Suicide Attempts in Women with Eating Disorders

Abstract

We evaluated whether the prevalence of lifetime suicide attempts/completions was higher in women with a lifetime history of an eating disorder than in women with no eating disorder and assessed whether eating disorder features, comorbid psychopathology, and personality characteristics were associated with attempts in women with anorexia nervosa, restricting subtype (ANR), anorexia nervosa, binge-purge subtype (ANBP), lifetime history of both anorexia nervosa and bulimia nervosa (ANBN), bulimia nervosa (BN), binge eating disorder (BED), and purging disorder (PD). Participants comprised the Swedish Twin study of Adults: Genes and Environment (n = 13,035) cohort. Lifetime suicide attempts were identified using diagnoses from the Swedish National Patient and Cause of Death Registers. General linear models were applied to evaluate whether eating disorder category [ANR, ANBP, ANBN, BN, BED, PD, or no eating disorder (no ED)] was associated with attempts, and to identify factors associated with attempts. Relative to women with no ED, lifetime suicide attempts were significantly more common in women with all types of eating disorder. None of the eating disorder features or personality variables was significantly associated with attempts. In the ANBP and ANBN groups, the prevalence of comorbid psychiatric conditions was higher in individuals with than without a lifetime suicide attempt. The odds of suicide were highest in presentations that included purging behavior (ANBN, ANBN, BN, and PD), but were elevated in all eating disorders. To improve outcomes and decrease mortality, it is critical to be vigilant for suicide and identify indices for those who are at greatest risk.

Introduction

Eating disorders are serious mental illnesses that occur in 1–5% of women (Hudson, Hiripi, Pope, & Kessler, 2007) and can have poor long-term outcome (Berkman et al., 2006; Hudson et al., 2007). A comprehensive meta-analysis of mortality in eating disorders including 36 studies reported standardized mortality ratios (SMR) of 5.86 (95% CI [4.17, 8.26]) for anorexia nervosa (AN) and 1.93 (95% CI [1.44, 2.59]) for bulimia nervosa [BN; (Arcelus, Mitchell, Wales, & Nielsen, 2011)], suggesting that individuals with either disorder are at increased risk of death compared with their age and gender matched peers.

Suicide is a common cause of this elevated mortality in eating disorders. A meta-analysis by Preti and colleagues (2011) analyzed data from 40 studies comprising 16,342 patients with AN followed over a mean of 11.1 years, and yielded an SMR = 31.0. A companion analysis of 16 studies on BN comprising 1768 patients, with a mean follow-up of 7.5 years, yielded an SMR = 7.5 (Preti, Rocchi, Sisti, Camboni, & Miotto, 2011). Only three studies examining suicide in individuals with binge eating disorder (BED) were available for inclusion in this meta-analysis. With only 246 patients, no completed suicides were identified and an SMR was unable to be calculated. Purging disorder (PD), characterized by purging in the absence of binge eating behavior is described in depth by Keel and Striegel-Moore (2009) and is included as a named condition in the DSM-5 in the Other Specified Feeding and Eating Disorders section (American Psychiatric Association, 2013). To our knowledge, there are no published data on completed suicides in individuals with PD.

Suicide attempts in individuals with AN are also common, with estimates of lifetime attempts ranging from 3.0% to 29.7% (Bulik et al., 2008; Forcano et al., 2011; Franko & Keel, 2006). Further, these attempts are often serious and associated with the intention to die. In a sample of 432 non-treatment seeking participants with AN, of those who had attempted suicide, 78.3% wanted to die from their attempt(s), and 56.5% thought that they would die as a result of the attempt(s). Over half of these attempts required medical attention (Bulik et al., 2008). In another study of a treatment-seeking sample, 79% of those who had attempted suicide endorsed “moderate or severe intent” to die (Bulik, Sullivan, & Joyce, 1999).

Suicide attempts are also common in BN, with between 15% and 40% of individuals indicating a lifetime history of at least one suicide attempt (Bulik et al., 1999; Corcos et al., 2002; Favaro & Santonastaso, 1997; Forcano et al., 2009; Franko & Keel, 2006; Milos, Spindler, Hepp, & Schnyder, 2004). Of individuals with BN who attempt suicide, 34.1% had a “serious” or “extreme” first attempt; the proportion of “serious” or “extreme” attempts increased with the number of attempts. Over 60% of individuals with BN were hospitalized as a result of their first suicide attempt, and 100% of those who endorsed an “extreme” suicide attempt were hospitalized (Corcos et al., 2002).

Whether the risk of suicide attempts across eating disorder subtypes differs remains a matter of some disagreement, with some studies reporting no difference in the prevalence of attempts (Bulik et al., 1999; Herzog et al., 1999; Milos et al., 2004), some reporting higher prevalence of suicide attempts in individuals with BN than in those with AN (Favaro & Santonastaso, 1996, 1997), and others reporting higher prevalence of suicide attempts in individuals with AN than in those with BN (Franko et al., 2004). The differences are likely attributable to differences in subtyping diagnostics although one fairly consistent finding is higher risk amongst the binge-purge subtype of AN than individuals with the restricting subtype (Bulik et al., 2008; Favaro, Tenconi, & Santonastaso, 2006; Franko & Keel, 2006). However, most of the research has focused on differences in prevalence of suicide attempts across AN subtypes and BN. Research on the prevalence of suicide attempts in individuals with BED or PD is extremely limited. One recent study of patients with BED presenting for outpatient treatment found that 12.5% had a lifetime history of a suicide attempt, providing initial indication of elevated risk of suicide attempts in this population compared to individuals with no ED (Carano et al., 2012). PD has not been included in any of the large studies of suicide attempts in women with eating disorders. Additional studies are needed to further clarify the prevalence of suicide attempts in individuals with BED and individuals with PD, and compare suicide risk in the other ED groups. Large population-based studies have the potential to clarify differences in suicide risk across the ED diagnostic categories, particularly in these understudied disorders, and inform risk assessment in treatment settings.

Reports have been inconsistent when addressing whether specific patient profiles or characteristics are associated with suicide attempts in individuals with eating disorders. Factors identified as associated with suicide attempts in isolated studies of AN include older age, longer duration of illness, lower BMI, depression, greater number of past treatments, antidepressant use, elevated phobic anxiety, and drug and alcohol abuse (Favaro & Santonastaso, 1997; Forcano et al., 2009). For BN, associated factors include greater general psychopathology, greater number of past treatments, antidepressant use, lower education, lower minimum BMI, family history of alcohol abuse, increased impulsive behaviors including self-injury, and lower self-directedness, cooperativeness and reward dependence, and higher harm avoidance (Favaro & Santonastaso, 1997; Forcano et al., 2009). Current evidence suggests that suicidality is associated with anxious personality traits such as harm avoidance and neuroticism in various psychiatric and community samples of eating disorders (Bulik et al., 1999; Engstrom, Brandstrom, Sigvardsson, Cloninger, & Nylander, 2004; McGirr, Paris, Lesage, Renaud, & Turecki, 2007; Ruchkin, Schwab-Stone, Koposov, Vermeiren, & King, 2003; van Heeringen et al., 2003). Moreover, impulsivity, high novelty seeking and low self-directedness have also been reported (Forcano et al., 2009; McGirr et al., 2007; Zouk, Tousignant, Seguin, Lesage, & Turecki, 2006). The co-existence of anxious and impulsive traits may converge to increase suicidal risk.

Several comorbid psychiatric disorders have been associated with suicide attempts in individuals who have an eating disorder, including a lifetime history of major depression (Anderson, Carter, McIntosh, Joyce, & Bulik, 2002; Bulik et al., 2008; Corcos et al., 2002; Favaro & Santonastaso, 1997) with over 80% of individuals with AN who attempted suicide reporting that their worst or only attempt occurred during an active episode of major depressive disorder (Bulik et al., 2008). In addition, anxiety disorders (e.g., post-traumatic stress disorder, panic disorder, and a broad diagnosis of “any anxiety disorder”) (Bulik et al., 2008; Milos et al., 2004), substance abuse and alcohol abuse (Anderson et al., 2002; Corcos et al., 2002; Franko et al., 2004) have been associated with suicide attempts in AN and BN.

The present study clarifies and extends this body of research by employing data from the population-based Swedish Twin Registry (STR) in conjunction with the Swedish National Patient Register and the National Cause of Death Register. Conducting research with national registers is a valuable methodological approach, especially in Nordic countries (Allebeck, 2009) where data are reliable and participants are less likely to be lost to follow-up. Moreover, all deaths and hospitalizations in Sweden are captured in national registers (Ludvigsson et al., 2011; National Board of Health and Welfare, 2010). Using Swedish national registers, we evaluated whether suicide attempts/completions were more prevalent in individuals with AN subtypes, BN, lifetime history of both AN and BN (ANBN), BED and PD than in individuals without eating disorders and explored whether there were specific psychopathological, temperament, and personality features that associated with suicide attempts in individuals with eating disorders.

Method

Participants

Participants were female twins born between 1959–1985 and assessed as part of the Swedish Twin study of Adults: Genes and Environment [STAGE; n = 13,035 (Furberg et al., 2008; Lichtenstein et al., 2006)]. STAGE data were collected in 2005 using web-based questionnaires and phone interviews (response rate for full STAGE dataset = 59.6%). Questionnaires assessed demographic information; medical history; presence of psychiatric disorders including detailed information on eating disorders, alcohol and illicit substance use; and personality variables including neuroticism, extraversion, perfectionism and self-directedness. Participants were between 20 and 47 years of age at the time of assessment.

