10 years at start of first in- or outpatient hospital treatment for AN to ensure prepubertal onset (national data indicate the mean age of initiation of puberty among Danish girls is 10.88 years [Reference Juul, Teilmann, Scheike, Hertel, Holm and Laursen22]). Assigned 1 point.
-
2. Late onset was defined as age above 25 years at start of first in- or outpatient hospital treatment for AN to ensure adult onset. Assigned 1 point.
-
3. Inpatient admissions were defined as the number of recorded inpatient hospital admissions for AN. Assigned 2 points per admission.
-
4. Outpatient treatments were defined as the number of recorded outpatient treatment courses for AN, disregarding the first treatment course. Assigned 1 point per treatment.
-
5. Treatment length was defined as the combined lengths of inpatient and outpatient treatment. The length of an inpatient admission was defined as the difference between start and end dates. The length of an outpatient treatment was defined as the sum of the difference between start and end dates and the recorded number of treatment days (days with an appointment) during the outpatient treatment course, both multiplied by 0.5. Assigned 1 point per full year.
-
6. Illness duration was defined as time between first diagnosis and last treatment day occurring within 5 years postdiagnosis. Assigned 1 point per full year.
Outcomes
The main outcome was mortality from any cause. Secondary outcomes were mortality from eight specified causes of death: AN, other eating disorders, somatic anorexia diagnosis, suicide, accident or intention unknown, alcohol-related causes, other specified psychiatric diagnoses, and other causes of death. See the Supplementary Material for definitions.
Statistical analysis
We conducted survival analyses using Cox regression resulting in hazard ratios (HR) and 95% confidence intervals (CI) for associations between AN-RSI and mortality, adjusted for time since first diagnosis as the continuous underlying time scale, and calendar year of first diagnosis and birth year in 10-year strata.
Follow-up started 5 years after the first diagnosis, at which point AN-RSI was also evaluated, and ended at death, emigration, or on December 31, 2018, whichever came first.
The results of AN-RSI were evaluated in two ways: (1) the continuous index score resulting in HR per one point increase in AN-RSI; and (2) severe AN cases – defined as individuals who scored in the top 20% of AN-RSI scores 5 years after initial diagnosis – compared to less severe AN cases in the bottom 80% of AN-RSI. This cut-off is in line with previous studies showing that approximately 20% of individuals with AN do not improve with treatment [Reference Dalle Grave23].
We calculated the proportion of severe versus less severe AN cases receiving any in- or outpatient treatment for AN annually for up to 10 years postdiagnosis. Additionally, we assessed the prevalence of diagnosed psychiatric comorbidities prior to first AN diagnosis, during the severity evaluation period (0–5 years postdiagnosis), and in the subsequent 5 years (5–10 years postdiagnosis) for severe and less severe cases. Details on the definitions and groupings of the included psychiatric conditions can be found elsewhere [Reference Momen, Plana-Ripoll, Yilmaz, Thornton, McGrath and Bulik21]. Statistically significant differences between severe and less severe AN cases were evaluated using Pearson’s chi-squared tests.
Results
A total of 9167 individuals (8592 [93.73%] females, 575 [6.27%] males) were included in the main analysis. The distribution of severity variables contributing to the combined AN-RSI is shown in Figure 1. The main contributors were number of inpatient admissions, number of outpatient treatments, and illness duration.
Figure 1. Distribution of severity variables contributing to the combined anorexia nervosa register-based severity index evaluated 5 years after onset.
Based on AN-RSI, 2036 (22.21%) individuals were classified as severe AN cases (scoring 6 points or more) 5 years after onset, of whom 96 (4.72%) were male. Figure 2 shows the proportion of those still followed in the survival analysis who are in treatment annually up to 10 years postdiagnosis (not counting the date of onset in the first year) for severe and less severe AN cases, respectively. We found that a larger proportion of severe AN cases were in treatment during the first 5 years, as would be expected based on the construction of the AN-RSI. Importantly, severe AN cases were also more likely to be in treatment in the next 5 years after they had attained the threshold for being considered severe. Table 1 shows the number of diagnosed psychiatric conditions during each time period for severe and less severe AN cases, while Table 2 shows the proportions diagnosed with each psychiatric condition. Severe AN cases also had significantly higher prevalence of comorbid psychiatric conditions during and after the severity evaluation period compared to less severe cases, with the exception of anxiety disorders and other eating disorders having a higher prevalence in severe AN cases also prior to first AN diagnosis.
