Emotion dysregulation and suicidality in eating disorders

Abstract Objective Suicidality in eating disorders (EDs) is high, and identification of therapeutically targetable traits associated with past, current, and future suicidality is of considerable clinical importance. We examined overall and ED subtype‐specific associations among suicidal ideation, suicide attempts, and general and specific aspects of emotion dysregulation in a large sample of individuals with ED, at presentation for treatment and 1‐year follow‐up. Method Using registry data from 2,406 patients, scores on the Difficulties in Emotion Dysregulation Scale (DERS) at initial registration were examined as predictors of recent suicidal ideation and self‐report lifetime suicide attempts. Associations were examined in the full sample and in each ED subtype. In 406 patients, initial DERS scores were examined as predictors of suicidality at 1‐year follow‐up. Results Overall DERS was associated with suicidal ideation and suicide attempts, even when adjusting for ED psychopathology and current depression. Perceived lack of emotion regulation strategies showed unique associations with suicidal ideation and suicide attempts, both in the full sample and in most ED subtypes. Initial DERS was also associated with follow‐up suicidal ideation and suicide attempts, although this association did not remain when adjusting for past suicidality. Discussion Results suggest that emotion dysregulation may be a potential mechanism contributing to suicidality in EDs, beyond the effects of ED psychopathology and current depression. Although the prevalence of suicidality differs across ED subtypes, emotion dysregulation may represent a risk trait for future suicidality that applies transdiagnostically. Results support addressing emotion dysregulation in treatment in order to reduce suicidality.


| INTRODUCTION
Eating disorders (EDs) have amongst the highest standardized mortality ratios for death by suicide of any mental disorder (Chesney, Goodwin, & Fazel, 2014;Huas et al., 2013;Preti, Rocchi, Sisti, Camboni, & Miotto, 2011). In order to detect individuals at risk for suicide, address vulnerability factors in treatment, and prevent suicide, identification of traits associated with suicidal ideation and suicide attempts, hereafter referred to as suicidality, in this population is of considerable clinical importance.
Emotion dysregulation, characterized by difficulties in emotional awareness, clarity and acceptance, as well as difficulties managing emotions and refraining from impulsive behaviors when in distress (Gratz & Roemer, 2004), has been associated with suicidal behaviors in general and specifically in individuals with EDs (Gomez-Exposito et al., 2016;Pisetsky, Haynos, Lavender, Crow, & Peterson, 2017).
We explored associations between aspects of emotion dysregulation and suicidality in a large sample of patients across a range of ED subtypes.
Suicidal ideation (i.e., thinking about, considering, or planning suicide) often precedes suicide attempt (Nock et al., 2008), but has received less research attention in EDs. Lifetime suicidal ideation occurs in about one third of individuals with EDs (Favaro & Santonastaso, 1997;Milos, Spindler, Hepp, & Schnyder, 2004;Swanson, Crow, Le Grange, Swendsen, & Merikangas, 2011), and a nationwide Swedish ED treatment registry study indicated the highest prevalence estimates in  and the lowest in AN and other specified feeding and ED (OSFED; 26-30%). The presence of purging behavior is associated with greater risk of suicidality (Andersén & Birgegård, 2017;Swanson et al., 2011).
Using the Difficulties in Emotion Regulation Scale (DERS) (Gratz & Roemer, 2004) in 122 patients with bulimic symptoms (AN-BP, BN, BED, OSFED), Gómez-Expósito and colleagues reported that individuals with previous suicide attempts exhibited higher impairment than patients without suicide attempts in most aspects of emotion dysregulation, except difficulties in emotional awareness (Gomez-Exposito et al., 2016). Similarly, Smith and colleagues examined emotion dysregulation in 100 patients with a range of EDs (AN, BN, BED, OSFED, unspecified feeding and ED [UFED]) and found higher perceived lack of adaptive emotion regulation strategies in those who reported previous suicide attempts than in those who did not (Smith, Forrest, et al., 2018). Conversely, Pisetsky and colleagues reported no significant differences in aspects of emotion dysregulation between ED patients with and without suicide attempts (N = 110 patients; AN, BN, BED, OSFED), although a trend with small effect size emerged for higher levels of difficulties with impulse control and emotional clarity in those with lifetime suicide attempts (Pisetsky, Haynos, et al., 2017). Small sample sizes precluded subgroup analyses and these studies were cross-sectional.
The present study extends the existing literature by exploring associations between suicidality and emotion dysregulation in a large, well-characterized sample of individuals with a range of EDs. Specifically, we examined associations between general and specific aspects of emotion dysregulation assessed at initial registration for treatment and recent suicidal ideation and lifetime suicide attempts. As exploratory analyses, we also investigated ED subtype-specific associations between emotion dysregulation at initial registration and suicidality.
Lastly, we explored whether emotion dysregulation at initial registration predicted suicidality at 1-year follow-up in a subset of the sample with available data.
The sample was drawn from the Stepwise database that includes data from patients entering ED specialist treatment at any of 45 treatment units in Sweden since 2005 (Birgegård, Björck, & Clinton, 2010).
Stepwise inclusion criteria include referral to an ED treatment unit, a DSM-IV ED diagnosis (American Psychiatric Association, 2000), and established intent to treat. Stepwise initial assessment includes semistructured interviews, clinical ratings and self-reports (both mandatory and optional; DERS is optional); all recorded with software on clinicians' computers. ED diagnoses are based on the Structured Eating Disorder Interview that demonstrates good reliability and validity (de Man Lapidoth & Birgegård, 2010). The 45-min assessment is performed by trained professionals at the third clinic visit. The Stepwise 1-year assessment, similar to the initial assessment, is performed one year after the initial evaluation, within a 10-week window (±5 weeks). Attrition between initial and 1-year assessment occurs and can be due to both patient and clinician/treatment unit factors (drop-out, time constraints, follow-ups not encouraged). Stepwise attrition is around 40-64%, with a trend of higher attrition, the longer Stepwise has been running (Andersén & Birgegård, 2017;Ekeroth & Birgegård, 2014;Forsén Mantilla, Norring, & Birgegård, 2019).

