The Impact of Alcohol and Other Substance Use Disorders on Mortality in Patients With Eating Disorders: A Nationwide Register-Based Retrospective Cohort Study

Eating disorders are associated with high mortality (1–10). Anorexia nervosa (AN) is the second most lethal psychiatric disorder, surpassed only by substance use disorders (SUDs) (11). A meta-analysis from 2011 (1) found that patients with AN had an all-cause mortality rate nearly six times higher than that of the general population (standardized mortality ratio [SMR]=5.86). The corresponding rates for patients with bulimia nervosa (BN) and eating disorder not otherwise specified were lower, but still significantly elevated (BN: SMR=1.93; eating disorder not otherwise specified: SMR=1.92), which emphasizes the seriousness of these disorders (1). Research conducted since 2011 has confirmed elevated all-cause mortality in AN (3–8) and BN (3, 5, 9), whereas studies on mortality in eating disorder not otherwise specified have been limited and unable to replicate prior findings (5). Regarding cause-specific mortality, elevated rates of suicide have been found in patients with AN (2–4) and BN (12) compared with the general population, with the highest rates occurring in AN. Other work has shown increased rates of mortality from both external (suicide, accidents, and homicide) and internal (somatic and psychiatric disorders) causes in AN compared with the general population, with the highest rates reported for external causes of death (6, 7).

Individuals with eating disorders frequently present with comorbid SUDs (13), which are also associated with excess mortality, independently of eating disorders (10, 11). However, as is the case with eating disorders, some types of SUDs are more lethal than others. In a meta-review (11), the highest all-cause mortality rates were observed for alcohol and hard drugs (opioids, amphetamines, and cocaine). Since eating disorders and SUDs are deadly disorders individually, one might expect that SUD comorbidity would increase mortality risk even more in patients with eating disorders. However, studies on the impact of SUDs on mortality in eating disorders are scarce. Comorbid alcohol use disorder (AUD) has been found to predict increased all-cause mortality in a pooled sample of patients with eating disorders (14) and patients with AN (3). Conversely, other research has shown no significant association between any comorbid SUD (pooled category) and all-cause mortality in patients with AN or BN (8, 9). These studies did not include a control group, which is necessary to establish whether the effect of comorbid SUD is additive or synergistic. Among studies that did include a control group, one of them (15) found that the rate of all-cause mortality in female patients with an eating disorder and comorbid AUD was elevated compared with patients with an eating disorder without AUD, but not compared with control subjects with AUD. The remaining studies (6, 7) reported a synergistic effect of comorbid AUD on all-cause mortality in male and female patients with AN. Hence, these studies only examined the influence of AUD on mortality in AN or pooled eating disorders, and the results are ambiguous.

To enhance possibilities for mortality prevention, it is necessary to examine the contribution of AUD as well as other types of SUDs to elevated mortality across the full spectrum of eating disorder patients compared with control subjects. Therefore, the aim of this study was to assess the impact of SUDs involving alcohol, cannabis, or hard drugs on mortality in patients with AN, BN, and unspecified eating disorder separately compared with matched control subjects.

METHODS

Study Design and Population

This was a register-based retrospective cohort study. Data from several Danish nationwide registers were linked to unique personal identification numbers from the Danish Civil Registration System (16). The Danish National Patient Register (17) was used to identify a cohort of all individuals born in Denmark after December 31, 1967, and registered with an eating disorder diagnosis according to ICD-10 (18) between January 1, 1994, and December 31, 2015. The eating disorder cohort was separated into three subcohorts: AN (ICD-10 code F50.0, anorexia nervosa; F50.1, atypical anorexia nervosa), BN (code F50.2, bulimia nervosa; F50.3, atypical bulimia nervosa), and unspecified eating disorder (code F50.8, other eating disorders; F50.9, eating disorder, unspecified). The index date was defined as the date of first contact with an inpatient or outpatient treatment facility for eating disorders.

If a patient received more than one eating disorder diagnosis during the inclusion period, we used the first registered diagnosis. In cases where two eating disorder diagnoses were recorded on the same day, patients with a diagnosis of AN and either BN or unspecified eating disorder were coded as BN, and those with a diagnosis of BN and unspecified eating disorder were coded as BN. To ensure that the cohort was characterized by first-time eating disorder diagnoses, patients who had received an ICD-8 eating disorder diagnosis (codes 306.50, 306.58, 306.59) between 1976 and 1994 were excluded (ICD-9 has never been implemented in Denmark). We also excluded patients who had received an eating disorder diagnosis before age 8 (to reduce misclassification, e.g., feeding disorders in infancy and childhood) and those who emigrated between the index date and the end of the inclusion period in 2015.