Determination of zygosity was based on responses to standard twin similarity questions, which were validated with a panel of 47 single-nucleotide polymorphisms in a random sample of 198 twin pairs. Ninety five percent (n = 188) were correctly classified. This zygosity algorithm has also previously been validated with similar results (Lichtenstein et al., 2002). Of the twins included in the present study, 42.4% were from monozygotic twin pairs, 30.2% were from same-sex dizygotic twin pairs, 25.2% are from opposite-sex twin pairs, and 2.3% were of unknown zygosity.

Identification of Attempted and Completed Suicide

All Swedish citizens since 1947 and, therefore, all participants in STAGE, have an assigned unique personal identification number [national registration number; (Ludvigsson, Otterblad-Olausson, Pettersson, & Ekbom, 2009)]. Via this number, the STAGE database can be linked to any Swedish national register. In order to identify all recorded suicide attempts and completions, STAGE was linked with following registers:

• National Patient Register. The National Board of Health and Welfare maintains the National Patient Register (National Board of Health and Welfare, 2010), which covers all public inpatient hospitalizations in Sweden. Each record contains admission and discharge dates, primary discharge diagnosis, and up to eight secondary diagnoses using the International Classification of Diseases (ICD) 8, ICD-9, or ICD-10 depending on the year of hospitalization (World Health Organization, 1967, 1978, 1992). The attending physician documented the diagnoses. This register captures all inpatient psychiatric care in Sweden since 1973. The register also routinely captures suicide attempts using ICD codes (codes E950-E959 in ICD-8 and ICD-9, and X60-X84 in ICD-10). This register was searched for any discharge diagnoses indicating suicide attempts.

• The Cause of Death Register. All deaths in Sweden from 1958–2009 are contained in the Cause of Death Register (National Board of Health and Welfare). The diagnoses and causes of death are coded according to ICD codes. The register routinely codes suicide as a cause of death. This database was searched to identify all cases of death by suicide (ICD-10 X60-X84 for years 2005–2009).

• Suicide Attempts/Completions. Information on suicide attempts was extracted from the National Patient Register for the years 1969–2009. Therefore, all suicide attempts not resulting in death prior to participation in STAGE and suicide attempts after participation in STAGE were captured. To participate in STAGE, individuals needed to be alive in 2005 and suicide information was available up to 2009. As such, information on completed suicides was only available for the interval between 2005–2009. Due to the restricted interval to capture completed suicides, the number of completed suicides was hypothesized a priori to be too small to be adequately powered to conduct independent analyses across groups. Therefore, we created a composite variable that included suicide attempts and completions. In addition, the number of individuals with an initial suicide attempt subsequent to their participation in STAGE was hypothesized a priori to be too small to comprise a sufficiently large group to be examined independently (i.e., suicide attempt prior to STAGE participation versus after STAGE participation). For this reason we identified the presence or absence of lifetime suicide attempts as the primary outcome variable.

Eating Disorder Diagnosis

Narrow.

Two sets of diagnostic criteria were used to define eating disorders. The first reflected narrow definitions from the Diagnostic and Statistical Manual for Psychiatric Disorders Fourth Edition [DSM-IV; (American Psychiatric Association, 2000)]. Specifically, AN was coded if a participant: 1) indicated that she had a period of time when she weighed much less than other people thought she should weigh and reported a BMI < 17.55; 2) indicated being very or extremely afraid of gaining weight or becoming fat; and 3) indicated feeling very or extremely fat when at low weight. Amenorrhea was not required for the diagnosis as it is an unreliable diagnostic criterion for AN (American Psychiatric Association, 1994; Bulik, Sullivan, & Kendler, 2000) and was ultimately eliminated in the DSM-5 (American Psychiatric Association, 2013). Information about binge eating (eating an unusually large amount of food in a short period of time with at least slight loss of control) and purging (defined as vomiting, diuretic use, or laxative use during the time when binge eating was occurring or at least weekly) was also collected and used to further classify women as anorexia nervosa, restricting subtype (ANR; absence of both lifetime binge eating and purging) or anorexia nervosa, binge-purge subtype, (ANBP; presence of lifetime binge eating and or purging).

Narrow BN was defined as meeting DSM-IV criteria A (binge eating), B (inappropriate compensatory behaviors), C (binge eating episodes occur at least eight times a month for at least three months), and criterion D (body weight or shape are important or the most important factors in self-evaluation).

Narrow BED was defined as meeting DSM-IV criteria A (binge eating), B (endorsing at least three of the following symptoms: eating faster than usual, eating until uncomfortably full, eating large amounts of food when not hungry, eating alone due to embarrassment, and feeling disgusted/depressed/guilty after overeating), C (distress or upset over binge episodes), D (at least eight binge episodes a month for at least three months), and E (did not engage in inappropriate compensatory behaviors during the time when they were binge eating). A diagnosis of BED was not made if the participant had a history of either AN or BN.

If a participant indicated that she engaged in vomiting, laxative use, diuretic use at least weekly, indicated that body weight or shape are important or the most important factors in self-evaluation, had no history of binge eating and did not have a lifetime history of AN, she was scored positive for narrow PD.

In addition to the specific questions described above, participants were also asked the general prompt “do you have or have you ever had any of the following problems?” and one of the listed problems was “anorexia/bulimia/eating disorders” as part of a medical checklist (n = 11,117). Participants responded “yes” or “no” to this question. This question was used in the present study only to identify agreement between the response on the medical checklist and eating disorder diagnosis as established by the diagnostic algorithms (κ = 0.40). Most individuals who responded yes to the checklist item were also given an eating disorder diagnosis (n = 261). Of the 10,391women who were not classified as meeting a narrow diagnosis for an eating disorder, 642 (6.10%) self-identified as having an eating disorder. There were 50 individuals who said no to the checklist item, but were classified as having an eating disorder diagnosis by the algorithm

Broad.

For the broad diagnoses, the criteria were modified for each disorder. Specifically, AN was coded if a participant: 1) indicated that she had a period of time when she weighed much less than other people thought she should weigh and had a BMI < 18.55; 2) indicated being slightly, somewhat, very, or extremely afraid of gaining weight or becoming fat; and 3) indicated feeling slightly, somewhat, very, or extremely fat when at low weight. For broad BN, criterion C was modified: a reduced frequency of binge eating of four or more times per month was required. In addition, criterion D was defined as body weight or shape at least moderately influences self-evaluation. This definition has been used previously (Root et al., 2010) and has been shown to improve the detection of binge eating behavior without significantly increasing the prevalence of the disorder (Trace et al., 2012). Broad BED was defined as meeting DSM-IV criteria with a reduced frequency of at least four binge eating episodes in a one-month period (broadened criterion D). For broad purging disorder, vomiting, laxative use, or diuretic use had to occur at least weekly and body weight or shape at least moderately influenced self-evaluation.

Agreement between those identified as having a broadly defined eating disorder diagnosis from the above-described criteria and those who self-identified as having an eating disorder on the medical checklist was calculated (κ = 0.60). Many individuals identified by the algorithms as having an eating disorder diagnosis also answered yes to the eating disorder question on the medical checklist (n = 524). However, 202 women met criteria for an eating disorder using the algorithms but responded no to the checklist item. Of those who were not classified as meeting a broad diagnosis for an eating disorder, 407 self-identified as having an eating disorder.

The eating disorder categories used in the analyses were non-overlapping. For both the narrow and broad definitions, each person was classified as either having a lifetime diagnosis of ANR, lifetime diagnosis of ANBP (the respondent was scored as positive for binge eating, defined below, and/or indicated engaging in at least one purging behavior weekly or daily), lifetime diagnosis of BN, lifetime diagnosis of BED, lifetime diagnosis of PD, or no lifetime diagnosis of an eating disorder (no ED). Any participant who had a lifetime history of both AN and BN (narrow definition, N = 23, broad definition, N = 103) was classified as ANBN. Participants who met broad criteria for an eating disorder but not narrow criteria (N = 388) were excluded from the analyses for the narrow diagnoses, as those individuals could not be considered cases or controls. Therefore, the final sample size for the analyses of the broad definitions of illness was 13,035.

Eating Disorder Features

Binge Eating.

The main binge eating question was: “Have you ever had binges when you ate what most people would regard as an unusually large amount of food in a short period of time?” with response options yes, no, and don’t know/refuse. Positive responses were followed by, “When you were having eating binges, did you feel your eating was out of control?” Response options were: not at all, slightly, moderately, very much, extremely, and don’t know/don’t wish to answer. Binge eating was scored as present if the individual responded yes to the first question and indicated feeling slightly, moderately, very much, or extremely out of control.

Eating Disorder Behaviors.

Two questions evaluated weight control methods. Individuals who endorsed binge eating were asked whether they engaged in compensatory behaviors (vomiting, laxative use, diuretic use, diet pills) during the time of binge eating. Those who responded that they engaged in compensatory behaviors during the time of binge eating were scored as positive for the respective method. Individuals who did not endorse binge eating were asked whether they ever engaged in vomiting, laxative use, diuretic use, or diet pill use at any point in their lifetime to control shape or weight. Response options were never, once or twice, weekly, or daily. Those who engaged in any of the behaviors weekly or daily were scored as positive for the respective weight control method.

Each participant was asked whether she ever fasted to control her shape or weight or had not eaten for 24 hours or more (present/absent). Excessive exercise reflected exercising more than 2 hours per day to control shape and weight. Those who endorsed “daily” were scored positive for excessive exercise.

Amenorrhea.

Women were asked to recall age at menarche. Those who got their first period at age 16 or later, those who had not yet experienced menarche prior to AN onset, and those who reported missing 3 or more periods were classified as having amenorrhea.

BMI.

