Suicide attempts in US adults with lifetime DSM-5 eating disorders

In 1999–2016, the rates of suicide in the United States (US) increased in all states, except for Nevada, with more than 30% increase in 25 states, and this increasing pattern was observed regardless of sex, racial/ethnic groups, and urbanity [1]. Since 2008, suicide has remained the 10th leading cause of deaths in the US, [2] making suicide prevention an urgent public health issue. Research has increasingly focused on identifying risk factors and predictors of suicidal behaviors and suicide completion; identifying reliable variables, however, has been challenging for many reasons including methodological limitations (e.g., variability in target populations, cross-sectional vs. longitudinal, short follow-up periods) [3, 4]. Nonetheless, critical and meta-analytic reviews of the literature have identified some variables that may predict suicide risk reasonably consistently across studies [3]. For example, certain sociodemographic characteristics have been associated with increased risk for suicide, including male gender, sexual orientation, race (being Caucasian), and lower education [3, 5]. Psychiatric illnesses are also one of the strongest dynamic, malleable factors associated with suicide [3, 6]. Systematic reviews have consistently reported that approximately 90% of individuals had diagnosable psychiatric disorders at the time when they died by suicide [7, 8].

Mood disorders, substance use disorders (SUD), and anxiety disorders are the most common predictors of suicide [9–11]. In addition, analyses of psychiatric-patient samples have reported that eating disorders (EDs) also appear to be associated with suicide [12, 13]. EDs, which includes include anorexia nervosa (AN), bulimia nervosa (BN), and binge eating disorder (BED) according to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5, [14]), are prevalent psychiatric disorders that occur across age groups, sex, and race/ethnicity [15, 16]. Preti and colleague [17] reported that among 16,342 patients with AN pooled from 40 studies, the standardized mortality ratio (SMR; i.e., mortality risk relative to age- and gender-matched general population) was 31.0 (95% CI = 21.0–44.0, mean follow-up = 11.1 years). The same study reported that among 1768 patients with BN pooled from 16 studies, SMR was 7.5 (95% CI = 1.6–11.6, mean follow-up = 7.5 years); it was not possible to calculate a SMR for BED because there were only 3 studies and none reported suicide incidence [17].

A recent meta-analysis of 2611 longitudinal studies also found that ED diagnosis was associated with significantly increased risk for suicide attempt (SA) [18] although the rate of SA varied considerably across studies. In clinic samples, the rate of SA in patients with AN ranges from 3 to 20% [19], 25–35% in BN [19], 12.5% in BED [20], and 20.8% when combining all EDs [21]. Crow et al. [22], which specifically examined suicidal behaviors in US adolescents and adults with data from the National Comorbidity Survey-Replication (NCS-R; N = 2980) [23] and the National Comorbidity Survey-Adolescent Supplement (NCS-A; N = 10,123) [24, 25], estimated that the prevalence of SA in adults was 8.8% in AN, 21.3% in BN, and 15.2% in BED, whereas it was only 1.2% in adults without a lifetime psychiatry disorder.

Collectively, these studies suggest associations between EDs and suicidal behaviors, although there remain important questions to be answered. First, because BED was not a formal specific diagnosis until DSM-5, most prior studies included BED as part of the larger and heterogeneous “eating disorder not otherwise specified” category. Research on the relationship between suicidal behaviors and BED has thus been limited relative to AN or BN. In their systematic review, Conti and colleagues [26] considered 17 original studies on the relationship between BED and suicidal behaviors; the prevalence of suicide ideation ranged from 26.3 to 51.7% and the prevalence of SA ranged 5.5 to 33.6% across the various studies. The five studies that considered BED reported significantly increased odds of suicidal behaviors relative to individuals without BED [26]; all these studies, however, were based on previous DSM-IV criteria, and most focused on patients with BED and comorbid conditions. Furthermore, besides several studies that examined comorbid psychiatric conditions [21, 27, 28], body mass index (BMI) [21, 29], and ED pathologies [21, 29] as risk factors for SA in EDs, research on correlates of suicidal behaviors in EDs has been narrow and limited.