Figure 2. Severe and less-severe anorexia nervosa cases (evaluated 5 years after onset) in treatment during each year since onset, calculated as the proportion of those still followed in the survival analysis.
Table 1. Number of comorbid psychiatric conditions diagnosed prior to AN diagnostic onset; during severity evaluation (years 0–5 after AN onset); and during the first 5 years after severity evaluation (i.e., years 5–10 after AN onset) for severe and less severe AN cases
Table 2. Psychiatric conditions diagnosed prior to AN onset, during severity evaluation (years 0–5 after AN onset), and the first 5 years after severity evaluation (i.e., years 5–10 after AN onset) for severe and less severe AN cases
We observed 132 deaths during the follow-up period. Of the AN cases classified as severe, 46 (2.26%) cases died, while 86 (1.21%) of the less severe AN cases died. Figures 3 and 4 show the associations between AN-RSI and overall and cause-specific mortality per 1 point increase in AN-RSI score and for severe compared to less severe AN cases, respectively. In addition to mortality from any cause (HR = 1.03 [95% CI: 1.02, 1.04]), causes of mortality most strongly associated with increase in AN-RSI score were AN (1.07 [1.04, 1.10]), somatic anorexia (1.07 [1.02, 1.11]), other psychiatric disorders (1.07 [1.03, 1.11]), and suicide (1.02 [1.00, 1.04]). The only outcome negatively associated with increase in AN-RSI score was mortality from other eating disorders, though these results were based on few deaths (n = 6) and not statistically significant (0.95 [0.79, 1.14]). In the categorical analysis, the HR of mortality from any cause was 1.92 (1.34, 2.75) for severe AN cases compared to less severe cases, and being classified as a severe case was significantly associated with mortality from AN (6.74 [2.49, 18.24]), anorexia (somatic diagnosis) (4.62 [1.23, 17.33]), and alcohol-related causes (3.58 [1.03, 12.44]).
Figure 3. Hazard ratios (HR) with 95% confidence intervals (CI) for overall and cause-specific mortality per 1 point increase of the anorexia nervosa register-based severity index evaluated 5 years after onset.
Figure 4. Hazard ratios (HR) with 95% confidence intervals (CI) for overall and cause-specific mortality for severe anorexia nervosa cases compared to less-severe anorexia nervosa cases evaluated 5 years after onset.
Discussion
This study is the first to construct a register-based severity index as a proxy for AN severity in the absence of detailed clinical records. The development of AN-RSI addresses a significant gap in the field, especially given the limited treatment options and inconsistency in treatment approaches for the potentially fatal SE-AN [Reference Wonderlich, Bulik, Schmidt, Steiger and Hoek8, Reference Wonderlich, Dodd, Sondag, Jorgensen, Blumhardt and Evanson24]. Our findings indicate that both higher AN-RSI scores and the classification of severe AN cases were associated with overall mortality. However, the specific causes of mortality significantly associated with severity varied depending on whether the continuous AN-RSI or the categorical classification of AN cases was used. The continuous AN-RSI was associated with increased mortality from AN, somatic anorexia, other psychiatric disorders, and suicide. In contrast, classification of severe AN cases was not significantly associated with suicide or other psychiatric disorders. Instead, severe cases had higher rates of death from AN, somatic anorexia, and alcohol-related causes. These distinctions highlight the importance of considering both a continuous measure of AN severity and a binary severity classification when assessing prognoses and outcomes. That these specific causes of mortality are associated with AN-RSI suggests that AN severity is closely linked to underlying mental health problems and behaviors directly related to the disorder, and that severe AN may exacerbate vulnerabilities to other psychiatric disorders, alcohol use, and suicidal ideation, which in turn increase mortality risk. These findings highlight the need for effective treatment approaches that address not only AN itself but also a broader spectrum of mental health problems and behaviors associated with severe AN. Notably, death from other eating disorders was not associated with AN severity, suggesting that the factors driving mortality in severe AN are distinct and do not apply to individuals with AN who diagnostically transition to another eating disorder.
