Abstract
Objective:
Eating disorders are common and have a high public health burden. However, existing clinically relevant data sources are scarce, limiting the capacity to accurately measure the burden of eating disorders. This study tests the feasibility of generating a large clinically relevant cohort of individuals with eating disorders using health administrative data.
Methods:
We developed 3 clinically relevant eating disorder prevalence cohorts using health administrative data from Ontario, Canada, between 1990 and 2014. Cohort 1 included patients with a hospitalization where an eating disorder diagnosis was the primary diagnosis, cohort 2 included patients with a hospitalization where an eating disorder diagnosis was any diagnosis, and cohort 3 included cohort 2 plus any patient with an emergency department visit with an eating disorder diagnosis.
Results:
Cohort 1 had 7268 patients, cohort 2 had 13,197 patients, and cohort 3 had 17,373 patients. As cohort size increased, the proportion of eating disorder patients with diagnoses of bulimia nervosa and eating disorder not otherwise specified increased. Although the cohorts differed according to demographic and clinical characteristics, these differences were small compared to the degree to which they differed from the Ontario population.
Discussion:
It is feasible to use health administrative data to measure the clinically relevant burden of eating disorders. The cohorts differed significantly in the eating disorder diagnostic composition. Eating disorders have a high burden, but poor data availability has resulted in fewer public health–related eating disorders studies in comparison to other mental disorders. The use of administrative data can address this evidence gap.
Keywords: eating disorders, epidemiology, disease surveillance, mental health
Abstract
Objectif :
Les troubles alimentaires sont communs et constituent un fardeau de santé publique élevé. Toutefois, les sources existantes de données pertinentes sur le plan clinique sont rares, et limitent ainsi la capacité de mesurer avec exactitude le fardeau des troubles alimentaires. Cette étude vérifie la faisabilité de générer une vaste cohorte pertinente sur le plan clinique de personnes souffrant de troubles alimentaires à l’aide des données de santé administratives.
Méthodes :
Nous avons formé trois cohortes ayant une prévalence de troubles alimentaires pertinente sur le plan clinique à partir des données de santé administratives de l’Ontario, au Canada, entre 1990 et 2014. La cohorte 1 comprenait des patients ayant eu une hospitalisation quand le diagnostic de trouble alimentaire était le principal diagnostic; la cohorte 2 comportait des patients ayant eu une hospitalisation quand un diagnostic de trouble alimentaire était un quelconque diagnostic; et la cohorte 3 regroupait des patients de la cohorte 2 en plus de patients qui comptaient une visite au service d’urgence avec un diagnostic de trouble alimentaire.
Résultats :
La cohorte 1 comportait 7 268 patients, la cohorte 2, 13 197 patients, et la cohorte 3, 17 373 patients. À mesure que s’accroissait la taille des cohortes, la proportion des patients des troubles alimentaires présentant des diagnostics de boulimie et de trouble alimentaire non spécifié augmentait également. Même si les cohortes différaient à l’égard des caractéristiques démographiques et cliniques, ces différences étaient minimes comparativement au degré auquel elles différaient de la population ontarienne.
Discussion :
Il est faisable d’utiliser les données de santé administratives pour mesurer le fardeau des troubles alimentaires pertinent sur le plan clinique. Les cohortes différaient significativement en ce qui concerne la composition des diagnostics de troubles alimentaires. Les troubles alimentaires constituent un fardeau élevé, mais la mauvaise disponibilité des données a fait en sorte que moins d’études des troubles alimentaires liées à la santé publique aient été menées en comparaison d’autres troubles mentaux. Le recours aux données administratives peut combler ces lacunes des données probantes.
Eating disorders, which include anorexia nervosa, bulimia nervosa, binge-eating disorder, and other specified feeding and eating disorders, are common psychiatric disorders, with a lifetime prevalence of approximately 7% in women and 3% in men in the United States.1,2 The most recent prevalence of eating disorders in Canada is of bulimia nervosa from nearly 25 years ago.3,4 Eating disorders can persist over many years, are often challenging to treat, and can be associated with significant medical and psychiatric comorbidity.5,6 Eating disorders are also associated with one of the highest mortality rates of any psychiatric disorder.7 Despite substantial literature examining aspects of epidemiology, comorbidity, and mortality in individuals with eating disorders, few studies are clinically relevant. Little work has been done looking at health care utilization, access, or costs in this group, and very little work has been done studying any of these issues in Canada.8 Moreover, the high burden of eating disorders is often overlooked by health care professionals and policy makers, who view eating disorders as rare and/or episodic, time-limited conditions.9,10 The public health impact of other mental disorders, such as depression,11 has been recognized through a concerted effort to generate high-quality data and information to facilitate careful burden measurement and evaluation. Based on existing evidence, there is interest in similar analyses focusing on the public health and mental health burden of eating disorders, but data are limited.12 The inability to measure the burden of eating disorders to the same degree as other mental disorders like depression has led to the false impression that these disorders are insignificant.13
Ontario, with 13.5 million people, is Canada’s most populated province. Routine administrative data, including diagnosis, are collected within Ontario’s universal health care system, and this presents an opportunity to develop clinically relevant cohorts of eating disorder patients to facilitate research on the burden of eating disorders. The development of clinically relevant cohorts within large health administrative data sets allows for the surveillance of the development of psychiatric and medical comorbidity in large, representative samples, as well as research on eating disorder–related mortality.