Most previous studies on suicide and EDs were based on clinical samples, which suffer from various treatment-seeking and selection confounds [30–32]. The very few available epidemiological studies often did not include EDs, particularly those conducted based in the US. For example, there are two studies that reported significantly increased risk for suicidal attempts [10, 11] and suicidal ideation [33] in individuals with psychiatric disorders using the NCS-R and the WHO World Mental Health Surveys, but neither considered EDs in their analyses of suicide behaviors, making an earlier mentioned Crow et al. [22] study the only epidemiological study on the relationship between EDs (but specifically BN) and suicide risk in the US adults. Thus, there exist no epidemiological studies in the US that have reported on the relationship between DSM-5-defined EDs and suicide. A number of substantive changes in diagnostic criteria for EDs were made between DSM-IV-TR and DSM-5. In DSM-5, the AN diagnosis no longer requires amenorrhea and now defines low-weight as less than minimally normal for age. The BN diagnosis now has a frequency requirement of once-weekly for binge eating and weight-compensatory behaviors, which is lower than the twice-weekly frequency requirement in the DSM-IV. BED, now a “formal” diagnosis, is also defined with a lower frequency requirement of once-weekly binge eating for a duration of 3 months (less than a 6-month criterion in previous research using the DSM-IV “research definition”) to parallel the BN diagnosis. These substantive changes across the DSM-5-defined EDs dictate the need for further research on the prevalence and correlates of SA in this group.

The National Epidemiologic Survey on Alcohol and Related Conditions III (NESARC-III) [34] is the first large psychiatric epidemiology study with a representative sample of US adults since DSM-5 was published. Using the NESARC-III, the goals of the present study were to (1) obtain the prevalence estimates of SA in US adults with lifetime DSM-5 ED diagnoses, (2) examine sociodemographic, clinical, and psychiatric correlates of SA in EDs, and (3) examine temporary relationships between DSM-5 ED diagnoses and first SA. In addition to the specific ED diagnoses of AN, BN, and BED, we aimed to also examine the two subtypes of AN—restricting and binge/purging subtypes of AN (AN-R and AN-BP, respectively)—given prior findings suggesting that prevalence of suicide completion was greater for AN-BP than for AN-R [29].

Using the data from the 2012–2013 NESARC-III, this was the first study to examine the prevalence of SAs in a representative sample of the US adults with lifetime DSM-5 ED diagnoses. SAs are common in US adults with lifetime AN, BN, and BED diagnoses. We generated the following prevalence estimates for SAs across the ED diagnosis: 24.9% for AN, 31.4% for BN, and 22.9% for BED; importantly, for AN, the prevalence rate for SAs was substantially higher for the AN-BP subtype (44.1%) than for the AN-R subtype (15.7%). Adjusting for sociodemographic variables, compared with those without lifetime specific-ED diagnosis, odds of SAs were 5.40, 6.33, and 4.83 times higher in respondents with lifetime AN, BN, and BED, respectively; adjusted odds of SAs calculated separately for AN-BP and AN-R subtypes were 12.09 and 3.16, respectively. The total number of SAs was also significantly greater in respondents with lifetime specific-ED diagnosis. In those with SA history, reported age at first SA or the number of SAs were not significantly different by lifetime ED diagnoses, except for greater number of SAs in BED.

This study also attempted to address possible implications of comorbidity among different EDs as well as diagnostic cross-over [42–44] in a series of supplemental analyses. We found that the odds of SAs, although remaining statistically significant, decreased for BN and BED when we applied the hierarchical convention of AN>BN>BED for assigning ED diagnoses (Additional file 2). Among respondents who met more than one lifetime ED diagnosis, the prevalence and odds of SAs were significantly elevated (41.5% and AOR = 10.63, respectively), relative to respondents without lifetime history of specific EDs (Additional file 3). These findings are similar to those previously reported by Pisetsky et al. [29] that the highest odds of suicidal behaviors were found in those who met criteria for AN and BN. We note, however, that a more recent larger cohort study with treatment-seeking adults from the Swedish registry (N = 9622) found that transition between AN, BN, and BED was rare [45]. Our study also found that the comorbidity rate for different lifetime ED diagnosis was only 12% of those who met lifetime diagnosis of any EDs. Collectively, these findings suggest that while comorbidity or transition across ED diagnoses might be less common than clinically presumed, those individuals who do transition between different ED diagnoses may be at higher risk for suicide attempts.

Prevalence of SAs in our sample of US adults with lifetime DSM-5-defined EDs was higher than that of previously reported by Crow et al. [22] based on the smaller NCS-R which used DSM-IV definitions (8.8% in AN, 21.3% in BN, and 15.2% in BED). The reasons for differences are not clear and may be due partly to several methodological differences between the two studies, in addition to the many complexities inherent in assessing and generating diagnostic estimates in large epidemiological studies [46]. One notable difference between two studies is that while a question of ever attempting suicide was asked for all NESARC-III respondents, in the NCS-R this question was asked only when respondents endorsed suicidal ideation. Given the two surveys were conducted over 10 years apart, it is also possible that some of the difference partly reflect general increases in prevalence of suicide in the US [1, 47].