AN cases classified as severe based on AN-RSI 5 years after first diagnosis continued to exhibit higher rates of any type of treatment for AN in the 5 years after they had attained the threshold for being considered severe. Although the construction of the AN-RSI did not specifically focus on long-standing AN, as ongoing AN diagnoses in the registers during or after the 5-year severity establishment period were not required, the sustained need for treatment underscores the chronic nature of severe AN. Combined with the increased prevalence of psychiatric comorbidities before and after onset of AN, this finding shows that AN-RSI captures several important aspects of illness course.
Of note, AN-RSI captured a higher psychiatric comorbidity burden in individuals with severe versus less severe AN. This further reflects the robustness of the AN-RSI as comorbidity has often been associated with severity despite not being part of the primary definition [Reference Wildes, Forbush, Hagan, Marcus, Attia and Gianini25, Reference Hay and Touyz26]. More specifically, we observed higher prevalence of comorbid psychiatric conditions after AN diagnosis, before and after the period in which we assessed severity. One possible explanation is that contact with the healthcare system for AN diagnosis may have resulted in assessment and diagnosis of other psychiatric disorders [Reference Berkson27]. Overall, all psychiatric diagnoses – grouped based on ICD chapters – were more commonly diagnosed in severe cases compared to less severe cases. However, only anxiety and other eating disorders were more prevalent in severe AN cases prior to first AN diagnosis, which reflects the large body of research on AN risk factors [Reference Wildes, Forbush, Hagan, Marcus, Attia and Gianini25, Reference Bulik, Sullivan, Fear and Joyce28–Reference Schaumberg, Jangmo, Thornton, Birgegard, Almqvist and Norring33]. Taken together, our findings highlight the important relationship between AN-based severity and the higher burden of comorbid psychiatric disorders.
Strengths and limitations
Our study has several strengths that enhance the robustness and reliability of our findings. First, the use of a large cohort provides substantial statistical power and generalizability to our results. The comprehensive registers offer rich and detailed information, allowing for thorough analysis and accurate classification of AN severity. Additionally, the reliance on register data minimizes recall bias. Furthermore, validity of register-based diagnoses in Denmark has been established for a multitude of psychiatric disorders [Reference Nissen, Powell, Koch, Crowley, Matthiesen and Grice34–Reference Bock, Bukh, Vinberg, Gether and Kessing36], including eating disorders [Reference Egedal, Støving, Lynggaard, Laursen, Vinholt and Hansen37]. Eating disorder diagnoses have also been validated in the similar Swedish health registers [Reference Birgegard, Forsen Mantilla, Dinkler, Hedlund, Savva and Larsson38]. Furthermore, the public healthcare system in Denmark reduces selection bias and enhance the applicability of our findings to similar healthcare settings.