Each participant reported lowest weight in kilograms (kg) since age 18 and current height in meters (m). For women who did not have a history of AN, lowest adult BMI (kg/m²) was calculated. For women with a history of AN, lowest BMI was calculated from lowest weight during AN and height at the time of low weight. Current BMI was calculated using current height (m) and participant reported current weight (kg). Highest BMI was calculated using current height (m) and participant reported highest weight (not including pregnancy; kg). BMI difference was calculated by subtracting the lowest BMI value from the highest BMI value.

Age of Onset.

Age of onset of ANR was defined as age at lowest illness-related weight. Age of onset of BN and of BED was defined as age at first binge. Age of onset of ANBP and ANBN was defined as age at first binge or age at lowest illness-related weight, which ever was younger. No age of onset information was provided for inappropriate compensatory behaviors so these data are unavailable for participants with PD.

Psychiatric Comorbidity

Other psychiatric disorders were assessed using detailed self-report questionnaires based on the Structured Clinical Interview for DSM-IV [SCID; (First, Spitzer, Gibbon, & Williams, 2002)]. Depression was coded as present if criterion A (five symptoms of depression, including depressed mood and/or anhedonia, associated with a change of functioning) and criterion C (significant impairment caused by the symptoms) were met. Participants needed to endorse the symptoms of depression for two or more weeks in a row and experience these symptoms “all day long” or “most of the day” to be coded as meeting criteria for depression. Participants were also asked whether they “have or have ever had… depression” with a response option of “yes” or “no.” Agreement between those identified as having lifetime depression from the above-described algorithm and those self-identified as having lifetime depression from the yes/no question was calculated (κ = 0.56).

Generalized anxiety disorder (GAD) was considered present if DSM-IV-TR (American Psychiatric Association, 2000) criteria A (excessive anxiety and worry) and C (at least three symptoms resulting from anxiety and worry) were met. Lifetime prevalence of specific phobias, obsessive-compulsive disorder (OCD), and panic disorder were assessed with the question “have you ever had any of the following problems?” Each disorder was then listed and response options were “yes” and “no.” A composite “any anxiety disorder” variable was also assessed: if the participant had a history of GAD, phobias, OCD, or panic disorder, they were scored positive for any anxiety disorder.

Alcohol abuse/dependence was assigned based on DSM-IV criteria (American Psychiatric Association, 1994); participants were asked about the presence (and frequency where appropriate) of each abuse and dependence criterion. If a participant responded positively to one abuse criterion or three or more dependence criteria, she was given a positive diagnosis for alcohol abuse/dependence. Substance use was defined as using marijuana/hash, opioids, stimulants, hallucinogens, sedatives and/or hormones 10 times or more in one month.

Temperament and Personality

Concern over mistakes (α = 0.82), personal standards (α = 0.81), and doubts about actions (α = 0.90) were each evaluated using four items from the subscales from the Frost Multidimensional Perfectionism Scale (Frost, Marten, Lahart, & Rosenblate, 1990). Extraversion was evaluated using nine items (α = 0.78) and neuroticism was evaluated using 18 items (α = 0.90) of the Eysenck Personality Inventory (Schapiro et al., 2001). Self-directedness was measured using five items (α = 0.84) from the Temperament and Character Inventory (TCI) (Cloninger, 1994).

Statistical Analyses

All data management and analyses were conducted using SAS 9.2 (SAS Institute Inc., 2008). The STAGE database was first linked with the Swedish National Patient Register and the Swedish National Cause of Death Register. For each individual, the total number of lifetime suicide attempts resulting in hospitalization was based on the number of unique dates of hospital discharge entries into the Swedish National Patient Register with a suicide attempt code. Individuals who died as a result of a suicide attempt were identified from a cause of death code for suicide in the Swedish National Cause of Death Register. Identified completed suicides were added to the number of suicide attempts generated from the Swedish Hospital Discharge Register if the date of the death was different from the last Swedish National Patient Register entry.

Each individual’s age at first attempt was calculated from the date of discharge or death. ICD codes were used to categorize the method used for each attempt. Methods were classified as “violent” (e.g., stabbing, hanging, jumping from a high place) and “non-violent” (see Table 1 for classification by ICD code). Any participant who had more than one suicide attempt with at least one attempt classified as violent was coded as having had a violent attempt. The prevalence of at least one lifetime suicide attempt and of a violent suicide attempt among those who had ever attempted suicide were computed for ANR, ANBP, ANBN, BN, BED, PD, and no ED groups by narrow and broad eating disorder definitions. Means and standard deviations for number of attempts and age at first attempt were calculated for the all groups.

Table 1:

International Classification of Diseases (ICD) codes used to identify suicide attempts and completions and the violence categorization for each code

ICD-8 and ICD-9 Codes for Suicide Attempts and Completions
E950	Suicide and self-inflicted poisoning by solid or liquid substances	Non-violent
E951	Suicide and self-inflicted poisoning by gases in domestic use	Non-violent
E952	Suicide and self-inflicted poisoning by other gases and vapors	Non-violent
E953	Suicide and self-inflicted injury by hanging strangulation and suffocation	Violent
E954	Suicide and self-inflicted injury by submersion (drowning)	Violent
E955	Suicide and self-inflicted injury by firearms, air guns, and explosives	Violent
E956	Suicide and self-inflicted injury by cutting and piercing instrument	Violent
E957	Suicide and self-inflicted injuries by jumping from high place	Violent
E958	Suicide and self-inflicted injury by other and unspecified means	Non-violent
E959	Late effects of self-inflicted injury	Non-violent
ICD-10 Codes for Suicide Attempts and Completions
X60	Intentional self-poisoning by and exposure to non-opioid analgesics, antipyretics and anti-rheumatics	Non-violent
X61	Intentional self-poisoning by and exposure to antiepileptic, sedative-hypnotic, anti-parkinsonism and psychotropic drugs, not elsewhere classified	Non-violent
X62	Intentional self-poisoning by and exposure to narcotics and psychodysleptics [hallucinogens], not elsewhere classified	Non-violent
X63	Intentional self-poisoning by and exposure to other drugs acting on the autonomic nervous system	Non-violent
X64	Intentional self-poisoning by and exposure to other and unspecified drugs, medicaments and biological substances	Non-violent
X65	Intentional self-poisoning by and exposure to alcohol	Non-violent
X66	Intentional self-poisoning by and exposure to organic solvents and halogenated hydrocarbons and their vapours	Non-violent
X67	Intentional self-poisoning by and exposure to other gases and vapours	Non-violent
X68	Intentional self-poisoning by and exposure to pesticides	Non-violent
X69	Intentional self-poisoning by and exposure to other and unspecified chemicals and noxious substances	Non-violent
X70	Intentional self-harm by hanging, strangulation and suffocation	Violent
X71	Intentional self-harm by drowning and submersion	Violent
X72	Intentional self-harm by handgun discharge	Violent
X73	Intentional self-harm by rifle, shotgun, and larger firearm discharge	Violent
X74	Intentional self-harm by other and unspecified firearm discharge	Violent
X75	Intentional self-harm by explosive material	Violent
X76	Intentional self-harm by smoke, fire, and flames	Violent
X77	Intentional self-harm by steam, hot vapors, and hot objects	Violent
X78	Intentional self-harm by sharp object	Violent
X79	Intentional self-harm by blunt object	Non-violent
X80	Intentional self-harm by jumping from a high place	Violent
X81	Intentional self-harm by jumping or lying before moving object	Violent
X82	Intentional self-harm by crashing of motor vehicle	Violent
X83	Intentional self-harm by other specified means	Non-violent
X84	Intentional self-harm by unspecified means	Non-violent

Logistic regression analyses (using PROC GENMOD in SAS) with generalized estimating equations (GEE) were applied to assess differences in the prevalence of suicide attempts/completion across the ANR, ANBP, ANBN, BN, BED, PD and no ED groups. GEE was used to account for the nesting of the data within twin pairs in this and all subsequent full model analyses. GEE assumes a relationship within clusters (twin pairs) and this relationship is modeled and treated as a nuisance variable. For our analyses, the exchangeable correlation matrix was used to model the relationship for the analyses of the ANR, ANBP, ANBN, and PD groups and for the personality features for the BN and BED groups. The identity matrix was used for the analyses of the eating disorder features and comorbidity in the BN groups because the models for these groups would not converge. Models for the eating disorder features and comorbidity for the BED groups could not be applied. Age at assessment was entered as a covariate in all models. Type 3 score statistics were used to determine the significance of the independent variable in the models. Post hoc contrasts, which use adjusted means, were requested to assess pairwise group differences for the omnibus tests that were significant. The score statistics are presented as χ² for all analyses. Using the guidelines presented by Chen, Cohen and Chen (2010), odds ratios of 1.68, 3.47, and 6.71 are considered equivalent to Cohen’s d = 0.2 (small), 0.5 (medium), and 0.8 (large), respectively (Chen, Cohen, & Chen, 2010). Therefore, odds ratios greater than 1.68 indicate a small effect size and are presented even if the model did not reach significance.

Among those who ever attempted suicide, differences in the prevalence of violent suicide attempts and in age at first suicide attempt across eating disorder category were assessed using general linear models with GEE corrections. To determine whether eating disorder category was associated with the total number of lifetime suicide attempts, a Poisson regression with GEE corrections was conducted.

Associations between suicide attempts and: lifetime lowest BMI, age of onset of eating disorder, specific eating disorder features (including history of vomiting, laxative use, diet pill use, diuretic use, excessive exercise, fasting, other inappropriate compensatory behaviors, and amenorrhea), psychiatric comorbidity, and personality traits were evaluated for each eating disorder group separately using general linear models with GEE corrections. All p-values of omnibus tests were corrected for multiple testing using the methods of False Discovery (Benjamini & Hochberg, 1995).