The objective of this study was to demonstrate the feasibility of generating clinically relevant cohorts of eating disorder patients using Ontario’s health administrative data. The first cohort includes hospitalized patients where the eating disorder diagnosis is the main diagnosis. The second cohort adds hospitalized patients where the eating disorder diagnosis is a secondary diagnosis to the first cohort. Finally, the third cohort adds patients who have had an emergency department visit with an eating disorder diagnosis to the second cohort. To our knowledge, this is the first study to systematically develop clinically relevant eating disorder cohorts. We hypothesized that the cohorts will differ such that the first cohort will represent the most specific cohort of eating disorder patients with the highest illness severity biased towards anorexia nervosa, with the progression of increased sensitivity and increased representation of eating disorders other than anorexia nervosa. The development of 3 different cohorts may provide the opportunity to use cohorts tailored to research questions based on the cohort characteristics detailed in our study.
Methods
Study Design and Setting
We developed 3 different clinically relevant eating disorder prevalence cohorts using health administrative data from Ontario, Canada, between 1990 and 2014. The 3 cohorts were defined as 1) a patient with a hospitalization where an eating disorder diagnosis was the primary diagnosis, 2) a patient with a hospitalization where an eating disorder diagnosis was any diagnosis, and 3) cohort 2 plus any patient with an emergency department visit associated with an eating disorder diagnosis. The cohorts were developed with decreasing specificity of diagnosis, treatment provided, and illness severity but increasing sensitivity. The first cohort only includes patients with a hospitalization diagnosis of eating disorder as the main diagnosis for that hospitalization. The second cohort includes hospitalizations where an eating disorder is any of multiple diagnoses provided for that hospitalization. Finally, the third cohort includes hospitalizations with eating disorder diagnoses as any diagnoses plus emergency department visits with eating disorders as diagnoses. A more detailed description of cohort development is described below. This study was approved by the Research Ethics Board at Sunnybrook Health Sciences Centre, Toronto, Canada.
Data
We used administrative health care records available through the Institute for Clinical Evaluative Sciences (ICES) in Toronto, Ontario. The ICES data repository includes individual-level, linkable, longitudinal data on most publicly funded health care services (and some related services outside health) for all Ontario residents eligible for universal health care insurance. Patient records were linked using unique, anonymized, encrypted identifiers developed using the health card number of each participant across multiple Ontario health administrative databases containing information on all publicly insured hospital and physician services. Eating disorder patients were identified using the Discharge Abstract Database (DAD) for non–mental health hospital admissions that includes the most responsible diagnosis (MRD) and other diagnoses, the Ontario Mental Health Reporting System (OMHRS) for all hospitalizations occurring in mental health–designated hospital beds that includes the MRD, the National Ambulatory Care Reporting System (NACRS) for emergency department visits, and the Ontario Drug Benefits data to determine, via prescriptions, whether individuals receive benefits based on financial need. Information about patient demographics and death were obtained from the Registered Persons Database (RPDB). Immigration status was acquired from the Immigration, Refugee, and Citizenship Canada database that describes immigration status. The Ontario Health Insurance Plan (OHIP) database contains information from physician billings and was used for comorbidity calculations (Johns Hopkins Adjusted Diagnostic Groups14) and for the development of medical comorbidity registries. Neighbourhood income was derived from Statistics Canada 2011 census estimates. These data sets were linked using unique encoded identifiers and analyzed at the ICES. The use of data in this project was authorized under section 45 of Ontario’s Personal Health Information Protection Act, which does not require review by a research ethics board.