| Participants
Initial data included 6,713 potential cases ≥13 years of age with a DSM-IV ED registered between April 7, 2014 (date when DERS was included in Stepwise), and October 16, 2019 (date of data extraction).

| Measures
The Structured Clinical Interview for DSM-IV axis I (SCID-I) (First, Spitzer, Gibbon, & Williams, 2002)  Suicidality at initial assessment and 1-year follow-up was extracted from the Riksät National Quality Registry for ED (embedded within Stepwise). Lifetime/follow-up suicide attempts were recorded using the question "Has the patient ever (initial registration) / since [date for initial registration] (1-year assessment) attempted suicide?" with fixed responses "Never", "1-2 times", and "more than 3 times".
Suicidal ideation was recorded using the question "How often during the past three months has the patient had suicidal thoughts, intentions or plans?" with the responses "Never", "Occasionally", and "Every week or more". Suicidal ideation and suicide attempts at both time-points were dichotomized such that "Never" was coded as 0 (=no suicidal ideation/suicide attempts), and alternatives "1-2 times" and "3 or more" for suicide attempts and "Occasionally" and "Every week or more" for suicidal ideation were combined into 1 (=any suicidal ideation/suicide attempt). Although mandatory, clinicians can note if sufficient information is lacking. At initial registration, small numbers of patients lacked information on suicidality (<1%); as no patient lacked information in both variables, all remained in the full sample. At follow-up, all patients had information on both suicidality outcomes.
The EDE-Q provides a mean Global Score and four subscales (restraint, eating concern, shape concern, and weight concern) with higher scores indicating greater pathology. The Swedish adult and adolescent ver- and Adolescents; OSFED, other specific feeding and EDs; SA, occurrence of lifetime suicide attempts; SA follow-up, occurrence of suicide attempts last 12 months; SCID-I, Structured Clinical Interview for DSM-IV axis I; SI, occurrence of suicidal ideation during the last three months; SI follow-up, occurrence of suicidal ideation the last three months.