In addition, sign tests were conducted to evaluate whether there was a systematic difference in reporting of comorbid conditions between individuals with a history of suicide attempts and those who had never attempted suicide. Specifically, we evaluated whether those with suicide attempts reported higher prevalences across a majority of comorbid conditions than those without suicide attempts. These analyses were stratified by eating disorder group, were one-tailed, and were not corrected for multiple testing.

Results

Demographics

Narrow diagnostic groups.

Demographic information is presented across the narrow eating disorder categories in Table 2a. There were no differences in age at assessment across the eating disorder groups. BMI at time of assessment differed across eating disorders (χ² = 48.79, p < .0001). Post-hoc analyses revealed that the BED group had a significantly higher mean BMI at assessment than all of the other ED categories. Additionally, the mean BMI at assessment for the ANR and ANBP group were significantly lower than the BN, PD, and no ED groups. Education level differed across groups (χ² = 15.38, p < .02); post-hoc analyses revealed that education levels differed between the BN and no ED groups. Civil status differed across groups (χ² = 13.83, p < .04). The prevalence of being married or cohabiting with a partner was lower in the BN and BED groups than the no ED group.

Table 2a:

Mean (SD) of Age and BMI at time of STAGE assessment and prevalence, N (%), of education and relationship status by eating disorder category for the narrow definitions of eating disorders.

Narrow Eating Disorder Category	ANR (N = 65)	ANBP (N = 75)	ANBN (N = 23)	BN (N = 124)	BED (N = 22)	PD (N = 83)	No ED (N = 12,255)
	M (SD)	M (SD)	M (SD)	M (SD)	M (SD)	M (SD)	M (SD)
Age at Assessment	32.20 (7.78)	30.68 (7.32)	30.09 (6.09)	32.59 (7.14)	33.09 (7.32)	31.78 (7.36)	33.59 (7.65)
BMI at Assessment	21.37 (4.05)	21.07 (2.89)	22.94 (4.99)	22.76 (3.76)	28.72 (7.18)	23.07 (3.83)	23.17 (3.90)
	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)
Education
Some Secondary or Secondary Graduate	26 (40.63)	26 (34.67)	7 (30.43)	36 (29.27)	6 (27.27)	36 (43.90)	5113 (42.16)
University	30 (46.88)	38 (50.67)	12 (52.17)	66 (53.66)	12 (54.55)	29 (35.37)	5603 (46.20)
Other (Military, Vocational, or Folk)	8 (12.50)	11 (14.67)	4 (17.39)	21 (17.07)	4 (18.18)	17 (20.73)	1412 (11.64)
Civil Status
Married or Cohabiting	39 (61.90)	41 (55.41)	12 (52.17)	68 (54.84)	10 (45.45)	54 (66.67)	8283 (68.40)
Living Alone	24 (38.10)	33 (44.59)	11 (47.83)	56 (45.16)	12 (54.55)	27 (33.33)	3826 (31.60)

Abbreviations: ANR, restricting subtype anorexia nervosa; ANBP, binge-purge subtype anorexia nervosa; ANBN, lifetime diagnosis of both anorexia nervosa and bulimia nervosa; BN, bulimia nervosa; BED, binge eating disorder; PD, purging disorder; No ED, no eating disorder

Broad diagnostic groups.

Table 2b presents the demographic information across the broadly defined eating disorder categories. There were no differences in age at assessment across the eating disorder groups. BMI at time of assessment differed across eating disorders groups (χ² = 99.57, p < .0001). Post-hoc analyses revealed that the BED group had a significantly higher mean BMI than all of the other ED categories and the mean BMI for the ANBN group was lower than the BN, PD and no ED groups. Education level differed across groups (χ² = 19.89, p < .01), with post-doc analyses revealing differences between the BN group and both the PD and no ED groups. Civil status differed across groups (χ² = 19.32, p < .04). The prevalence of being married or cohabiting with a partner was lower in the BN group than the PD and no ED groups.

Table 2b:

Mean (SD) of Age and BMI at time of STAGE assessment and prevalence, N (%), of education and relationship status by eating disorder category for the broad definitions of eating disorders.

Broad Eating Disorder Category	ANR (N =242)	ANBP (N = 127)	ANBN (N = 103)	BN (N = 251)	BED (N = 64)	PD (N = 89)	No ED (N = 12,159)
	M (SD)	M (SD)	M (SD)	M (SD)	M (SD)	M (SD)	M (SD)
Age at Assessment	31.93 (7.52)	30.96 (7.35)	32.10 (7.40)	31.77 (7.19)	30.95 (7.59)	31.92 (7.42)	33.67 (7.65)
BMI at Assessment	21.28 (3.82)	21.57 (3.47)	21.74 (3.75)	23.46 (4.62)	26.20 (6.02)	23.49 (3.79)	23.18 (3.87)
	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)
Education
Some Secondary or Secondary Graduate	88 (37.29)	50 (39.37)	30 (29.41)	81 (32.40)	20 (31.25)	41 46.59)	5087 (42.33)
University Graduate	123 (52.12)	61 (48.03)	54 (52.94)	137 (54.80)	34 (53.13)	31 (35.23)	5506 (45.81)
Other (Military, Vocational, or Folk)	25 (10.59)	16 (12.60)	18 (17.65)	32 (12.80)	10 (15.63)	16 (18.18)	1425 (11.86)
Civil Status
Married or  Cohabiting	144 (60.76)	71 (56.80)	67 (65.05)	135 (53.78)	34 (53.13)	59 (67.05)	8251 (68.77)
Living Alone	93 (39.24)	54 (43.20)	36 (34.95)	116 (46.22)	30 (46.88)	29 (32.95)	3747 (31.23)

Abbreviations: ANR, restricting subtype anorexia nervosa; ANBP, binge-purge subtype anorexia nervosa; ANBN, lifetime diagnosis of both anorexia nervosa and bulimia nervosa; BN, bulimia nervosa; BED, binge eating disorder; PD, purging disorder; No ED, no eating disorder

Prevalence of Suicide Attempts

Narrow diagnostic groups.

Suicide attempts were identified in 260 of the 12,647 (2.06%) individuals included in the analysis applying the narrow ED diagnoses, representing 11.99% of individuals with eating disorders and 1.74% of the referent group. Three individuals died subsequent to their participation in STAGE as a result of a suicide attempt: one had a lifetime diagnosis of narrowly defined BN and two had no lifetime eating disorder diagnosis.

Table 3 presents the lifetime prevalence of suicide attempts across the ANR, ANBP, ANBN, BN, BED, PD, and no ED groups for narrow and broad definitions of illness. The prevalence of suicide attempts was significantly different across narrow eating disorder groups (χ² = 31.39, p_fdr < .004). Suicide attempts were significantly more common in all eating disorder groups than in the referent group (Table 4). There were no other significant pairwise differences across the narrow eating disorder groups.

Table 3.

Lifetime prevalence, N (%), of at least one suicide attempt by eating disorder group for narrow and broad eating disorder definitions.

	No Lifetime Suicide Attempt	Lifetime Suicide Attempt
Eating Disorder Group	N (%)	N (%)
Narrow
ANR (N = 65)	60 (92.31)	5 (7.69)
ANBP (N = 75)	65 (86.67)	10 (13.33)
ANBN (N = 23)	19 (82.61)	4 (17.39)
BN (N = 124)	108 (87.10)	16 (12.90)
BED (N = 22)	19 (86.36)	3 (13.64)
PD (N = 83)	74 (89.16)	9 (10.84)
No ED (N = 12,255)	12,042 (98.26)	213 (1.74)
Broad
ANR (N = 242)	232 (95.87)	10 (4.13)
ANBP (N = 127)	116 (91.34)	11 (8.66)
ANBN (N = 103)	90 (87.38)	13 (12.62)
BN (N = 251)	222 (88.45)	29 (11.55)
BED (N = 64)	59 (92.19)	5 (7.81)
PD (N = 89)	77 (86.52)	12 (13.48)
No ED (N = 12,159)	11,969 (98.44)	190 (1.56)

Abbreviations: ANR, restricting subtype anorexia nervosa; ANBP, binge-purge subtype anorexia nervosa; ANBN, lifetime diagnosis of both anorexia nervosa and bulimia nervosa; BN, bulimia nervosa; BED, binge eating disorder; PD, purging disorder; No ED, no eating disorder

Table 4.

Results of post hoc pairwise comparison evaluating differences in suicide prevalence between eating disorder groups for narrow and broad eating disorder definitions.