For the control group, the Danish Civil Registration System was used to identify individuals from the general population (1:4 ratio) who had no record of an ICD-8 eating disorder diagnosis between 1976 and 1994 or an ICD-10 eating disorder diagnosis before 2015 and had not emigrated between the index date and the end of the inclusion period. Eligible control subjects were matched on sex, month and year of birth, and ethnicity.

Register-based studies in Denmark do not require approval by the Research Ethics Committee or informed participant consent. The study was approved by the Danish Data Protection Agency. All data were anonymized before the analyses were conducted.

Procedures

Eating disorder patients and control subjects were classified as having an SUD diagnosis if during the follow-up period they had been registered 1) with a relevant ICD-10 code (F10.1–F19.2, excluding tobacco abuse and dependence [F17.1 and F17.2]) in the Danish National Patient Register or Psychiatric Central Research Register (19), or 2) as having received treatment for SUD (illegal drugs only) in the Register of Substance Abusers in Treatment (1996–2018) (20) or treatment for AUD in the National Alcohol Treatment Register (2007–2018) (21). In cases where eating disorder patients and control subjects appeared in more than one register, we used the registration that came first.

The Danish National Prescription Register was not accessed because using receipt of prescription drugs as a proxy measure for SUDs can be problematic, for several reasons. First, it is rare for individuals to develop an SUD based on prescribed drugs, given the strict control and legislation in Denmark on prescription drugs with addiction potential. Second, many prescription drugs with addiction potential are also used to treat other psychiatric and somatic disorders. Third, SUDs are often treated with off-label prescription drugs (e.g., anticonvulsants, first- and second-generation antipsychotics). Fourth, very few prescription drugs are used only to treat one specific SUD. One example is disulfiram, which is used to treat AUD. However, one cannot be sure whether individuals receiving disulfiram fulfill the diagnostic criteria for AUD, and even if they did, they would appear in the Danish National Patient Register, the Psychiatric Central Research Register, or the National Alcohol Treatment Register.

SUDs were categorized into alcohol, cannabis, and hard drugs (heroin and other opioids, sedative-hypnotics, cocaine and other stimulants, multiple substances, and other psychoactive substances, e.g., hallucinogens, volatile solvents, designer drugs, etc.).

Data on date and primary cause of death were extracted from the Cause of Death Register (22). Causes of death were grouped into external causes (suicide, accident, murder) and internal causes (SUD, eating disorder, other psychiatric disorder, somatic disorder, other). The categories are mutually exclusive, and we classified overdose as an external cause of death. Data on birth date, sex, ethnicity (Danish or immigrant/descendant), cohabitation status at the index date (cohabiting, single, or under age 18 and living with a caregiver), and dates of emigration were retrieved from the Danish Civil Registration System. We obtained information on highest achieved education (primary/unknown, lower secondary, upper secondary, or higher) and employment status (employed, unemployed, other) at the index date from the Danish Education Register (23) and Danish Income Statistics Register (Employment Classification Module) (24), respectively. Information on the original classifications of sociodemographic variables and how these variables were recategorized is provided in the online supplement.

Statistical Analysis

We compared sociodemographic characteristics, rates of SUDs, age at first SUD diagnosis, rates of all-cause and cause-specific mortality, and age at death 1) between all eating disorder patients and matched control subjects, and 2) across the eating disorder types, using chi-square tests, t tests, or analysis of variance.

For the main analysis, we used Cox regression (hazard ratios and 95% confidence intervals) to assess all-cause mortality among eating disorder patients both with and without SUDs compared with control subjects without SUDs. In the subanalysis, we repeated the previous analysis including an additional interaction term between case status (0=control, 1=case) and type of eating disorder. Here, patients with AN, BN, and unspecified eating disorder were compared with their respective control subjects. The method of Fine and Gray (25) was used to estimate subhazard ratios for cause-specific mortality. It was only possible to do this for the entire eating disorder cohort (main analysis) because of scarce counts in the specific eating disorder groups. Eight SUD categories were examined in the main analysis: no SUD; alcohol only; cannabis only; hard drugs only; alcohol and cannabis; alcohol and hard drugs; cannabis and hard drugs; and alcohol, cannabis, and hard drugs. We were unable to include all these categories in the subanalysis because of scarce counts in the specific eating disorder groups. Instead, we examined 1) no SUD, 2) alcohol and/or cannabis, and 3) hard drugs alone or in combination with alcohol and/or cannabis. Furthermore, we used the likelihood ratio test to investigate the possibility of an interaction effect between case status (0=control, 1=case) and type of SUD. This was done in both the main analysis and the subanalysis, and the SUD categories remained unchanged.