Patients
We included all individuals eligible for public health care insurance in 2014 aged 10 years and older who were hospitalized with a main diagnosis of an eating disorder at any point since April 1990—in other words, all prevalent cases with an acute eating disorder. For cohort 1, we selected all patients who had a hospitalization with a diagnosis of anorexia nervosa (International Classification of Diseases, Ninth Revision [ICD-9] code 307.1; International Classification of Diseases, Tenth Revision [ICD-10] codes F50.0 and F50.1; and Diagnostic and Statistical Manual of Mental Disorders, 4th edition [DSM-IV] code 307.1), bulimia nervosa (ICD-9 code 307.51, ICD-10 codes F50.2 and F50.3, and DSM-IV code 307.51), and EDNOS (EDNOS) (ICD-9 code 307.50, ICD-10 codes F50.8 and F50.9, and DSM-IV code 307.50) as a main diagnosis. Eating disorder not otherwise specified is an eating disorder that does not meet the criteria for anorexia nervosa or bulimia nervosa. Individuals with EDNOS usually fall into 1 of 3 groups: subthreshold symptoms of anorexia or bulimia, mixed features of both disorders, or extremely atypical eating behaviors that are not characterized by either of the other established disorders.2 The psychiatric hospitalization data from OMHRS only had DSM-IV diagnoses at the time of cohort development. Cohort 2 included all patients in cohort 1 but also included patients who had the above eating disorder diagnoses as any (e.g., secondary) diagnoses. Finally, cohort 3 included all patients in cohorts 1 and 2 but also included any patients who had an emergency department visit with one of the above eating disorder diagnoses. We excluded all patients who were not eligible for provincial public health care insurance in 2014 and with missing information on age and/or sex. In addition, we excluded all individuals diagnosed with Prader-Willi syndrome (ICD-9 759.81 code; no specific ICD-10 code is currently available), a rare genetic disorder associated with excessive eating and obesity but not typically considered an eating disorder like anorexia nervosa or bulimia nervosa. Given data limitations, we were not able to select patients in ambulatory settings who would have less severe or subthreshold forms of eating disorders, due to the lack of a specific diagnostic code for these disorders within Ontario physician billings.
Measures
For each cohort, the following variables were included: demographics (age, sex, ethnicity, migrant status, neighbourhood income quintile, receipt of social assistance, and rural residence) and medical comorbidities as measured by specific medical conditions that have been validated at ICES and by Johns Hopkins Adjusted Diagnostic Groups (JH-ADGs).14 The JH-ADGs are a case-mix method developed by Johns Hopkins that has been used as a general measure of comorbidity and as an accurate predictor of mortality within the general population15 and in a clinically relevant sample of individuals with schizophrenia.16 For female patients, we measured whether they had delivered live children in our database to determine the proportion who were parents. All variables were calculated as of January 1, 2014, with the exception of neighbourhood income quintile, which was derived from the 2011 census and the JH-ADGs, which were derived based on a 2-year lookback of health service utilization prior to January 1, 2014. To provide clinically relevant context to our comparisons between cohorts, we also generated characteristics for the entire Ontario population as of January 1, 2014.
Analyses
Descriptive statistics to compare the cohorts included analysis of variance (ANOVA) for continuous variables and chi-square tests for categorical variables. All analyses used SAS Version 9.4 (SAS Institute, Cary, NC).
Results
Cohort 1, which was developed using hospitalizations with an eating disorder as the main diagnosis, had 7268 patients (Table 1). The cohort using hospitalizations with an eating disorder as any diagnosis (cohort 2) included all patients from cohort 1 plus 5929 patients with an eating disorder diagnosis as a secondary hospitalization diagnosis for a total of 13,197 patients. The cohort that added individuals who had emergency department visits related to eating disorders to cohort 2 (cohort 3) added 4176 patients who had an eating disorder–related emergency department visit for a total of 17,373 patients. These cohorts represented 0.05%, 0.1%, and 0.13% of the Ontario population aged 10 years and older.
Table 1.
Eating Disorder Diagnoses within Each Cohort.
| Characteristic | Cohort 1, n (%) | Cohort 2, n (%) | Cohort 3, n (%) | Chi-Square P Values |
|---|---|---|---|---|
| Anorexia nervosa only | 3550 (49.4) | 4750 (36.3) | 5305 (31.07) | <0.0001 |
| Bulimia nervosa only | 682 (9.5) | 1379 (10.5) | 2237 (13.1) | <0.0001 |
| Eating disorder NOS only | 1596 (22.2) | 3808 (29.1) | 6343 (37.2) | <0.0001 |
| Anorexia and bulimia | 59 (0.8) | 91 (0.7) | 119 (0.7) | 0.53 |
| Anorexia and eating disorder NOS | 40 (0.6) | 44 (0.3) | 53 (0.3) | 0.011 |
| Bulimia and eating disorder NOS | 1075 (14.9) | 2737 (20.9) | 2749 (16.1) | <0.0001 |
| Anorexia, bulimia, and eating disorder NOS | 188 (2.6) | 264 (2.0) | 267 (1.6) | <0.0001 |
NOS, not otherwise specified.