		Narrow		Broad
Comparison eating disorder group	Referent	OR (95% CI)	χ2 (p-value)	OR (95% CI)	χ2 (p-value)
ANR	No ED	4.77 (1.94, 11.72)	11.56 (.001)	2.70 (1.40, 5.20)	8.79 (.003)
ANBP	No ED	7.92 (3.86, 16.20)	32.07 (.001)	5.79 (3.04, 10.95)	28.81 (.001)
ANBN	No ED	10.74 (3.70, 31.25)	19.03 (.001)	8.46 (4.53, 15.81)	44.80 (.001)
BN	No ED	8.12 (4.50, 14.65)	48.34 (.001)	7.83 (5.07, 12.09)	86.41 (.001)
BED	No ED	7.84 (2.08, 29.59)	9.25 (.003)	5.09 (1.99, 13.06)	11.47 (.001)
PD	No ED	6.59 (3.18, 13.65)	25.72 (.001)	9.16 (4.81, 17.45)	45.46 (.001)
ANBP	ANR	1.66 (0.54, 5.10)	0.79 (.376)	2.14 (0.88, 5.19)	2.84 (.093)
ANBN	ANR	2.26 (0.57, 8.97)	1.33 (.249)	3.14 (1.31, 7.54)	6.53 (.011)
BN	ANR	1.70 (0.59, 4.90)	0.98 (.324)	2.90 (1.37, 6.17)	7.71 (.006)
BED	ANR	1.65 (0.33, 8.08)	0.38 (.542)	1.89 (0.60, 5.93)	1.19 (.277)
PD	ANR	1.38 (0.44, 4.31)	0.31 (.577)	3.40 (1.40, 8.28)	7.25 (.008)
ANBN	ANBP	1.36 (0.38, 4.83)	0.22 (.637)	1.47 (0.62, 3.48)	0.75 (.387)
BN	ANBP	1.03 (0.42, 2.53)	0.00 (.956)	1.36 (0.65, 2.85)	0.65 (.420)
BED	ANBP	0.99 (0.22, 4.40)	0.00 (.991)	0.88 (0.29, 2.71)	0.05 (.827)
PD	ANBP	0.83 (0.31, 2.26)	0.13 (.719)	1.59 (0.66, 3.83)	1.06 (.304)
BN	ANBN	0.76 (0.23, 2.52)	0.21 (.649)	0.93 (0.45, 1.90)	0.04 (.833)
BED	ANBN	0.73 (0.13, 3.98)	0.13 (.716)	0.60 (0.20, 1.84)	0.79 (.374)
PD	ANBN	0.61 (0.17, 2.20)	0.57 (.453)	1.08 (0.45, 2.59)	0.03 (.858)
BED	BN	0.97 (0.23, 4.03)	0.00 (.963)	0.65 (0.24, 1.79)	0.69 (.406)
PD	BN	0.81 (0.32, 2.03)	0.20 (.656)	1.17 (0.55, 2.47)	0.17 (0.682)
PD	BED	0.84 (0.21, 3.44)	0.06 (.809)	1.80 (0.62, 5.25)	1.15 (.283)

Abbreviations: ANR, restricting subtype anorexia nervosa; ANBP, binge-purge subtype anorexia nervosa; ANBN, lifetime diagnosis of both anorexia nervosa and bulimia nervosa; BN, bulimia nervosa; BED, binge eating disorder; PD, purging disorder; No ED, no eating disorder

Characteristics of the suicide attempts among those with narrow diagnostic classifications who attempted suicide are presented in Table 5. No significant differences emerged across the narrow eating disorder groups and the referent in terms of the percentage of individuals who experienced violent suicide attempts (Table 5). The age at first attempt for individuals with at least one suicide attempt ranged from 13 to 50 years and did not differ across narrow eating disorder and referent groups (Table 5). The total number of suicide attempts ranged from 1 to 34 (Table 5). Although the number of suicide attempts did not differ across groups, there was a small effect observed for the ANBN group (OR = 1.92, 95% [0.87, 2.98]) having a higher mean number of attempts than the referent.

Table 5.

Suicide characteristics in women who attempted suicide by eating disorder category and results for models assessing differences in these features across eating disorder category for both narrow and broad eating disorder definitions.

		ANR	ANBP	ANBN	BN	BED	PD	No ED	Results
	Narrow	(N = 5)	(N = 10)	(N = 4)	(N = 16)	(N = 3)	(N = 9)	(N = 213)
	Broad	(N =10)	(N = 11)	(N = 13)	(N = 29)	(N = 5)	(N = 12)	(N = 190)
		N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	χ2 (pfdr -value)
Violent attempt	Narrow	2 (40.00)	3 (30.00)	3 (75.00)	3 (18.75)	0 (0.00)	1 (11.11)	30 (14.08)	12.33 (.147)*
	Broad	3 (30.00)	2 (18.18)	4 (30.77)	4 (13.79)	1 (20.00)	3 (25.00)	27 (14.21)	3.34 (.766)
		M (SD)	M (SD)	M (SD)	M (SD)	M (SD)	M (SD)	M (SD)	χ2 (pfdr -value)
Age at first attempt	Narrow	30.40 (4.93)	22.90 (6.26)	25.25 (9.63)	24.20 (6.39)	28.00 (10.82)	31.11 (10.41)	25.73 (8.48)	6.56 (.492)
	Broad	28.90 (4.48)	20.55 (5.82)	25.77 (8.27)	24.11 (6.75)	23.80 (9.73)	29.08 (9.89)	26.07 (8.64)	5.95 (.492)
Total number of attempts	Narrow	2.60 (3.05)	5.80 (6.64)	15.50 (18.95)	4.56 (5.50)	1 (0.00)	2.33 (1.41)	2.25 (3.78)	8.85 (.366)
	Broad	2.20 (2.20)	4.00 (6.20)	6.85 (11.63)	3.34 (4.53)	1.60 (0.89)	2.50 (1.93)	2.31 (3.95)	6.33 (.492)

Abbreviations: ANR, restricting subtype anorexia nervosa; ANBP, binge-purge subtype anorexia nervosa; ANBN, lifetime diagnosis of both anorexia nervosa and bulimia nervosa; BN, bulimia nervosa; BED, binge eating disorder; PD, purging disorder; No ED, no eating disorder

^*Model applied without correcting for non-independence of the sample.

Broad diagnostic groups.

Suicide attempts were identified in 270 of 13,035 (2.07%) individuals included in the analyses when applying broad diagnoses, representing 9.13% of individuals with eating disorders and 1.56% of the referent group. The prevalence of suicide attempts was also significantly different across the broad eating disorder groups (χ² = 51.88, p_fdr < .004). Post hoc pairwise comparisons (Table 4) revealed that suicide attempts were significantly more common in all broad eating disorder groups than in the referent. Additionally, suicide attempts were significantly more prevalent in the ANBN, BN, and PD groups than in the ANR group.

No significant differences emerged across the broad eating disorder groups and the referent in terms of the percentage of individuals who experienced violent suicide attempts (Table 5). The age at first attempt for all individuals with at least one suicide attempt ranged from 13 to 50 years and did not differ across broad eating disorder and the referent (Table 5). The total number of suicide attempts ranged from 1 to 43 and also did not differ across all groups (Table 5).

Eating Disorder Features Associated with Suicide Attempts

Narrow diagnostic groups.

Table 6a presents descriptive statistics for eating disorder features by eating disorder diagnosis for individuals with narrow eating disorders with and without suicide attempts. None of the eating disorder features was significantly associated with suicide attempts in any of the eating disorder groups. Although all of the models investigating the association between eating disorder features and suicide attempts in each of the eating disorder groups failed to reach statistical significance, several of the models produced medium and large effect sizes. The Cohen’s d for BMI difference was .52 and for age of onset was 1.24 in the ANR group, with women with suicide attempts having a greater BMI difference and an older age of onset. For the ANBN group, women with suicide attempts had lower lowest BMI values (Cohen’s d = 0.60), higher highest BMI values (Cohen’s d = 0.63), greater BMI differences (Cohen’s d = 0.63), and an older age of onset (Cohen’s d = 0.87). A greater BMI difference was also observed for women with suicide attempts in the BN group (Cohen’s d = 0.57). In the BED group, women with suicide attempts had higher lowest BMI values (Cohen’s d = 1.04) and higher highest BMI values (Cohen’s d =.75) than women who had no suicide attempts. In the PD group, those with suicide attempts had higher highest BMI values (Cohen’s d = 0.74) and a greater BMI difference (Cohen’s d = 0.84). The remaining models did not produce a remarkable effect size or did not converge. Models that did not converge are indicated in the table.

Table 6a:

Lifetime prevalence, N (%), of eating disorder features and mean (SD) characteristics of women with a lifetime history of a suicide attempt by eating disorder category for the narrow definitions of eating disorders.

Narrow Eating Disorder Category	ANR (N = 65)		ANBP (N = 75)		ANBN (N = 23)		BN (N = 124)		BED (N = 22)		PD (N = 83)
Suicide attempt	NO 60 (92.31)	YES 5 (7.69)	NO 65 (86.67)	YES 10 (13.33)	NO 19 (82.61)	YES 4 (17.39)	NO 108 (87.10)	YES 16 (12.90)	NO 19 (86.36)	YES 3 (13.64)	NO 74 (89.16)	YES 9 (10.84)
Eating disorder features	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)
Vomiting	NA	NA	50 (76.92)	9 (90.00)	17 (89.47)*	4 (100.00)*	93 (86.11)*	16 (100.00)*	NA	NA	65 (87.84)	7 (77.78)
Laxative use	NA	NA	15 (23.08)	3 (30.00)	6 (31.58)	3 (75.00)	22 (20.37)	4 (25.00)	NA	NA	13 (17.57)	2 (22.22)
Diuretic use	NA	NA	5 (7.69)	2 (20.00)	7 (36.84)	1 (25.00)	8 (7.41)	6 (37.50)	NA	NA	9 (12.16)*	0 (0.00)*
Diet pill use	2 (3.33)*	0 (0.00)*	6 (9.23)	4 (40.00)	5 (26.32)	1 (25.00)	18 (16.67)	3 (18.75)	2 (10.53)*	0 (0.00)*	13 (17.57)	5 (55.56)
Fasting	20 (33.33)	3 (60.00)	40 (61.54)	6 (66.67)	10 (52.63)*	4 (100.00)*	47 (43.52)	10 (62.50)	1 (5.26)*	0 (0.00)*	19 (25.68)	3 (33.33)
Excessive exercise	17 (28.33)	4 (80.00)	46 (70.77)	4 (40.00)	13 (68.42)*	4 (100.00)*	46 (42.59)	5 (31.25)	0 (0.00)*	1 (33.33)*	23 (31.08)	1 (11.11)
Amenorrhea	31 (65.96)	4 (80.00)	37 (67.27)	5 (55.56)	15 (83.33)	2 (66.67)	30 (46.15)	4 (36.36)	2 (33.33)*	0 (0.00)*	12 (31.58)	2 (66.67)
	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)
Lowest BMI	15.09 (1.91)	14.50 (2.42)	15.20 (1.74)	14.90 (1.45)	15.69 (1.26)	14.36 (2.86)	19.07 (2.69)	19.06 (3.73)	21.97 (3.13)	28.47 (8.25)	19.10 (2.26)	17.70 (4.30)
Highest BMI	22.93 (3.58)	26.16 (10.98)	23.69 (3.64)	24.88 (6.34)	26.32 (6.95)	31.44 (9.20)	25.62 (4.03)	27.66 (6.08)	31.86 (9.19)	38.28 (7.91)	25.80 (5.04)	31.55 (9.79)
BMI difference (Highest-Lowest)	7.80 (3.98)	11.66 (9.60)	8.49 (4.11)	9.86 (6.75)	10.48 (6.96)	16.50 (11.54)	6.49 (3.59)	8.60 (3.87)	8.33 (4.97)	9.81 (6.50)	6.55 (4.71)	13.84 (11.39)
Eating disorder age of onset	17.63 (4.06)	25.80 (8.35)	17.71 (4.23)	19.30 (2.98)	17.26 (3.23)	20.75 (4.65)	18.92 (4.53)	20.94 (7.03)	19.78 (5.56)	20.00 (9.90)	NA	NA