Within each ED category, there were only a few significant differences in sociodemographic characteristics between those who have ever versus never attempted suicide, a pattern that differs somewhat from in the general population [5]. Nonetheless, there were several significant differences in clinical characteristic between those with and without SA histories. In BN and BED, SA history was associated with earlier age of ED onset, which is partially consistent with findings from a clinical sample [21]. In all EDs, the lowest BMI was lower in respondents with a SA history than those without SA history. SA history was significantly associated with longer duration (more years) of ED episodes in AN, whereas it was associated with significantly shorter duration of ED episodes in BN. These findings, which require replication and extension in future studies, are curious and a challenge to explain. One possible clinical speculation is that SA in AN might perhaps be related to chronicity whereas SA in BN might be more related to severity. Given the relatively small sample sizes when focusing only on respondents with ED diagnoses, our study was likely underpowered to adequately statistically examine such potential reasons. We also found that among those with lifetime AN, the proportion of those reporting that ED symptoms interfered with normal daily activities or caused serious problems fulfilling responsibilities was higher among those with SA history. In addition, among those with lifetime BN, the proportion of those reporting that ED symptoms caused serious problems with fulfilling responsibilities was higher among those with SA histories. Among those with BED, the prevalence of psychosocial impairments did not differ significantly by SA history.

Consistent with previous studies [21, 22, 27, 28], SA history was significantly associated with lifetime diagnoses of other psychiatric disorders. Mood disorders, SUD, and personality disorders were prevalent across all EDs with a history of SA, with 73–96% meeting criteria for those diagnoses. Prevalence of anxiety disorders was also high in BED, with over 70% meeting criteria for diagnosis at some point in their lifetime. Within each specific ED, odds for all comorbid psychiatric disorders, even after adjusting for sociodemographic characteristics, were substantially elevated in those with SA history relative to those without SA history. This was particularly prominent in BN where many of the significant AORs were greater than 10. Further research, particularly with longitudinal designs, is necessary to replicate our findings and to examine how elevated psychiatric comorbidity in BN may contribute to future SAs as well as to other negative health and psychosocial outcomes.

In terms of sequence of ED and SA onsets, we found that for the majority of respondents with BED (71.2%), BED onset preceded the first SA. In contrast, among respondents with AN and BN, approximately 50% reported ED onset prior to the first SA. The age of onset is significantly later in BED than AN and BN [16], and the observed difference between the three forms of ED diagnoses may reflect differences in the clinical course. Our finding on BED onset preceding SA onset contrasts with a study based on the NCS-R that found that their adult respondents (in contrast to their findings for adolescents) reported SA prior to onset of BED [48]. We, however, note that due to missing data, that sample size in the Forest et al. study was extremely small (n = 1 for adults 18–29, n = 3 for 30–44, and n = 3 for 45–59, and n = 1 for 60+ years old), suggesting need for caution in interpreting those findings.

Our findings suggest that many individuals, particularly those with BED, report experiencing SAs prior to the onset of their ED symptoms. This is consistent with a recent meta-analysis by Smith [18], suggesting that while ED diagnosis and disordered eating are significantly associated with SA, they are only weak predictors of SA. Given high comorbidity between EDs and other psychiatric disorders, it is possible that the relationship between EDs and SA may be mediated by mood disorders and/or SUD, which are established significant predictors of SA [33]. Our study was cross sectional and did not investigate the temporal relationship between EDs, other psychiatric disorders, and SAs. Our findings and the findings from the prior studies highlight the need for further research specifically focused on the relationship between ED and suicide behaviors. With these caveats in mind, we state that our findings of significantly increased risk for SAs in those with lifetime EDs and that ED onset preceded first SAs in roughly half of ED cases highlights the importance of assessing and closely monitoring ED symptoms along with other psychiatric disorders by health-care providers and clinicians.

Our study also found several important differences between the two AN subtypes. We found that 32.3% (SE = 2.88) of those with lifetime AN-BP (prevalence among the total NESARC sample: AN-R = 0.54% [SE = 0.05], AN-BP = 0.26% [SE = 0.03]). Thus, when considered in this manner, across the EDs included in this study, AN-BP had the highest risk for SA whereas AN-R had the lowest, albeit significantly elevated. Our findings regarding especially elevated risk of SA in AN-BP are consistent with previous findings [29, 49].