The 3 cohorts differed in terms of the distribution of eating disorder diagnosis categories (Table 1). Cohort 1 had the highest proportion of patients with only a diagnosis of anorexia nervosa (49.1%), and the lowest proportion of bulimia nervosa only (9.4%) and EDNOS only (22.4%). By contrast, cohorts 2 and 3 had a progressive decrease in the proportion of patients with only a diagnosis of anorexia nervosa, as well as a progressive increase in the proportion of patients with diagnoses of bulimia nervosa only and EDNOS only (Table 1). In cohorts 1 and 2, anorexia nervosa only was the most common diagnostic category, whereas in cohort 3, EDNOS was the most prevalent.
Demographic, clinical, and comorbidity characteristics are presented in Table 2. A similar proportion of eating disorder patients were female (90.7% to 92.8%) in all 3 cohorts, and the mean age of patients in each cohort was also similar, ranging from a mean of 30 to 33 years. The proportion of eating disorder patients of Chinese or South Asian descent or immigrants to Ontario was similar, but ethnicity and immigration were less common than in the Ontario population. Among female eating disorder patients, between 71.1% (cohort 2) and 75.2% (cohort 1) of patients had not had children, a proportion that is much higher than the Ontario population (25.7%). Cohorts 2 and 3 had similar neighbourhood income distributions as the Ontario population, whereas in cohort 1, 24.8% of the patients were in the highest income quintile versus 20.2% in the Ontario population. Cohorts 2 and 3 had a higher proportion of patients receiving social assistance (37.8% and 36.5%, respectively) than cohort 1 (31.5%) and a much higher proportion compared to the Ontario population (27.9%). Finally, a large proportion of patients resided in rural settings (approximately 10%), which is similar to the 11% of Ontario residents who reside in rural settings.
Table 2.
Demographic, Diagnostic, and Comorbidity Characteristics across Different Eating Disorder Cohorts.a
| Variable | Cohort 1 (n = 7190) | Cohort 2 (n = 13,0730 | Cohort 3 (n = 17,073) | Standardized Differenceb | Ontario Population (n = 12,615,820) |
|---|---|---|---|---|---|
| Sex | |||||
| Female | 6683 (92.9) | 12,068 (92.3) | 15,592 (91.3) | 0.98-1.05 | 6,421,436 (50.9) |
| Male | 507 (7.1) | 1005 (7.7) | 1481 (8.7) | 6,194,384 (49.1) | |
| Age, mean ± SD, y | 30.44 ± 13.15 | 33.43 ± 14.70 | 32.64 ± 14.83 | 0.48-0.66 | 44.22 ± 20.11 |
| Migrant status | |||||
| Immigrant | 239 (3.3) | 412 (3.2) | 630 (3.7) | 0.35-0.37 | 1,756,232 (13.9) |
| Refugee | 54 (0.8) | 92 (0.7) | 150 (0.9) | 0.13-0.14 | 329,946 (2.6) |
| Nonimmigrant | 6897 (95.9) | 12,569 (96.1) | 16,293 (95.4) | 0.38-0.41 | 10,529,642 (83.5) |
| Motherhood | |||||
| No | 5018 (75.1) | 8565 (71.0) | 11,220 (72.0) | 0.02-0.07 | 4,675,775 (72.8) |
| Yes | 1665 (24.9) | 3503 (29.0) | 4372 (28.0) | 1,745,661 (27.2) | |
| Neighbourhood income quintile | |||||
| 1 (low) | 1314 (18.3) | 2735 (20.9) | 3591 (21.0) | 0.02-0.11 | 2,399,776 (19.0) |
| 2 (medium-low) | 1224 (17.0) | 2395 (18.3) | 3171 (18.6) | 2,457,383 (19.5) | |
| 3 (medium) | 1306 (18.2) | 2406 (18.4) | 3137 (18.4) | 2,507,518 (19.9) | |
| 4 (medium-high) | 1536 (21.4) | 2641 (20.2) | 3438 (20.1) | 2,653,433 (21.0) | |