Abbreviations: ANR, restricting subtype anorexia nervosa; ANBP, binge-purge subtype anorexia nervosa; ANBN, lifetime diagnosis of both anorexia nervosa and bulimia nervosa; BN, bulimia nervosa; BED, binge eating disorder; PD, purging disorder; No ED, no eating disorder

^*The model evaluating the association with the variable and suicide attempt for the specific group did not converge.

Broad diagnostic groups.

Table 6b presents descriptive statistics for eating disorder features by group for individuals with broad eating disorders with and without suicide attempts. None of the eating disorder features was significantly associated with suicide attempts in any of the eating disorder groups. Regarding effect sizes, results were similar to those for the narrow eating disorders definitions: Cohen’s d = 0.80 for age of onset of eating disorder for the ANR group; Cohen’s d = 0.68 for lowest BMI and Cohen’s d = 0.64 for BMI difference for the ANBN group; Cohen’s d = 0.52 for BMI difference for the BN group; Cohen’s d = 0.71 for lowest BMI and Cohen’s d = 0.68 for highest BMI for the BED group; and, for the PD group, Cohen’s d = 0.62 for lowest BMI (those with suicide attempts had lower lowest BMI values), Cohen’s d = 0.58 for highest BMI and Cohen’s d = 0.78 for BMI difference. The remaining models produced effect sizes less than 0.50 or did not converge. Models that did not converge are indicated in the table.

Table 6b:

Lifetime prevalence, N (%), of eating disorder features and mean (SD) characteristics of women with a lifetime history of a suicide attempt by eating disorder category for the broad definitions of eating disorders.

Broad Eating Disorder Category	ANR (N =242)		ANBP (N = 127)		ANBN (N = 103)		BN (N = 251)		BED (N = 64)		PD (N = 89)
Suicide attempt	NO 232 (95.87)	YES 10 (4.13)	NO 116 (91.34)	YES 11 (8.66)	NO 90 (87.38)	YES 13 (12.62)	NO 222 (88.45)	YES 29 (11.55)	NO 59 (92.19)	YES 5 (7.81)	NO 77 (86.52)	YES 12 (13.48)
Eating disorder features	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)
Vomiting	NA	NA	84 (72.41)	10 (90.91)	68 (75.56)	12 (92.31)	170 (76.58)	26 (89.66)	NA	NA	64 (83.12)	10 (83.33)
Laxative use	NA	NA	20 (17.24)	3 (27.27)	24 (26.67)	7 (53.85)	32 (14.41)	6 (20.69)	NA	NA	11 (14.29)	3 (25.00)
Diuretic use	NA	NA	11 (9.48)	2 (18.18)	15 (16.67)	3 (23.08)	14 (6.31)	6 (20.69)	NA	NA	11 (14.29)*	0 (0.00)*
Diet pill use	8 (3.45)*	0 (0.00)*	11 (9.48)	4 (36.36)	10 (11.11)	4 (30.77)	31 (13.96)	4 (13.79)	3 (5.08)*	0 (0.00)*	10 (12.99)	5 (41.67)
Fasting	44 (18.97)	4 (40.00)	59 (50.86)	8 (72.73)	48 (53.33)	6 (50.00)	81 (36.49)	16 (55.17)	1 (1.69)*	0 (0.00)*	17 (22.08)	6 (50.00)
Excessive exercise	42 (18.10)	5 (50.00)	62 (53.45)	3 (27.27)	56 (62.22)	8 (61.54)	80 (36.04)	13 (44.83)	0 (0.00)*	1 (20.00)*	20 (25.97)	2 (16.67)
Amenorrhea	88 (46.32)	5 (55.56)	46 (52.27)	4 (40.00)	53 (60.92)	5 (45.45)	39 (42.86)	6 (33.33)	4 (21.05)*	0 (0.00)*	11 (33.33)	3 (75.00)
	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)
Lowest BMI	16.45 (1.89)	15.55 (2.37)	16.36 (1.53)	16.01 (1.73)	16.36 (1.82)	15.01 (2.16)	20.04 (2.97)	18.99 (3.13)	21.20 (2.75)	25.17 (7.45)	19.68 (2.28)	17.78 (3.65)
Highest BMI	23.12 (3.93)	24.35 (8.11)	24.39 (4.36)	25.43 (5.97)	24.23 (4.14)	26.84 (7.50)	26.06 (4.97)	27.81 (7.93)	28.61 (7.11)	33.96 (8.56)	25.90 (5.09)	30.04 (8.62)
BMI difference (Highest-Lowest)	6.64 (4.12)	8.56 (7.75)	8.05 (4.44)	9.20 (6.69)	7.84 (4.55)	11.90 (7.79)	5.94 (3.90)	8.83 (6.89)	6.88 (4.39)	8.79 (5.07)	6.15 (4.63)	12.25 (9.97)
Eating disorder age of onset	18.94 (5.25)	24.10 (7.43)	18.69 (4.16)	20.18 (4.75)	17.48 (3.44)	18.69 (5.36)	18.72 (4.35)	20.14 (6.64)	19.98 (5.16)	20.75 (6.13)	NA	NA

Abbreviations: ANR, restricting subtype anorexia nervosa; ANBP, binge-purge subtype anorexia nervosa; ANBN, lifetime diagnosis of both anorexia nervosa and bulimia nervosa; BN, bulimia nervosa; BED, binge eating disorder; PD, purging disorder; No ED, no eating disorder

^*The model evaluating the association with the variable and suicide attempt for the specific group did not converge.

Psychiatric Comorbidity and Personality Features Associated with Suicide Attempts

Narrow diagnostic groups.

Table 7a presents comorbid psychiatric disorders and personality variables across narrow eating disorder groups by the presence or absence of suicide attempts. None of the comorbid psychiatric disorders was significantly associated with suicide attempts. None of the models yielded medium or large effect sizes. Models that did not converge are indicated in the table.

Table 7a:

Lifetime prevalence, N (%), of comorbid psychiatric disorders and means (SD) of temperament characteristics by eating disorder category and by lifetime suicide attempt/completion status for narrow eating disorder definitions.