Individuals were followed from the index date until death, emigration, or the end of the study period. For cause-specific mortality, individuals were censored when they died from “other” causes. All analyses were adjusted for age at the index date, sex, and birth year. We included individuals exposed to SUDs during the entire follow-up period irrespective of whether the SUD diagnosis was registered before or after the index date. This was done because the development of SUD probably started before the date when treatment started. We conducted a sensitivity analysis to evaluate this approach, in which SUD was treated as a time-varying covariate, where SUD prior to the index date was regarded to be present throughout the study period.

All statistical analyses were performed with STATA, version 16 (www.stata.com), and the significance threshold was set at a p value of 0.01.

RESULTS

Population Characteristics

We identified 20,759 patients with a first diagnosis of an eating disorder and 83,036 matched control subjects. The eating disorder patients were followed up for 227,538 person-years, and the control subjects for 939,628 person-years.

At the time of matching, over 90% of the eating disorder patients and control subjects were under 30 years of age, female, and Danish (Table 1). The mean follow-up time did not differ significantly between the eating disorder patients (11.0 years) and the control subjects (11.3 years). Significantly more eating disorder patients than control subjects presented with each type of SUD (alcohol: 4.7% compared with 1.0%; cannabis: 4.3% compared with 1.3%; hard drugs: 4.7% compared with 1.3%). Eating disorder patients were significantly older than control subjects at the time of first diagnosis of abuse of or dependence on hard drugs (mean age, 23.9 years compared with 23.0 years), but they did not differ significantly from control subjects with respect to age at the time of first diagnosis of alcohol abuse or dependence (mean age, 27.4 years compared with 27.0 years) or cannabis abuse or dependence (mean age, 22.6 years compared with 22.0 years). In assessing the relative timing of first SUD diagnosis among eating disorder patients and control subjects, the index date of the patients to whom the control subjects were matched was used for both groups. Compared with their matched control subjects with SUDs (N=2,276), eating disorder patients with SUDs (N=2,093) were significantly more likely to have received their SUD diagnosis <1 year before the date of the patients’ incident eating disorder (6.7% compared with 4.7%), on the same day as the incident eating disorder (8.7% and 0.0%), and <1 year after the date of the incident eating disorder (9.6% compared with 5.9%), and they were significantly less likely to have received their SUD diagnosis >1 year before the date of the incident eating disorder (20.8% compared with 23.9%), and >1 year after the date of the incident eating disorder (54.2% compared with 65.5%). The proportion of deaths during follow-up was significantly higher in the entire eating disorder cohort compared with the entire control group (11.4% compared with 0.4%).

Characteristics also varied significantly across the eating disorder types (AN, N=8,108; BN, N=5,485; unspecified eating disorder, N=7,166) (Table 2). At the time of matching, BN patients were the oldest (mean age, 23.0 years), followed by unspecified eating disorder patients (mean age, 20.3 years), and then AN patients (mean age, 18.8 years).

The unspecified eating disorder group had the highest proportion of male patients (unspecified eating disorder, 10.8%; AN, 6.3%; BN, 2.3%) as well as of patients who had cannabis abuse or dependence (unspecified eating disorder, 5.5%; AN, 3.7%; BN, 3.5%); who had hard drug abuse or dependence (unspecified eating disorder, 5.4%; BN, 5.1%; AN, 3.1%); whose SUD diagnosis was made >1 year before incident eating disorder (unspecified eating disorder, 23.9%; AN, 19.4%; BN, 18.1%); whose SUD diagnosis was made <1 year before incident eating disorder (unspecified eating disorder, 7.6%; BN, 7.0%; AN, 5.3%); whose SUD diagnosis was made on the same day as incident eating disorder (unspecified eating disorder, 9.6%; BN, 9.5%; AN, 7.1%); whose SUD diagnosis was made <1 year after incident eating disorder (unspecified eating disorder, 10.6%; BN, 9.1%; AN, 8.9%); and who died during follow-up (unspecified eating disorder, 1.8%; AN, 1.4%; BN, 1.0%).