| 5 (high) | 1782 (24.8) | 2845 (21.8) | 3662 (21.4) | 2,550,409 (20.2) | |
| Missing | 28 (0.4) | 51 (0.4) | 74 (0.4) | 47,301 (0.4) | |
| Social assistancec | |||||
| No | 4916 (68.4) | 8115 (62.1) | 10,789 (63.2) | 0.08-0.21 | 8,965,515 (71.1) |
| Yes | 2274 (31.6) | 4958 (37.9) | 6284 (36.8) | 3,650,305 (28.9) | |
| Rural residence | |||||
| No | 6491 (90.3) | 11,775 (90.1) | 15,361 (90.0) | 0.03-0.04 | 11,212,297 (88.9) |
| Yes | 699 (9.7) | 1298 (9.9) | 1712 (10.0) | 1,403,523 (11.1) | |
| Clinical characteristics | |||||
| Age at first diagnosis | |||||
| Mean ± SD | 20.64 ± 10.02 | 23.70 ± 12.48 | 24.16 ± 13.00 | NA | — |
| Median (IQR) | 17 (15-23) | 18 (15-29) | 19 (15-29) | NA | — |
| Age group at first diagnosis | |||||
| 10-13 | 1033 (14.4) | 1288 (9.9) | 1523 (8.9) | NA | |
| 14-17 | 3094 (43.0) | 4766 (36.5) | 5766 (33.8) | ||
| 18-24 | 1467 (20.4) | 2699 (20.6) | 4122 (24.1) | ||
| 25-44 | 1301 (18.1) | 3316 (25.4) | 4226 (24.8) | ||
| 45-64 | 272 (3.8) | 835 (6.4) | 1123 (6.6) | ||
| >64 | 23 (0.3) | 169 (1.3) | 313 (1.8) | ||
| Comorbidity | |||||
| Arthritis | 67 (0.9) | 133 (1.0) | 169 (1.0) | 0.0-0.01 | 118,338 (0.9) |
| Asthma | 1972 (27.4) | 3734 (28.6) | 4880 (28.6) | 0.31-0.34 | 1,866,390 (14.8) |
| Cancer | 107 (1.5) | 308 (2.4) | 417 (2.4) | 0.12-0.19 | 628,994 (5.0) |
| Chronic kidney disease | ≤5 (0.0) | 12 (0.1) | 23 (0.1) | 0.01-0.03 | 6842 (0.1) |
| COPD | 335 (4.7) | 915 (7.0) | 1076 (6.3) | 0.01-0.09 | 896,795 (7.1) |
| Congestive heart failure | 64 (0.9) | 200 (1.5) | 262 (1.5) | 0.04-0.1 | 282,691 (2.2) |
| Crohn’s disease/colitis | 89 (1.2) | 196 (1.5) | 230 (1.3) | 0.06-0.08 | 84,051 (0.7) |
| Diabetes | 430 (6.0) | 1086 (8.3) | 1319 (7.7) | 0.07-0.16 | 1,389,260 (11.0) |
| HIV | — | 22 (0.2) | 30 (0.2) | 0.01-0.03 | 17,682 (0.1) |
| Hypertension | 648 (9.0) | 1560 (11.9) | 1949 (11.4) | 0.29-0.38 | 3,023,579 (24.0) |
| Myocardial infarction | 21 (0.3) | 64 (0.5) | 80 (0.5) | 0.10-0.12 | 191,452 (1.5) |
| Psychosocial ADGs | |||||
| ADG23—Psychosocial: Time Limited—Minor | 1177 (16.4) | 2445 (18.7) | 3219 (18.9) | 0.44-0.50 | 442,357 (3.5) |
| ADG24—Psychosocial: Recurrent or Persistent—Stable | 4811 (66.9) | 8840 (67.6) | 11,500 (67.4) | 1.07-1.08 | 2,616,695 (20.7) |
| ADG25—Psychosocial: Recurrent or Persistent—Unstable | 2037 (28.3) | 4261 (32.6) | 5466 (32.0) | 0.68-0.78 | 565,571 (4.5) |
| Nonpsychosocial ADGs | |||||
| ADG3—Time Limited: Major | 537 (7.5) | 1097 (8.4) | 1422 (8.3) | 0.12-0.15 | 616,594 (4.9) |
| ADG4—Time Limited: Major—Primary Infections | 783 (10.9) | 1545 (11.8) | 2050 (12.0) | 0.11-0.15 | 979,172 (7.8) |
| ADG9—Likely to Recur: Progressive | 159 (2.2) | 409 (3.1) | 498 (2.9) | 0.01-0.07 | 269,287 (2.1) |
| ADG11—Chronic Medical: Unstable | 1473 (20.5) | 2783 (21.3) | 3422 (20.0) | 0.14-0.17 | 1,957,102 (15.5) |
| ADG16—Chronic Specialty: Unstable— Orthopedic | 109 (1.5) | 257 (2.0) | 310 (1.8) | 0.0-0.02 | 200,609 (1.6) |
| ADG22—Injuries/Adverse Effects: Major | 1779 (24.7) | 3581 (27.4) | 4648 (27.2) | 0.26-0.32 | 1,836,022 (14.6) |
| ADG32—Malignancy | 309 (4.3) | 667 (5.1) | 848 (5.0) | 0.03-0.07 | 759,987 (6.0) |
ADG, Aggregate Diagnosis Group; COPD, chronic obstructive pulmonary disease; HIV, human immunodeficiency virus; IQR, interquartile range; SD, standard deviation.