Narrow Eating Disorder Category	ANR (N = 65)		ANBP (N = 75)		ANBN (N = 23)		BN (N = 124)		BED (N = 22)		PD (N = 83)
Suicide attempt	NO 60 (92.31)	YES 5 (7.69)	NO 65 (86.67)	YES 10 (13.33)	NO 19 (82.61)	YES 4 (17.39)	NO 108 (87.10)	YES 16 (12.90)	NO 19 (86.36)	YES 3 (13.64)	NO 74 (89.16)	YES 9 (10.84)
Psychiatric Comorbidity	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)
Major depression	27 (50.94)	3 (60.00)	39 (63.93)*	10 (100.00)*	12 (70.59)*	4 (100.00)*	66 (65.35)	14 (93.33)	12 (66.67)*	3 (100.00)*	37 (56.06)*	9 (100.00)*
Generalized anxiety disorder	2 (5.41)	1 (50.00)	10 (25.00)	3 (42.86)	4 (33.33)	2 (66.67)	19 (29.69)	8 (66.67)	4 (33.33)*	2 (100.00)*	15 (28.30)	1 (20.00)
Specific phobia	5 (12.82)	1 (50.00)	8 (18.18)	3 (60.00)	3 (23.08)	2 (66.67)	14 (22.58)*	0 (0.00)*	2 (18.18)*	0 (0.00)*	10 (20.41)*	0 (0.00)*
Obsessive compulsive disorder	7 (13.46)*	0 (0.00)*	12 (23.08)	2 (25.00)	3 (20.00)	1 (33.33)	21 (25.30)	4 (36.36)	1 (6.67)*	0 (0.00)*	4 (6.56)*	0 (0.00)*
Panic disorder	18 (34.62)*	0 (0.00)*	17 (32.69)	7 (77.78)	6 (40.00)	2 (66.67)	40 (47.62)*	11 (100.00)*	6 (40.00)*	2 (100.00)*	16 (26.67)	5 (62.50)
Any Anxiety Disorder	21 (56.76)*	2 (100.00)*	29 (70.73)*	7 (100.00)*	10 (76.92)	3 (75.00)	55 (78.57)*	15 (100.00)*	8 (57.14)*	2 (100.00)*	29 (64.44)	5 (71.43)
Alcohol abuse / dependence	7 (11.67)	2 (40.00)	10 (15.38)	4 (40.00)	5 (27.78)	2 (50.00)	29 (27.88)	5 (31.25)	5 (26.32)*	0 (0.00)*	10 (13.70)	2 (25.00)
Substance use	6 (10.00)	2 (40.00)	4 (6.15)	6 (60.00)	3 (15.79)	2 (50.00)	12 (11.11)	8 (50.00)	3 (15.79)	2 (66.67)	11 (14.86)	4 (44.44)
Personality Features	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)
Concern over mistakes (MPS)	12.05 (4.02)	11.50 (2.89)	11.51 (4.47)	14.67 (4.18)	15.20 (2.86)	15.33 (1.15)	12.39 (4.39)	15.54 (2.84)	12.44 (5.20)	12.50 (0.71)	11.15 (4.16)	9.50 (4.28)
Doubts about actions (MPS)	8.77 (3.15)	10.00 (1.82)	9.34 (3.76)	11.78 (4.24)	12.60 (4.14)	11.33 (2.08)	10.76 (4.14)	12.67 (4.96)	9.25 (3.64)	10.00 (2.83)	8.12 (3.75)	9.17 (4.49)
Personal standards (MPS)	14.38 (2.90)	14.25 (3.50)	13.57 (3.24)	13.78 (2.68)	17.40 (1.72)	15.75 (3.10)	13.19 (3.24)	14.17 (3.97)	12.25 (4.60)	10.50 (3.54)	12.96 (3.24)	10.00 (5.06)
Extraversion (EPI)	5.43 (2.10)	5.50 (1.73)	5.07 (2.19)	3.88 (1.89)	5.29 (1.90)	3.00 (2.94)	4.53 (2.20)	4.09 (2.26)	4.75 (2.77)	2.00 (1.41)	5.25 (1.99)	5.80 (1.30)
Neuroticism (EPI)	6.92 (4.23)	7.25 (5.50)	8.88 (5.15)	10.57 (5.88)	10.62 (4.96)	12.00 (6.08)	10.61 (4.70)	12.83 (4.22)	9.00 (4.43)	12.00 (4.24)	7.76 (4.05)	9.57 (3.31)
Self-directedness (TCI)	19.25 (2.88)	16.00 (4.36)	17.65 (3.28)	16.00 (4.15)	15.40 (2.75)	15.75 (3.10)	16.31 (2.83)	14.92 (3.20)	16.56 (4.03)	12.50 (2.12)	17.63 (3.65)	15.33 (4.46)

Abbreviations: ANR, restricting subtype anorexia nervosa; ANBP, binge-purge subtype anorexia nervosa; BN, bulimia nervosa; BED, binge eating disorder; No ED, no eating disorder; MPS, Frost Multidimensional Perfectionism Scale; EPI, Eysenck Personality Inventory; TCI, Temperament and Character Inventory

^*The model evaluating the association with the variable and suicide attempt for the specific group did not converge.

Although there were no significant differences in the individual comorbid conditions, the sign test indicated that, in the ANBP and ANBN groups, the prevalence of all seven conditions was significantly higher in the group with suicide attempts compared with the group without suicide attempts (all p < .008).

None of the personality measures was significantly associated with suicide attempts in any of the eating disorder groups. However, several of these measures had medium or large effect sizes. Specifically, concern over mistakes was greater in those with suicide attempts than those without attempts in the ANBP (Cohen’s d = 0.73) and BN (Cohen’s d = 0.85) groups as was doubts about actions in the ANBP group (Cohen’s d = 0.61) and neuroticism in the BED group (Cohen’s d = 0.69). Personal standards was lower in those with suicide attempts than in those without attempts in the ANBN (Cohen’s d = 0.66) and PD (Cohen’s d = 0.70) groups. Extraversion was also lower in those with suicide attempts in the ANBP (Cohen’s d = 0.58), ANBN (Cohen’s d = 0.93), and BED (Cohen’s d = 1.25) groups. Those with suicide attempts in the ANR, BED, and PD groups also had lower self-directedness (Cohen’s d = 0.88, 1.26, and 0.56, respectively).

Broad diagnostic groups.

Table 7b presents comorbid psychiatric disorders and personality variables across broad eating disorder groups by the presence or absence of suicide attempts. None of the comorbid psychiatric disorders was significantly associated with suicide attempts and no medium nor large effect sizes were observed.

Table 7b:

Lifetime prevalence, N (%), of comorbid psychiatric disorders and means (SD) of temperament characteristics by eating disorder category and by lifetime suicide attempt/completion status for broad eating disorder definitions.

Broad Eating Disorder Category	ANR (N =242)		ANBP (N = 127)		ANBN (N = 103)		BN (N = 251)		BED (N = 64)		PD (N = 89)
Suicide attempt	NO 232 (95.87)	YES 10 (4.13)	NO 116 (91.34)	YES 11 (8.66)	NO 90 (87.38)	YES 13 (12.62)	NO 222 (88.45)	YES 29 (11.55)	NO 59 (92.19)	YES 5 (7.81)	NO 77 (86.52)	YES 12 (13.48)
Psychiatric Comorbidity	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)
Major depression	94 (44.98)	7 (70.00)	59 (55.14)*	11 (100.00)*	43 (53.75)	12 (92.31)	117 (57.92)	23 (85.19)	30 (53.57)*	5 (100.00)*	40 (57.97)	9 (81.82)
Generalized anxiety disorder	24 (14.63)	3 (60.00)	14 (18.92)	4 (50.00)	13 (22.41)	5 (55.56)	29 (20.71)	8 (47.06)	8 (18.60)*	2 (100.00)*	13 (24.07)	1 (20.00)
Specific phobia	21 (12.73)	1 (20.00)	18 (21.95)	1 (25.00)	13 (20.97)	4 (57.14)	27 (19.71)	2 (15.38)	5 (12.82)	1 (50.00)	10 (18.87)*	0 (0.00)*
Obsessive compulsive disorder	19 (9.18)	1 (14.29)	17 (17.89)	2 (22.22)	13 (17.81)	3 (30.00)	27 (15.17)	4 (19.05)	6 (12.24)*	0 (0.00)*	3 (4.48)*	0 (0.00)*
Panic disorder	57 (27.67)	4 (57.14)	28 (29.79)	8 (88.89)	30 (40.54)	7 (63.64)	60 (33.52)	14 (66.67)	15 (30.61)*	3 (100.00)*	20 (29.85)	6 (60.00)
Any Anxiety Disorder	83 (53.55)*	6 (100.00)*	47 (62.67)*	8 (100.00)*	40 (64.52)	9 (90.00)	91 (65.00)	19 (95.00)	22 (53.66)*	3 (100.00)*	30 (60.00)	6 (75.00)
Alcohol abuse / dependence	24 (10.53)	2 (20.00)	13 (11.40)	4 (36.36)	18 (20.22)	6 (46.15)	48 (22.43)	5 (17.24)	9 (15.52)	1 (25.00)	12 (15.79)	2 (18.18)
Substance use	21 (9.05)	2 (20.00)	9 (7.76)	6 (54.55)	8 (8.89)	7 (53.85)	16 (7.21)	9 (31.03)	4 (6.78)	3 (60.00)	9 (11.69)	5 (41.67)
Personality Features	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)	Mean (SD)
Concern over mistakes (MPS)	11.42 (3.98)	13.50 (3.51)	11.14 (4.28)	14.14 (3.63)	12.96 (4.26)	13.18 (4.38)	11.41 (4.16)	13.14 (4.05)	11.90 (4.42)	11.50 (5.45)	10.75 (4.48)	9.62 (3.62)
Doubts about actions (MPS)	8.95 (3.66)	10.38 (2.72)	8.99 (3.59)	10.29 (3.59)	10.66 (4.11)	12.09 (3.33)	9.39 (4.21)	10.65 (4.62)	9.29 (3.83)	11.75 (5.19)	8.23 (3.78)	8.25 (4.17)
Personal standards (MPS)	13.46 (3.59)	13.86 (3.24)	13.39 (3.41)	13.00 (1.73)	14.36 (3.21)	13.83 (3.38)	12.52 (3.40)	13.22 (3.70)	12.26 (3.62)	11.50 (4.36)	12.95 (3.08)	9.62 (4.34)
Extraversion (EPI)	5.07 (2.28)	5.29 (1.60)	4.92 (2.29)	4.29 (2.56)	5.01 (2.26)	3.45 (1.86)	4.83 (2.26)	3.90 (2.43)	4.70 (2.55)	3.75 (2.22)	5.26 (2.09)	5.43 (1.51)
Neuroticism (EPI)	7.60 (4.53)	9.71 (5.82)	8.30 (4.72)	11.43 (4.76)	9.28 (5.06)	10.22 (6.12)	9.48 (4.85)	10.61 (4.15)	9.07 (4.59)	10.00 (5.29)	7.65 (4.34)	9.25 (3.20)
Self-directedness (TCI)	18.48 (3.14)	16.67 (3.01)	17.92 (3.28)	15.57 (4.54)	16.49 (3.02)	15.25 (2.83)	16.83 (3.30)	16.17 (2.71)	16.27 (3.34)	14.25 (2.36)	17.85 (3.67)	15.25 (3.81)

Abbreviations: ANR, restricting subtype anorexia nervosa; ANBP, binge-purge subtype anorexia nervosa; BN, bulimia nervosa; BED, binge eating disorder; No ED, no eating disorder; MPS, Frost Multidimensional Perfectionism Scale; EPI, Eysenck Personality Inventory; TCI, Temperament and Character Inventory

^*The model evaluating the association with the variable and suicide attempt for the specific group did not converge.

Models that did not converge are indicated in the table.

The sign test indicated that, in the ANR, ANBP and ANBN groups, the prevalence of all seven conditions was significantly higher in those with suicide attempts compared with those without suicide attempts (all p < .008).