The BN group had the highest proportion of patients with alcohol abuse or dependence (BN, 6.1%; unspecified eating disorder, 4.8%; AN, 3.8%). Also, the BN patients were the oldest at the time of first diagnosis of alcohol abuse or dependence (BN, mean age, 29.1 years; AN, 26.9 years; unspecified eating disorder, 26.9 years), cannabis abuse or dependence (BN, mean age, 23.7 years; AN, 22.4 years; unspecified eating disorder, 22.2 years), and hard drug abuse or dependence (BN, mean age, 25.2 years; unspecified eating disorder, 23.6 years; AN, 23.0 years). The AN group had the highest proportion of patients who had their SUD diagnosed >1 year after incident eating disorder (AN, 59.3%; BN. 56.3%; unspecified eating disorder, 48.3%).

Mortality Risk

Eating disorder patients exhibited a significant elevation in risk of all-cause and cause-specific mortality compared with control subjects without SUDs (Table 3). Regarding all-cause mortality, the elevation in risk was higher among eating disorder patients who abused alcohol only (adjusted hazard ratio=11.84), cannabis only (adjusted hazard ratio=4.55), hard drugs only (adjusted hazard ratio=14.16), alcohol and hard drugs (adjusted hazard ratio=19.59), cannabis and hard drugs (adjusted hazard ratio=13.67), or all three substance categories (adjusted hazard ratio=22.99) than among eating disorder patients without SUDs (adjusted hazard ratio=2.85). The elevation in risk of mortality from external causes was higher among eating disorder patients who abused alcohol only (adjusted subhazard ratio=13.87), cannabis only (subhazard ratio=9.63), hard drugs only (subhazard ratio=34.67), alcohol and hard drugs (subhazard ratio=41.09), cannabis and hard drugs (subhazard ratio=29.46), or all three substance categories (subhazard ratio=56.17) than among eating disorder patients without SUDs (subhazard ratio=3.02). Regarding mortality from internal causes, the elevation in risk was higher among eating disorder patients who abused alcohol only (adjusted subhazard ratio=11.35), hard drugs only (subhazard ratio=4.74), alcohol and hard drugs (subhazard ratio=9.68), or all three substances (subhazard ratio=8.35) than among eating disorder patients without SUDs (subhazard ratio=2.78).

For patients with AN or unspecified eating disorder, a significant elevation in risk of all-cause mortality was observed in those both with and without SUDs compared with control subjects without SUDs (Table 4). The elevation in risk was higher in AN and unspecified eating disorder patients who abused alcohol and/or cannabis (AN: adjusted hazard ratio=11.28; unspecified eating disorder: adjusted hazard ratio=10.86) or hard drugs alone or in combination with alcohol and/or cannabis (AN: adjusted hazard ratio=22.34; unspecified eating disorder: adjusted hazard ratio=15.53) than in AN and unspecified eating disorder patients without SUDs (AN: adjusted hazard ratio=3.21; unspecified eating disorder: adjusted hazard ratio=4.75). All-cause mortality risk in BN patients without SUDs did not differ significantly from that of control subjects without SUDs (p=0.21). However, BN patients who abused alcohol and/or cannabis (adjusted hazard ratio=5.86) or hard drugs alone or in combination with alcohol and/or cannabis (adjusted hazard ratio=11.43) exhibited an elevated risk compared with control subjects without SUDs.

We found no interaction effect between eating disorders and SUDs on mortality risk (Tables 5 and 6). The model is additive, and thus the estimates for the different SUDs are the same for the eating disorder and control groups, meaning that the ratio between the predicted hazard for those with any given SUD and without SUDs is independent of having an eating disorder. Therefore, the estimates apply to control subjects with SUDs compared with control subjects without SUDs, or eating disorder patients with SUDs compared with eating disorder patients without SUDs, respectively.

Regarding the sensitivity analysis (see Tables S1 and S2 in the online supplement), the results showed that treating SUD as a time-varying covariate did not alter the pattern of change in mortality risk among eating disorder patients and control subjects.