aValues are presented as number (%) unless otherwise indicated.
bStandardized differences were calculated by comparing each of the cohorts to the general population. The range created by these 3 comparisons is presented in the table.
cBased on whether an outpatient drug prescription was filled.
The 3 cohorts differed in terms of clinical characteristics. The age of cohort entry was younger for cohort 1 than for cohorts 2 and 3 (Table 2). For example, 66.8% of cohort 1 subjects entered the cohort between ages 10 and 18 years, compared with 45.9% and 42.0% in the same age category in cohorts 2 and 3, respectively. Cohort 2 and 3 patients had higher rates of cancer, chronic obstructive pulmonary disease, congestive heart failure, diabetes, and hypertension than patients in cohort 1. Higher rates of medical comorbidities in the general population were observed compared to eating disorder patients, but the eating disorder cohorts were younger, making the comorbidity differences difficult to interpret. Table 2 also presents the proportions of cohort members who also fall into JH-ADGs. All 3 cohorts were similar to each other in terms of the prevalence of psychosocial ADGs, and each had far higher prevalence of psychosocial ADGs than the Ontario population. The 3 cohorts were also fairly similar to each other in terms of nonpsychosocial ADGs, but eating disorder patients were more likely than Ontario residents to fall into nonpsychosocial ADGs, particularly for time-limited conditions, infections, and injuries.
Discussion
We developed 3 different eating disorder cohorts using health administrative data from the province of Ontario, Canada. As hypothesized, the 3 cohorts differed with respect to eating disorder diagnoses, with the patients hospitalized with a primary diagnosis of eating disorder containing half of the patients diagnosed with anorexia nervosa. As the cohort was expanded to include more individuals who had a secondary eating disorder diagnosis via hospitalization or an emergency department visit, the cohort included a greater proportion of individuals with bulimia nervosa or EDNOS. The 3 eating disorder cohorts differed from one another according to demographic and clinical characteristics, but these differences were small compared to the degree to which all 3 eating disorder cohorts differed from the Ontario population.
The purpose of this study was to demonstrate the feasibility of generating a clinically relevant cohort of eating disorder patients using health administrative data. The 3 cohorts differed, but the first cohort differed substantially from cohorts 2 and 3 in terms of proportion of patients with a diagnosis of anorexia nervosa and age of eating disorder detection. The patient population who requires hospitalization for an eating disorder specifically is likely different from a population who has an identifiable eating disorder but does not have eating disorder–specific hospitalizations. While there may be a particular research or policy interest in individuals who are identified as having eating disorders as secondary hospitalization diagnoses, it is most likely that cohort 1 will be used to focus on individuals who require intensive hospitalization treatment for eating disorders, and cohort 3 will be used to study a more population-representative sample of individuals with eating disorders. Using cohort 3 versus cohort 2 will result in a larger sample size by including individuals who have not been hospitalized or, at the very least, have not had hospitalizations where an eating disorder diagnosis was recorded. Each cohort will have utility for future research on the burden and outcomes of eating disorders as well as the outcomes of different levels of specialization and intensity of health services use.
The importance of developing the capacity to evaluate the clinically relevant burden of eating disorders is underscored by the existing evidence of the mortality burden of these disorders,7 the poor existing data availability,12 and the effects of poor data availability, which results in the impact of these disorders being overlooked by both health care professionals and policy makers. The ability to accurately estimate eating disorder patients in administrative data means that a burden of illness, including cost, on the health care system can be estimated, which is valuable for policy makers. For clinicians, information from health administrative data can be used to evaluate quality of care and health outcome information. Indeed, the World Health Burden of Disease Initiative has identified eating disorders as disadvantaged relative to other mental illnesses based on poor availability of quantitative data.17 This is also true in Canada, with reports confirming that existing data provide a very limited profile of eating disorders and that priority data needs include incidence and prevalence, access and use of mental health services, treatment outcomes, and exposure to suspected risk and protective factors.18,19 If eating disorders impose a significant burden, it is imperative to achieve parity in the capacity to measure this burden in order to effectively intervene.