| Statistical analyses
Differences between ED diagnostic subgroups on suicidality, DERS, and covariates were explored in the full and follow-up samples with χ 2 -tests for categorical and analysis of variance (ANOVA) for continuous variables. Associations between initial DERS and suicidality at initial registration (aim 1 and 2) and 1-year follow-up (aim 3) were examined using logistic regression. Several regression models were examined for each suicidality measure. To correct for multiple comparisons, the Bonferroni correction method was applied within families of tests (family defined as each aim).
Aim 1: Considering the full sample (i.e., all EDs at initial registration), DERS total score was examined as a predictor of lifetime suicide attempts and recent suicidal ideation at initial registration (Main analyses; Model 1, unadjusted). Then, all DERS subscales were entered simultaneously as predictors of lifetime suicide attempts and recent suicidal ideation, respectively (Secondary analyses; Model 2, unadjusted).
Aim 2: Model 2 was repeated in all ED diagnostic subgroups separately at initial registration (Exploratory analyses).
Aim 3: For the follow-up analyses, initial DERS total score was examined as a predictor of suicide attempts and suicidal ideation assessed at 1-year follow-up (Main analyses; Model 1, unadjusted) in the subset of participants with complete follow-up data (i.e., the follow-up sample). Then, initial DERS subscales were examined as predictors for follow-up suicide attempts and suicidal ideation (Secondary analyses; Model 2, unadjusted).
All models were re-fitted including age at initial registration, ED duration, EDE-Q Global Score, and depression status as covariates (Models 1 and 2, adjusted); 1-year models additionally included initial suicidality (suicide attempts and suicidal ideation assessed at initial registration). Statistical analyses were performed using the Statistical Package for the Social Sciences for Mac (SPSS-22/24).

| Sample representativeness
Sample representativeness was examined by comparing eligible patients with and without DERS at initial registration (Table S1), and participants in the full sample with and without follow-up assessment (  (Table S1) nor between adults with and without complete follow-up assessments (Table S2). Adolescents with and without complete follow-up differed in diagnostic distribution (mostly AN-R and OSFED), past year suicide attempt, BMI, age, and lower DERS awareness and clarity (Table S2).

| Sample characteristics
Sample description at initial registration and follow-up is shown in 3.2 | Associations between emotion dysregulation and suicidality at initial registration DERS total score was significantly associated with lifetime suicide attempts in the full sample, with a one-point increase in the total score indicating 1.6% higher odds of life-time suicide attempts (  Table S3).
No DERS subscale was significantly associated with lifetime suicide attempts in any ED diagnostic subgroup (Table 3;

| Associations between initial emotion dysregulation and suicidality at 1-year follow-up
Initial DERS total score was a significant predictor of follow-up suicide attempts, with a one-point increase in initial total score indicating 3.3% higher odds of at least one attempt during the following year (Table 5; Model 1). This association did not remain after accounting for covariates; occurrence of lifetime suicide attempts was the only significant predictor of follow-up attempts (OR 10.46, p = .001). Initial DERS total score was also a significant predictor of follow-up suicidal ideation, with a one-point increase in the total score indicating 2% higher risk of suicidal ideation at follow-up. After adjusting for covariates, this association did not remain significant; only initial suicidal ideation was significantly associated with suicidal ideation at follow-up (OR 6.65, p < .001).
When entering all initial DERS subscales simultaneously, no subscale predicted follow-up suicide attempts (Table 5; Model 2). When including covariates, only occurrence of lifetime suicide attempts was a significant predictor of follow-up attempts (OR 10.28, p = .001). Onepoint increase in initial strategies indicated 6% increased risk of occurrence of follow-up suicidal ideation. When accounting for covariates, only initial suicidal ideation was a significant predictor of follow-up suicidal ideation (OR 6.86, p < .001; all covariate values in Table S4).