DISCUSSION

This study examined the contribution of SUDs involving alcohol, cannabis, and hard drugs to mortality in different eating disorders. Patients with any eating disorder who did not have an SUD exhibited an elevated risk of all-cause mortality (adjusted hazard ratio=2.85) as well as mortality from both external causes (adjusted hazard ratio=3.02) and internal causes (adjusted hazard ratio=2.78) compared with control subjects without SUDs. These findings are in line with a previous nationwide register-based study from Denmark (10). However, while we found similar rates for mortality due to external and internal causes, the previous study reported a much higher rate for external than internal causes, attributed to the large impact of suicide. This discrepancy may partly be because that study had a longer follow-up period. Considering that mortality due to external but not internal causes in eating disorder patients has been found to increase with age (10), it is possible that our eating disorder cohort would exhibit a higher risk of mortality from external causes if followed for a longer time.

SUD comorbidity was found to affect mortality risk in patients with any eating disorder. Compared with control subjects without SUDs, eating disorder patients who abused alcohol, cannabis, or hard drugs or different combinations of these substances exhibited a much higher risk of mortality from all causes (hazard ratios ranging from 4.55 to 22.99) and from external causes (subhazard ratios ranging from 9.63 to 56.17) than did eating disorder patients without SUDs. As for mortality from internal causes, the risk compared with control subjects without SUDs was much higher among eating disorder patients who abused alcohol only or hard drugs only, alcohol together with hard drugs, or hard drugs combined with alcohol and cannabis (subhazard ratios ranging from 4.74 to 11.35) than among eating disorder patients without SUDs. Interestingly, regarding internal causes, eating disorder patients who abused alcohol alone had a higher risk (subhazard ratio=11.35) than eating disorder patients who abused hard drugs alone (subhazard ratio=4.74) or hard drugs combined with alcohol (subhazard ratio=9.68). This pattern was not observed for external causes. There is consistent evidence that both eating disorders and AUD cause considerable physical damage (26–28). Indeed, alcohol has been found to cause more physical damage than certain hard drugs such as cocaine and other stimulants (29), which may explain our results.

Regarding the different eating disorder types, we found that AN or unspecified eating disorder patients without SUDs exhibited an elevated risk of all-cause mortality compared with the respective control subjects without SUDs (AN: adjusted hazard ratio=3.21; unspecified eating disorder: adjusted hazard ratio=4.75). This is consistent with previous research (1, 3–8). However, BN patients without SUDs did not differ significantly from the respective control subjects without SUDs with respect to all-cause mortality. This contrasts with a previous meta-analysis (1), which found elevated rates of all-cause mortality in BN patients compared with the general population. However, at the individual level, numerous studies reported that BN was not significantly associated with all-cause mortality (1). One possible explanation for the divergent findings is that a considerable proportion of patients with BN may have a history of AN, which has not been considered in most studies. Because of sparse counts, we were unable examine cause-specific mortality risk in the specific eating disorder groups. Previous research has been restricted to examining either death rates by broad death categories (external and internal causes) in AN (6, 7) or suicide-specific mortality in AN (2–4) and BN (2). Thus, there is a need for more large-scale longitudinal studies examining the specific underlying causes of death across the full spectrum of eating disorders.

Furthermore, we observed a higher risk of all-cause mortality relative to the respective control subjects in AN or unspecified eating disorder patients who abused alcohol and/or cannabis (AN: adjusted hazard ratio=11.28; unspecified eating disorder: adjusted hazard ratio=10.86) or hard drugs alone or in combination with alcohol and/or cannabis (AN: adjusted hazard ratio=22.34; unspecified eating disorder: adjusted hazard ratio=15.53) than in AN or unspecified eating disorder patients without SUDs. Regarding BN, the risk of all-cause mortality compared with the respective control subjects was only elevated among BN patients who abused alcohol and/or cannabis (adjusted hazard ratio=5.86) or hard drugs alone or in combination with alcohol and/or cannabis (adjusted hazard ratio=11.43). Our finding that AN patients who abused hard drugs had the highest all-cause mortality risk is not unexpected, considering that independent of each other, AN and SUDs involving hard drugs are the deadliest of psychiatric disorders (11). The fact that mortality was not raised in BN patients without SUDs suggests that SUDs may be the main driver behind mortality in BN. This is not surprising, since there is consistent evidence that BN patients who abuse drugs are more impulsive (30–32), and mortality due to external causes has been linked to impulsivity (33–37).