To our knowledge, this is the first study to test the feasibility of using health administrative data to generate clinically relevant samples of individuals with eating disorders. Prior studies have used data from the US veteran health care system to estimate eating disorder burden, including psychiatric comorbidity20,21; however, these studies were limited to hospitalization data and to US veterans, limiting the generalizability of the estimates. The strength of our study is the availability of a large Ontario population and different types of health service utilization (e.g., hospitalizations and emergency department visits) that are linkable to generate cohorts. There are several notable limitations. First, the eating disorder diagnoses we used to generate the cohorts have not been validated, and therefore the psychometric properties (e.g., sensitivity and specificity) are unknown. Ontario hospitalization and emergency department data have been validated to identify other types of disorders,22,23 including psychiatric disorders.24 In general, algorithms using health administrative data are highly specific but less sensitive. This is likely the case in our cohorts, where the prevalence estimates are much lower than the population estimates. It is likely that all the individuals within our cohort have an eating disorder, but there are many Ontario residents with eating disorders who are not represented in our cohorts through diagnostic inaccuracy, only being treated in outpatient settings, not seeking eating disorder treatment, or all of the above. A second, related limitation is the inability to use outpatient settings to capture individuals with eating disorders. Most individuals with eating disorders, like most disorders, will not require access to emergency departments or hospitalizations. Our inability to capture eating disorder patients via outpatient diagnostic codes means that the cohorts are an underestimate and biased towards a more severely ill population. Finally, using health services to capture a clinically relevant cohort requires individuals to use services. If there are access barriers to services such that people do not use services that would generate an eating disorder diagnosis, this will add to the underestimate of eating disorder prevalence.25
Our study demonstrates that health administration data can be used to generate clinically relevant cohorts of eating disorder patients. We have further demonstrated that different algorithms generate different populations of interest, with a population of likely severely ill patients, many of whom have a diagnosis of anorexia nervosa, who require specialty hospitalization, and a larger population of individuals who have an eating disorder that is identified during the course of non–eating disorder–specific health service utilization. Despite limitations, the formation of these cohorts will lay the groundwork to continue to build and improve on development of eating disorder cohorts for health services research. Other disease registries based on administrative data have been enhanced over time through addition of ambulatory data and supplemental diagnostic information incorporated through subsequent partnerships. These different cohorts can be used to study eating disorders within Ontario, a large, representative, and diverse setting. Since there is a paucity of evidence generated from large samples of eating disorder patients, the utility of the cohorts we have developed outweighs the limitations in terms of generalizability.
Acknowledgements
This study was supported by the Institute for Clinical Evaluative Sciences (ICES), which is funded by an annual grant from the Ontario Ministry of Health and Long-Term Care (MOHLTC). The opinions, results, and conclusions reported in this paper are those of the authors and are independent from the funding sources. No endorsement by ICES or the Ontario MOHLTC is intended or should be inferred. Parts of this material are based on data and/or information compiled and provided by the Canadian Institute for Health Information (CIHI). However, the analyses, conclusions, opinions, and statements expressed in the material are those of the author(s) and not necessarily those of CIHI.
Footnotes
Data Access: The data set from this study is held securely in coded form at the Institute for Clinical Evaluative Sciences (ICES). While data-sharing agreements prohibit ICES from making the data set publicly available, access may be granted to those who meet prespecified criteria for confidential access, available at www.ices.on.ca/DAS. The full data set creation plan and underlying analytic code are available from the authors upon request, understanding that the computer programs may rely upon coding templates or macros that are unique to ICES and are therefore either inaccessible or may require modification.
Declaration of Conflicting Interests: The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding: The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was funded by a grant from the University of Toronto Department of Psychiatry Labatt Family Innovation Fund in Brain Health.