Our study also has several limitations that should be acknowledged. First, we relied solely on hospital diagnoses and treatments, which may not capture the full spectrum of AN severity, though severe AN cases are much more likely to be captured by the hospital-based diagnoses. The initial hospital diagnosis may also not accurately capture the actual time of onset. Second, the weighting of severity variables contributing to the combined AN-RSI was based on clinical judgment, which introduces a degree of subjectivity despite their grounding in the clinical research literature. Additionally, there may be differences in clinicians’ practices regarding use of primary diagnosis for treatment and cause of death, potentially affecting the consistency of our data. For instance, AN cases with personality disorders may have the latter listed as the primary diagnosis for treatment targeting AN since it may be viewed as pervasive; similarly, any somatic reason for treatment directly caused by AN may be entered as the primary diagnosis by a clinician. It is also possible that some deaths from suicide are recorded as accidents, cases of unknown intent, or attributed to other causes due to lack of evidence of intent or a desire to avoid the stigma associated with suicide. As we summarized earlier, we believe the instances of somatic anorexia as a cause of death may have been due to differences in language-specific diagnostic terminology, and this anomaly would explain the significant associations we observed with somatic anorexia and AN severity as well as severe AN. Moreover, we originally intended to include a cause-of-death category of somatic diagnoses potentially attributable to AN, but the number was too low to warrant its own category. The construction of the AN-RSI was based exclusively on AN-related measures, without incorporating other clinically relevant data such as BMI, symptom-level data, self-reported functionality and quality-of-life measures, or questionnaires, which are not available in the registers. The lack of comprehensive clinical data may affect the precision of our severity measures and the generalizability of our results to other settings. Finally, since AN-RSI relies heavily on treatment-seeking variables (i.e., inpatient admissions, outpatient readmissions, and treatment length), we were unable to capture individuals (especially severe AN cases) who are not in contact with the healthcare system or choose not to seek treatment. However, illness duration was included as a variable to account for nontreatment-seeking individuals as best as we can using register data, which were also at least partially captured by nonpsychiatric healthcare contacts that may be associated with AN-related complications or medical stabilization.
Conclusion
Our study demonstrates that AN-RSI is associated with comorbidity and mortality, and therefore, shows promise in its use and applications in register-based research. AN-RSI provides a method to quantify and measure AN severity in the absence of detailed clinical data, which is crucial given the limited knowledge about the biological and epidemiological risk factors associated with severe AN. Beyond the relevance of AN-RSI for other register-based studies, the results highlight the critical need for clinical attention to psychiatric comorbidity and suicidal ideation. Moreover, the use of a 5-year duration as a marker for severe AN warrants caution in categorizing a case as severe too early. Instead, the findings emphasize the importance of sustained intervention, even in cases with prolonged illness duration. Future studies should examine other correlates such as somatic comorbidities, validate AN-RSI in other datasets, add genetic indices such as polygenic scores, and explore how well it correlates with other severity indicators such as BMI and eating disorder symptoms. It is our expectation that AN-RSI can serve as a tool to examine epidemiological and genetic risk factors associated with the course and outcomes of AN.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1192/j.eurpsy.2025.10059.
Data availability statement
Access to individual-level Danish register data is regulated by Danish authorities due to its sensitive nature. Data can only be accessed via secure servers that prohibit the downloading of individual-level information. Each research project must receive prior approval from the relevant governing bodies, and researchers must be affiliated with a Danish research institution to obtain access.
Acknowledgements
We gratefully acknowledge the funding agencies whose support made this research possible, as well as the individuals in Denmark whose data, accessed through national registers, formed the foundation of this study.
Financial support
This work was funded by the Independent Research Fund Denmark (DFF) Sapere Aude (Z.Y., 1052-00029B) and the US National Institute of Mental Health (NIMH) (Z.Y., R01MH136156). Z.Y. acknowledges grant support from DFF (3166-00063B, 4309-00050B) and Lundbeck Foundation Ascending Investigator (R434-2023-269). L.V.P. acknowledges funding from Novo Nordisk Foundation Data Science Investigator – Ascending (NNF23OC0085941). L.C. acknowledges grant support from DFF (2096-00139B) and Danish Regions (R260-A6068). C.M.B. acknowledges support from the NIMH (R56MH129437, R01MH120170, R01MH119084, R01MH118278, R01MH124871); Swedish Research Council (Vetenskapsrådet, 538-2013-8864); and the Lundbeck Foundation (R276-2018-4581). N.M. acknowledges funding from the Novo Nordisk Foundation Laureate Grant Award (NNF22OC0071010).
Competing interests
C.M.B. is an author and royalty recipient from Pearson Education, Inc. All other authors declare no competing interests.
Open access
Comments
No Comments have been published for this article.