| DISCUSSION
Emotion dysregulation could be a potential mechanism contributing to suicidality in EDs. In this large sample of patients with EDs, higher overall emotion dysregulation was associated with increased odds of both lifetime suicide attempts and recent suicidal ideation at initial registration. Further, higher perceived lack of adaptive emotion regulation strategies was uniquely associated with both suicidality outcomes, and higher difficulties in emotional awareness was associated Note: Model 1-2 run in the follow-up sample (N = 406). SA follow-up: suicide attempts last 12 months (0 = no; 1 = yes); SI follow-up: suicidal ideation the last three months (0 = no; 1 = yes). Abbreviation: DERS: Difficulties in Emotion Regulation Scale.
with suicidal ideation. Previous research on emotion dysregulation and suicidality is limited to comparisons between patients with or without lifetime suicide attempts (Gomez-Exposito et al., 2016;Pisetsky, Haynos, et al., 2017;Smith, Forrest, et al., 2018). Our findings partially corroborate those of Smith et al., highlighting lack of adaptive emotion regulation strategies; however, as our methodology enabled identification of unique associations with suicidality, results are not directly comparable. In general, suicidality prevalence in this sample was in line with previous findings (Franko & Keel, 2006;Milos et al., 2004;Smith, Zuromski, et al., 2018;Swanson et al., 2011;Udo et al., 2019).
Rather than ED diagnosis-specific associations between aspects of emotion dysregulation and suicidality, our results suggest more of a transdiagnostic pattern of perceived lack of strategies influencing suicidal ideation (except in BED). No emotion dysregulation measure was uniquely associated with suicide attempts in any diagnostic group, despite diagnostic differences in suicide attempts prevalence (i.e., AN-R lowest, AN-BP and BED highest). Of note, DERS-defined emotion dysregulation and suicidal ideation both refer to current cognitive-emotional processes at initial registration, in contrast to self-report lifetime suicide attempts referring to behaviors at any previous time point. Thus, associations with suicidal ideation were likely easier to detect. Moreover, the small sample size in each ED group, and the conservative correction for multiple comparisons may have limited the power to detect diagnosis-specific patterns of associations between emotion dysregulation domains and suicidality, especially in BED which was underrepresented.
Emotion dysregulation independently contributed to suicidality, beyond the effect of ED psychopathology and current depression.
Our results extend the understanding of suicidality in EDs; emotion dysregulation specifically impacted on suicidality even when controlling for relevant clinical variables. ED diagnoses and symptoms clearly impact on suicidality, with strong evidence of associations between suicidality and bulimic spectrum EDs and compensatory behaviors (Ahn, Lee, & Jung, 2019). Up to 45% of patients with EDs with a history of suicidality report having attempted suicide before ED onset, indicating that shared factors underlying ED and suicidality should be considered (Udo et al., 2019). Although present results confirm higher prevalence of suicidality in AN-BP, BED, and BN, the specific features of eating psychopathology do not appear to influence the association between emotion dysregulation and suicidality. Contrary to prior research (Ahn et al., 2019;Bulik et al., 2008;Pisetsky et al., 2013;Pisetsky et al., 2015;Udo et al., 2019), depression was not associated with lifetime suicide attempts in the multivariate analyses, although it was associated with recent suicidal ideation. This is surprising, given that depression is reported 4.32-15.06-fold times more frequently in patients with EDs who report a history of suicide attempts than in those without (Udo et al., 2019). Additionally, previous research on BN indicated depression as the salient factor for lifetime suicide attempts when examined with emotion dysregulation-related personality factors (Pisetsky et al., 2015). Our results, in contrast, revealed a marked association between emotion dysregulation and suicidality that was both stronger than, and independent of, depression. This lack of agreement could reflect different assessment of emotion related concepts (i.e., personality facets instead of DERS); furthermore, associations might have emerged with a continuous measure of depression instead of the dichotomous depression variable used here.
Emotion dysregulation at initial registration longitudinally predicted future suicidality, indicating emotion dysregulation as a risk trait. Although the small sample, higher initial overall emotion dysregulation was associated with increased risk of both suicidality outcomes in the year following initial assessment. Only one prior study has shown preliminary evidence of the role of emotion dysregulation in predicting future suicidality in EDs. Franko and colleagues explored a wide range of clinical variables at initial assessment as predictors of suicide attempts over the following 9 years in individuals with AN and BN; those with BN exhibiting greater impairment in identifying internal states (i.e., similar to DERS awareness and/or clarity) had a greater risk of suicide attempts (Franko et al., 2004). Although difficulties in identifying internal states cannot be considered as a proxy of the entire emotion dysregulation construct, the study by Franko et al. was a unique contribution to understanding longitudinal correlates of suicidality in EDs. The inclusion of initial suicidality in the models diminished the association of emotion dysregulation with future suicidality. This is consistent with prior studies identifying prior suicidality as the strongest predictor of future suicidality (Cavanagh, Carson, Sharpe, & Lawrie, 2003;Franko et al., 2004;Harris & Barraclough, 1997). Its replicated strength as a predictor should not detract from exploring the role of emotion dysregulation in being a clinical warning sign for future suicide risk.