References
- 1. Hudson JI, Hiripi E, Pope HG, Jr, et al. The prevalence and correlates of eating disorders in the National Comorbidity Survey Replication. Biol Psychiatry. 2007;61(3):348–358. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. American Psychiatric Association. Diagnostic and statistical manual of mental disorders 4th ed Washington, DC: American Psychiatric Association; 2000. [Google Scholar]
- 3. Garfinkel PE, Lin E, Goering P, et al. Purging and nonpurging forms of bulimia nervosa in a community sample. Int J Eat Disord. 1996;20(3):231–238. [DOI] [PubMed] [Google Scholar]
- 4. Garfinkel PE, Lin E, Goering P, et al. Bulimia nervosa in a Canadian community sample: prevalence and comparison of subgroups. Am J Psychiatry. 1995;152(7):1052–1058. [DOI] [PubMed] [Google Scholar]
- 5. Mehler PS. Clinical practice: bulimia nervosa. N Engl J Med. 2003;349(9):875–881. [DOI] [PubMed] [Google Scholar]
- 6. Yager J, Andersen AE. Clinical practice: anorexia nervosa. N Engl J Med. 2005;353(14):1481–1488. [DOI] [PubMed] [Google Scholar]
- 7. Arcelus J, Mitchell AJ, Wales J, et al. Mortality rates in patients with anorexia nervosa and other eating disorders: a meta-analysis of 36 studies. Arch Gen Psychiatry. 2011;68(7):724–731. [DOI] [PubMed] [Google Scholar]
- 8. Su JC, Birmingham CL. Anorexia nervosa: the cost of long-term disability. Eat Weight Disord. 2003;8(1):76–79. [DOI] [PubMed] [Google Scholar]
- 9. Simon J, Schmidt U, Pilling S. The health service use and cost of eating disorders. Psychol Med. 2005;35(11):1543–1551. [DOI] [PubMed] [Google Scholar]
- 10. Striegel-Moore RH. Health services research in anorexia nervosa. Int J Eat Disord. 2005;37(suppl):S31–S34; discussion S41-S42. [DOI] [PubMed] [Google Scholar]
- 11. Baxter AJ, Vos T, Scott KM, et al. The global burden of anxiety disorders in 2010. Psychol Med. 2014;44(11):2363–2374. [DOI] [PubMed] [Google Scholar]
- 12. Ventura A, Baxter A, Pham A, et al. Summary of data collected and decision rules used in making regional and global estimates: eating disorders Brisbane, Australia: Policy and Economics Group, Queensland Centre for Mental Health Research, University of Queensland; 2008. [Google Scholar]
- 13. Murray CJL, Lopez AD. Evidence-based health policy—lessons from the global burden of disease study. Science. 1996;274(5288):740–743. [DOI] [PubMed] [Google Scholar]
- 14. Weiner JP. The Johns Hopkins ACG Case-Mix System Version 6.0 Release Notes Johns Hopkins University, MD: Johns Hopkins University Bloomberg School of Public Health; 2003. [Google Scholar]
- 15. Austin PC, van Walraven C, Wodchis WP, et al. Using the Johns Hopkins Aggregated Diagnosis Groups (ADGs) to predict mortality in a general adult population cohort in Ontario, Canada. Med Care. 2011;49(10):932–939. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Austin PC, Newman A, Kurdyak PA. Using the Johns Hopkins Aggregated Diagnosis Groups (ADGs) to predict mortality in a population-based cohort of adults with schizophrenia in Ontario, Canada. Psychiatry Res. 2012;196(1):32–37. [DOI] [PubMed] [Google Scholar]
- 17. Mathers CD, Loncar D. Projections of global mortality and burden of disease from 2002 to 2030. PLoS Med. 2006;3(11):e442. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Government of Canada. The human face of mental health and mental illness in Canada Ottawa, Canada: Minister of Public Works and Government Services Canada; 2006. [Google Scholar]
- 19. Health Canada. A report on mental illnesses in Canada Ottawa: Health Canada; 2002. [Google Scholar]
- 20. Striegel-Moore RH, Garvin V, Dohm FA, et al. Eating disorders in a national sample of hospitalized female and male veterans: detection rates and psychiatric comorbidity. Int J Eat Disord. 1999;25(4):405–414. [DOI] [PubMed] [Google Scholar]
- 21. Striegel-Moore RH, Garvin V, Dohm FA, et al. Psychiatric comorbidity of eating disorders in men: a national study of hospitalized veterans. Int J Eat Disord. 1999;25(4):399–404. [DOI] [PubMed] [Google Scholar]
- 22. Hux JE, Ivis F, Flintoft V, et al. Diabetes in Ontario: determination of prevalence and incidence using a validated administrative data algorithm. Diabetes Care. 2002;25(3):512–516. [DOI] [PubMed] [Google Scholar]
- 23. Tu K, Campbell NR, Chen ZL, et al. Accuracy of administrative databases in identifying patients with hypertension. Open Med. 2007;1(1):e18–e26. [PMC free article] [PubMed] [Google Scholar]
- 24. Kurdyak P, Lin E, Green D, et al. Validation of population-based algorithm to detect chronic psychotic illness. Can J Psychiatry. 2015;60(8):362–368. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Hart LM, Granillo MT, Jorm AF, et al. Unmet need for treatment in the eating disorders: a systematic review of eating disorder specific treatment seeking among community cases. Clin Psychol Rev. 2011;31(5):727–735. [DOI] [PubMed] [Google Scholar]