Lastly, our results revealed that each type of SUD exerted a negative additive effect on mortality among eating disorder patients. Control subjects with SUDs also exhibited an elevated risk of all-cause mortality relative to control subjects without SUDs, although to a much lesser extent than eating disorder patients with SUDs. In contrast, Suzuki and colleagues (15) found that comorbid AUD did not even have an additive effect on all-cause mortality in a pooled sample of eating disorder patients. That study was conducted in Japan, included only female eating disorder patients and control subjects, and had a shorter follow-up time, which may explain the discrepancy in findings. Studies conducted in Sweden (6, 7) showed that comorbid AUD exerted a negative synergistic effect on all-cause mortality in male and female patients with AN. These studies had a longer follow-up period (a mean of 21 years) than our study, and thus it may be the case that a synergistic effect of AUD, and possibly other SUDs, manifests later in life.

Major strengths of this study are the large nationwide sample of eating disorder patients, well-matched control subjects, and a long follow-up period. Also, this is the first study to examine the effects of different constellations of eating disorders and SUD comorbidities on mortality risk. Some limitations should also be taken into account. First, we only had access to data on individuals who were treated for eating disorders and/or SUDs. Thus, our findings most likely represent the more severe cases or cases with psychiatric comorbidity and cannot be generalized to those who receive care elsewhere, such as in general practice or private institutions, or those who avoid seeking treatment. Second, the unspecified eating disorder category according to ICD-10 criteria does not allow for the separate analysis of diagnostic subcategories, and therefore it is likely that the mortality pattern among the unspecified eating disorder patients will change with the implementation of ICD-11 (which will occur in 2022 in Denmark). For instance, binge-eating disorder, which is included in the unspecified eating disorder category (F50.8), will constitute a separate diagnosis in ICD-11, and it is well known that SUDs frequently co-occur with binge-eating disorder (38, 39).

A third limitation is that it was not possible to further distinguish the specific substances in the “hard drugs” category. This was because in all the registers providing data on SUDs, many eating disorder patients and control subjects with SUDs were coded as using more than one type of hard drug at the same time, or as using multiple substances and other psychoactive substances (F19), depending on the clinical tradition of applying the diagnostic nomenclature. Furthermore, numerous eating disorder patients and control subjects with SUDs were detected through the Register of Substance Abusers in Treatment, which covers the treatment of illegal drugs (not alcohol), and in that register, many of them were coded as treated for addiction to several illegal drugs.

A fourth limitation is that we were unable to adjust the analyses for tobacco use disorder (F17) because the register data are unreliable, and we did not have additional questionnaire data on the disorder. In Denmark, a rather large proportion of the population are smokers, and until recently, the diagnosis of tobacco use disorder has not been systematically assessed in clinical practice, since the Danish health care system did not offer treatment for this disorder as standard procedure. Conversely, there is a long tradition in Denmark of providing inpatient and outpatient treatment for other types of SUDs. Consequently, tobacco use disorder is seldom assessed in register-based studies, as it is considered a poor indicator of how many people have developed the disorder (40). Furthermore, a separate examination of mortality risk among patients who crossed over from one eating disorder type to another was outside the scope of this study. In addition, since our eating disorder cohort was relatively young, we cannot draw any conclusions about how SUDs may effect mortality among older eating disorder patients. Lastly, because of sparse counts, we were unable to calculate cause-specific mortality risk for the specific eating disorder groups.

In summary, this study showed that SUDs involving alcohol, cannabis, and hard drugs had a negative additive effect on all-cause mortality risk in a large sample of patients with AN, BN, or unspecified eating disorder. The pattern of risk differed across the eating disorder types. Patients with AN or unspecified eating disorder both with and without SUDs exhibited an elevated risk compared with the respective control subjects without SUDs, but the risk was much higher in AN and unspecified eating disorder patients with SUDs, in particular hard drug abuse or dependence. For patients with BN, only those with SUDs exhibited an elevated risk of all-cause mortality compared with the respective control subjects, and hard drug abuse or dependence was associated with the greatest risk. These findings highlight the importance of focusing on the prevention and treatment of SUDs to reduce excess mortality in eating disorder patients. This is particularly relevant for AN patients who abuse hard drugs, since they were found to be the most susceptible to premature death. Interestingly, the driving factor behind mortality in BN patients appears to be SUD, and thus the prevention and treatment of SUDs would go a long way toward reducing mortality in BN.