None of the personality measures was significantly associated with suicide attempts in any of the eating disorder groups. Medium or large effect sizes were observed for each measure in at least one broadly defined eating disorder group. Concern over mistakes was greater in those with suicide attempts than those without attempts in the ANR (Cohen’s d = 0.55) and ANBP (Cohen’s d = 0.76) groups as was doubts about actions in the BED group (Cohen’s d = 0.54) and neuroticism in the ANBP group (Cohen’s d = 0.66). Similar to the narrowly defined eating disorder groups, personal standards was lower in those with suicide attempts than in those without attempts in the PD group (Cohen’s d = 0.88) and extraversion was ANBN group(Cohen’s d = 0.75), and BED (Cohen’s d = 1.25) groups. Those with suicide attempts in the ANR, ANBP, BED, and PD groups also had lower self-directedness (Cohen’s d = 0.59, 0.59, 0.70, and 0.70, respectively).

Discussion

We shed further light on the risk for suicide in individuals with eating disorders by engaging the rich national registers in Sweden. Consistent with previous clinical reports, our results revealed significantly elevated prevalence of suicide attempts in individuals with all forms of eating disorders—ANR, ANBP, ANBN, BN, BED, and PD, with the highest odds relative to the referent being for individuals with narrow ANBN (OR = 10.74) and broad PD (OR = 9.16). In general, the prevalence estimates in this study are somewhat lower than published estimates from clinic-based investigations (Bulik et al., 1999; Corcos et al., 2002; Favaro & Santonastaso, 1997). This is as expected since: 1) ours represents a population-based sample of eating disorders, and 2) our definition of suicide attempts was rigorous insofar as an attempt had to be sufficiently severe as to be captured by the health care system and entered into the national registers. Clinic-based studies of eating disorders typically focus on a more severe subset of the eating disorders population. Likewise, studies that estimate the prevalence of suicide attempts based on self-report may also include attempts that do not come to the attention of the health care system. This supposition is supported by data from a non-treatment seeking sample of individuals with AN, in which only approximately 50% of those who reported a suicide attempt had ever sought medical treatment for the attempt (Bulik et al., 2008). However, the percentage of individuals seeking treatment for a suicide attempt was only slightly higher in a treatment-seeking sample of individuals with eating disorders (60%) (Corcos et al., 2002).

Our results underscore the value of the recommendation of Franko and Keel (2006) who emphasized the importance of subtyping AN into ANR and ANBP when assessing the prevalence of suicide attempts. Risk is elevated in both AN subtypes, narrowly and broadly defined, compared with the no ED group. Although there was no significant difference in prevalence of suicide attempts between the ANR and ANBP group, either narrowly or broadly defined, the broad ANR group had significantly lower prevalence of suicide attempts than the broad ANBN, BN, and PD groups.

It has long been known that diagnostic crossover is a common occurrence in individuals with AN, with around 50% of those with initial ANR migrating to a bulimic presentation at some point during the course of their illness (Bulik, Sullivan, Fear, & Pickering, 1997; Eddy et al., 2008; Fichter & Quadflieg, 2007; Tozzi et al., 2005). Suicide risk may fluctuate as the clinical presentation migrates between restricting and binge/purge forms. It is noteworthy that the highest prevalence of suicide attempts (17.39%) and the highest mean number of lifetime suicide attempts (15.50) in the present sample was in the narrow ANBN group, indicating that individuals who experience diagnostic crossover may be at particularly elevated risk. More granular investigations of longitudinal risk are required to confirm whether suicide risk fluctuates with symptom evolution or is more strongly related to trait factors.

We further extend our understanding of suicide and eating disorders by demonstrating significantly elevated risk for suicide in individuals with BED. Although the number of cases of BED was smaller than would have been expected on the basis of US population data (Hudson, Hiripi, Pope, & Kessler, 2007), suicide risk was significantly elevated in individuals with both narrow and broad BED diagnoses. As adjustments are made worldwide secondary to BED being as a bona fide diagnostic category in DSM-5, clinicians should remain vigilant for suicide risk in individuals with this disorder.

An additional extension of the knowledge base is the observation that individuals with broadly defined PD had elevated risk for suicide attempts relative to individuals with no ED (OR = 9.16) and ANR (OR = 3.40). These initial results are consistent with observations that, within those with AN and those with BN, individuals who purge are at greater risk for suicide attempts than those who do not purge (Favaro & Santonastaso, 1997) and highlight the need for additional research to better understand this high-risk yet understudied population.

Unlike previous investigations, we did not identify specific features of eating disorders that were associated with suicide attempts. The small cell sizes for some of the suicide groups most likely precluded the detection of significant effects. However, the overall pattern of results indicates that the highest odds ratios for suicide were observed in presentations that included purging (ANBP, ANBN, ANBP, and PD). This observation is consistent with findings that individuals who engage in purging behavior are at particularly elevated risk for suicide attempts.

Also consistent with previous reports, our results confirm that greater comorbid psychiatric burden (i.e., increased prevalence across seven comorbid conditions) is associated with elevated risk for suicide in individuals with ANBP and ANBN (Franko & Keel, 2006). By nature of having an eating disorder presentation with both AN and BN features, these individuals already experience elevated comorbidity burden. Individuals who present with a clinical picture of both low weight plus binge eating and/or purging are at increased risk for adverse medical outcomes (Takakura et al., 2006) and psychiatric comorbidity (Peat, Mitchell, Hoek, & Wonderlich, 2009). Our results reveal that this pattern holds in population-based samples as well clinical samples. In aggregate, these results underscore the critical importance of flagging these individuals in the clinical setting as high risk for an array of adverse outcomes including suicide attempts.

Unlike previous studies of women who sought treatment for eating disorders, we did not find clear differences in personality measures between those who did and did not attempt suicide. This may be an artifact of the fact that the personality measures in STAGE were not necessarily completed contemporaneously with the suicide attempts, whereas measurements in treatment-seeking samples capture extremes of personality evident during the acute phase of illness (Perkins et al., 2005). Moreover, some of the personality measures assessed in the STAGE cohort are not ones commonly associated with suicide risk (e.g., perfectionism).

Our sample is uniquely informative by referencing a population-based register to identify eating disorder cases. This approach eliminates the biases inherent in relying on samples of individuals seeking treatment for an ED, which skews observations toward the extreme. By coupling this population-based sample with hospital and cause of death registers, we were able to capture those suicide attempts that were sufficiently severe to warrant medical attention. This methodology yields a clear picture of risk of suicide attempts individuals with eating disorders in the general population. As there is no evidence of elevated risk of suicide in twins (Statham et al., 1998), our results from the STAGE sample are likely to generalize to the non-twin population.

Strengths

One strength of this study is the sample, which is large and population-based. Such a sample decreases the biases inherent in using treatment-seeking populations that are likely to have a more severe eating disorder and be more psychologically distressed. Eating disorders, particularly BN, often go undetected in the community (Hoek, 1991; Hudson et al., 2007). Therefore, a woman is likely to seek treatment if 1) her disorder is sufficiently severe to be noticed by others who encourage treatment; 2) she is highly distressed by her disorder; 3) she has a comorbid disorder for which she is seeking treatment; or 4) she is hospitalized for a suicide attempt and then referred for treatment of her eating disorder. Therefore, the previous reliance on samples of individuals seeking treatment for an eating disorder may have inflated the estimates of the prevalence of suicide attempts.

By estimating the prevalence of suicide attempts from hospital discharge registers and cause of death registers, we opted for a conservative estimate not subject to self-report bias. Furthermore, the use of the cause of death register, albeit for a restricted interval, allowed for the identification of completed suicides. A longer observation period would allow for direct comparisons of those individuals who completed assessments at a point in time after a suicide attempt with individuals who attempted suicide after completing assessments.

Limitations

Although this study makes a significant contribution to the literature, several limitations should be noted. First, the study is of Swedish twins. Although twin registers are often used to conduct research about the general population, twins may differ from the general population in significant ways. The population of Sweden is also fairly homogenous in terms of racial and ethnic demographics. Therefore, results must be interpreted within the context of limited generalizability.

By estimating the prevalence of suicide attempts via the hospital register, we were unable to identify those suicide attempts that did not warrant medical attention. This yields a conservative estimate; thus, the findings of this study may apply primarily to individuals with more severe suicide attempts.

STAGE was chosen for this study due to the large sample size and detailed information about eating disordered behaviors. However, we were nonetheless faced with fairly small cell sizes—especially in ANR and BED, which left us underpowered for some comparisons. Additionally, as only the lifetime presence of eating disordered behaviors and diagnoses were assessed, we did not have information on illness duration or status of illness (currently ill or recovered) at the time of assessment.

The STAGE assessment covers a wide range of physical and psychological variables. Unfortunately, due to the large number of variables being assessed, not all of the constructs were assessed in depth. For example, criterion B of GAD, difficulty controlling worry, was not assessed and therefore not required for participants to meet a diagnosis of GAD. Specific phobia, OCD, and panic disorder were assessed with a single “yes/no” self-report, limiting the validity of these diagnoses. Age of onset and status of illness (currently ill or recovered) at the time of assessment were not assessed for any of the comorbid diagnoses, which precluded the investigation of temporal precedence of eating disorders, comorbid disorders, and suicide attempts. Further, Axis II disorders were not assessed.

Conclusions

Suicide attempts in women with eating disorders in the general population are concerningly common. Suicide risk is elevated in all eating disorders studied here—ANR, ANBP, ANBN, BN, PD, and BED. In order to improve outcomes and decrease mortality, it is critical to identify individuals who, within this patient population, experience psychological suffering so intense that they feel compelled to take their own lives.