The associations between emotion dysregulation and past, and follow-up suicidality may relate to the interpersonal theory of suicide (ITPS) (Joiner, 2005 In order to prevent suicide in EDs, identification of therapeutically targetable traits associated with past and future suicide attempts is of considerable importance. Our results suggest that emotion dysregulation may represent such a trait that applies transdiagnostically. Although prior suicide attempt remains the most robust predictor of subsequent attempts, emotion dysregulation is both measurable and targetable therapeutically. Even though the mechanisms underlying the association between emotion dysregulation and suicidality in EDs are not fully understood (e.g., mediator, shared underlying psychological processes), both ED symptoms and suicidal behaviors are negatively reinforced by providing temporary relief from negative emotions-at the expense of more adaptive regulatory strategies (Skinner, Rojas, & Veilleux, 2017). Targeting emotion dysregulation may be beneficial in ED treatment for patients with and without past suicidality. A review of emotion dysregulation-oriented interventions for various psychiatric disorders (e.g., ED, depression, anxiety, borderline personality disorder) revealed that improving emotion regulation skills was associated with decreases in both the specific pathology being targeted as well as comorbid psychopathology (Sloan et al., 2017). Using such approaches to address ED symptoms may also serve to reduce suicidality. Emotion regulation-focused therapies such as Dialectical Behavior Therapy, Emotion Acceptance Behavior Therapy, and Integrative Cognitive-Affective Therapy have been developed or adapted for EDs (Berg & Wonderlich, 2013). Whether they reduce both ED pathology and suicidality remains to be examined.
Study strengths include a large, ecologically valid sample at initial registration. The sample also included a wide range of DSM-defined EDs, strengthening the representativeness of the results and enabling analyses of diagnostic subgroups. Several limitations should be considered. The DERS was optional, and clinicians' decisions to include DERS were not recorded. However, previous analysis of missing data showed that clinicians typically choose no optional measures (Monell et al., 2018), suggesting that rather than choosing measures based on patient characteristics, clinician/clinic variables (e.g., interest, time constraints, unit specific assessment routines) seemed most influential. Although no meaningful differences between patients with and without DERS emerged, unmeasured differences could have introduced bias. Even though depression assessment is mandatory, 16% lacked information on depression, meaning sample sizes for adjusted models were smaller. Moreover, in order to include adolescents, we had to dichotomize the depression variable, reducing statistical power.
Similarly, suicidality variables were dichotomized, since the structure of the Riksät suicidality response options does not lend itself to ordinal quantification (i.e., never, occasionally, weekly for suicidal ideation; never, 1-2, ≥3 for suicide attempts), again reducing power. Further, we had limited information on the timing and seriousness of previous suicide attempts. In the follow-up sample, only 4% and 28% reported suicide attempts and ideation, respectively; thus, analyses may have had too low power to detect a significant association between emotion dysregulation and suicidality beyond the effect of previous suicidality. Substantial attrition at one-year follow-up potentially threatened the representativeness of these analyses. However, multiple studies using the Stepwise registry reported small to negligible differences between patients with and without follow-up data, indicating that factors related to treatment units rather than patient factors lead to attrition. We observed some minor differences between adolescents with and without follow-up, indicating that our follow-up results may be more generalizable to adolescents with AN-R and those without past year suicide attempts (as these groups were more likely to have follow-up data), and to those with slightly more difficulties in emotional clarity and awareness. We were unable to link our data to the Swedish Death Registry. Accordingly, other deaths could have occurred but not been recorded by the clinician.
Lastly, as the sample comprised Swedish patients seeking active treatment, results may not generalize to nontreatment seeking groups or to more culturally diverse samples.

| CONCLUSIONS
Suicidality in EDs is high, and robust and clinically relevant predictors of future attempts are needed. Emotion dysregulation was associated with both lifetime suicide attempts, recent ideation, and suicidality at 1-year follow-up in patients with a wide range of EDs, even when ED psychopathology and depression were accounted for. Results suggest that although suicidality differed across different EDs, emotion dysregulation may be a transdiagnostic trait influencing suicidality. Finally, these results encourage further longitudinal studies examining the specific contribution of emotion dysregulation to suicidality in EDs.