AN-R and AN-BP subtypes of AN also had several different patterns of significant associations between SA history and other psychiatric disorders, though interpretations of those “correlates” are more complicated due to various reasons, including the fact that it is not possible to examine the complex patterns of timing associated with all the lifetime psychiatric comorbidities. With that caveat, we cautiously note some overall differences in the lifetime comorbidity findings associated with SA for the AN subtypes. In AN-R, SA history was associated with significantly elevated risk for bipolar I disorder, all forms of SUDs, and all forms of personality/conduct disorders. In contrast, in AN-BP, SA history was associated with any anxiety disorder, panic disorder, and specific phobia. Clinically, AN-BP is thought to be associated with higher impulsivity and substance use than AN-R [49–51], although differences between naturalistic, epidemiological, and treatment-seeking patterns of comorbidity may make comparisons difficult. Treatment-seeking confounds and/or motivations might, for example, partly explain why clinically AN-BP and BN show greater impulsivity and problems such as SUD, in that those externalizing and florid problems lead to seeking help as opposed to the restriction/control leading to decreased help-seeking in those with AN-R. Thus, it is possible that in naturalistic samples, SA history is associated with increased risk for SUD in those persons with AN-R. Lastly, we note that the risk for major depressive disorder was especially elevated in AN-BP with SA histories, with observed AORs similar to observed in BN.

Our findings regarding elevated risk for SA in AN-BP and BN are consistent with findings reported by Pisetsky et al. [29] that the highest rates of suicide completion occurred in those forms of EDs involving purging behaviors (comorbid AN and BN, AN-BP, BN, and purging disorder). The associations between SA and major depressive disorder were significantly increased in all EDs, and they were also particularly increased in AN-BP and BN. Thus, it is possible that the elevated risk for SAs in these forms of EDs may be partly influenced by depression. Diagnostic and clinical utility of AN subtypes has long been debated as clinical research did not always support predictive validity (e.g., [49, 51]). Our findings, however, suggest that two AN subtypes (AN-BP and AN-R) differ significantly in terms of SA risk and correlates of SA; these findings support the DSM-5 retention of the subtyping scheme.

Strengths of the current study include use of the largest psychiatric epidemiology survey on a representative sample of US adults and that all psychiatric disorders were assessed with a lay-administered structured interview. A possible limitation is that the validity and reliability of the AUDADIS-5 has not been tested for EDs, although it has been tested for all other psychiatric conditions and the findings are adequate [37, 39]. This is a cross-sectional study that retrospectively assessed lifetime history of ED symptoms and thus is susceptible to recall bias, although use of a structured diagnostic interview might have attenuated such effects to some degree. Due to the patterns of skip-out questions and missing data, we were unable to include subthreshold BN and BED or otherwise specified feeding and eating disorders (OSFED). There are other possible assessment limitations with the AUDADIS-5 that might have contributed to lower prevalence of EDs in general in this study than other reports [16, 46]. Previous studies on EDs and suicidal behaviors have also investigated the relationship between SA and specific eating pathologies, such as compensation methods (e.g., vomiting, purging) [21, 22, 29], but we were unable to explore such possible relationship with SAs due to the nature of the AUDADIS-5 questions for BN or BED [16, 52]. Relative to AN and BED, the sample size of respondents with lifetime BN diagnosis was smaller, which may have contributed to some non-statistically significant results (e.g., AORs for risk for SAs in those with and without lifetime BN diagnosis); the smaller sample size also precluded us from investigating possible sex or ethnic/racial differences. Similarly, we were unable to obtain stable estimates of the age-cohort effects on the relationship between EDs and SA, which has been recommended when investigating lifetime prevalence of psychiatric disorders with a mix-age sample [53, 54].

Given that suicide remains a major cause of deaths in the US, it is crucial to develop better approaches for earlier identification and recognition of risk, which may, in turn, enhance suicide prevention strategies across the lifespan and across different clinical groups that might differ in their suicide risk. Our findings suggest that adults with a lifetime diagnosis of EDs are at substantially elevated risk for SA and suggest that health-care workers consider these findings in their screening and evaluation protocols. The prevalence of SAs in our epidemiological sample was similar or greater than those reported for clinical samples [19–21]. This is concerning as treatment utilization is very low among adults with diagnosable EDs in the general population, particularly among men and ethnic/racial minorities [30–32, 55]. Thus, routine screening for EDs along with SA history by health-care providers could inform comprehensive treatment planning and appropriate treatment referrals.