Intensive inpatient treatment for bulimia nervosa: Statistical and clinical significance of symptom changes.

Objective: This study examines the statistical and clinical significance of symptom changes during an intensive inpatient treatment program with a strong psychotherapeutic focus for individuals with severe bulimia nervosa. Method: 295 consecutively admitted bulimic patients were administered the Structured Interview for Anorexic and Bulimic Syndromes-Self-Rating (SIAB-S), the Eating Disorder Inventory-2 (EDI-2), the Brief Symptom Inventory (BSI), and the Beck Depression Inventory-II (BDI-II) at treatment intake and discharge. Results: Results indicated statistically significant symptom reductions with large effect sizes regarding severity of binge eating and compensatory behavior (SIAB-S), overall eating disorder symptom severity (EDI-2), overall psychopathology (BSI), and depressive symptom severity (BDI-II) even when controlling for antidepressant medication. The majority of patients showed either reliable (EDI-2: 33.7%, BSI: 34.8%, BDI-II: 18.1%) or even clinically significant symptom changes (EDI-2: 43.2%, BSI: 33.9%, BDI-II: 56.9%). Patients with clinically significant improvement were less distressed at intake and less likely to suffer from a comorbid borderline personality disorder when compared with those who did not improve to a clinically significant extent. Conclusions: Findings indicate that intensive psychotherapeutic inpatient treatment may be effective in about 75% of severely affected bulimic patients. For the remaining non-responding patients, inpatient treatment might be improved through an even stronger focus on the reduction of comorbid borderline personality traits.

Objetivo: Este estudo examina a significância estatística e clínica da mudança sintomática durante um programa intensivo de tratamento hospitalar com forte foco psicoterapêutico para indivíduos com bulimia nervosa grave. Método: 295 pacientes bulímicos foram admitidos consecutivamente e responderam a Structured Interview for Anorexic and Bulimic Syndromes-Self-Rating (SIAB-S), o Eating Disorder Inventory-2 (EDI-2), o Brief Symptom Inventory (BSI), e o Beck Depression Inventory-II (BDI-II) no início e no fim do tratamento. Resultados: Indicaram redução sintomática estatisticamente significativa com tamanhos de efeito maiores em relação à gravidade da compulsão alimentar e comportamento compensatório (SIAB-S), gravidade do sintoma de transtorno alimentar geral (EDI-2), psicopatologia geral (BSI) e gravidade dos sintomas depressivos (BDI-II) mesmo quando controlado para medicação antidepressiva. A maioria dos pacientes mostrou ou mudança sintomática confiável (EDI-2: 33.7%, BSI: 34.8%, BDI-II: 18.1%) ou mesmo mudança sintomática clinicamente significativa (EDI-2: 43.2%, BSI: 33.9%, BDI-II: 56.9%). Pacientes com melhora clinicamente significativa apresentaram menor sofrimento psicológico no início do tratamento bem como eram menos propensos a apresentar transtorno de personalidade Borderline como comorbidade quando comparados com aqueles que não melhoraram de forma clinicamente significativa. Conclusões: Os achados indicam que o tratamento psicoterapêutico intensivo hospitalar pode ser eficaz em cerca de 75% dos pacientes com bulimia grave. Para os demais pacientes que não respondem, o tratamento hospitalar pode ser melhorado através de um foco ainda mais forte na redução dos traços de personalidade Borderline comórbidos.

Ziel: Diese Studie untersuchte die statistische und klinische Signifikanz von Symptomveränderungen während eines intensiven stationären Behandlungsprogramms mit einem starken psychotherapeutischen Fokus für Individuen mit einer schweren Bulimia nervosa. Methode: Bei 295 aufeinanderfolgend aufgenommenen Bulimiepatienten wurden das Strukturierte Interview für Anorektische und Bulimische Essstörungen – Selbsteinschätzung (SIAB-S), das Eating Disorder Inventory-2 (EDI-2), das Brief Symptom Inventory (BSI) und das Beck-Depressions-Inventar-II (BDI-II) bei Aufnahme und Entlassung eingesetzt. Ergebnisse: Die Ergebnisse zeigten statistisch signifikante Abnahmen von Symptomen mit großen Effektgrößen bezüglich der Schwere von Essattacken und kompensatorischem Verhalten (SIAB-S), der allgemeinen Symptomschwere der Essstörung (EDI-2), der allgemeinen Psychopathologie (BSI) und der depressiven Symptomschwere (BDI-II), selbst wenn für die Einnahme von antidepressiver Medikation kontrolliert wurde. Die Mehrzahl der Patienten zeigten entweder reliable (EDI-2: 33,7%, BSI: 34,8%, BDI-II: 18,1%) oder sogar klinisch signifikante Symptomveränderungen (EDI-2: 43,2%, BSI: 33,9%, BDI-II: 56,9%). Patienten mit einer klinisch signifikanten Verbesserung waren bei der Aufnahme weniger belastet und hatten eine geringere Wahrscheinlichkeit an einer komorbiden Borderline-Persönlichkeitsstörung zu leiden im Vergleich zu Patienten, die sich nicht klinisch signifikant verbesserten. Schlussfolgerungen: Die Befunde legen nahe, dass eine intensive stationäre psychotherapeutische Behandlung bei ca. 75% der Patienten mit einer schweren Bulimie wirksam sein kann. Bei den übrigen Patienten, die nicht auf die Behandlung ansprechen, könnten stationäre Behandlungen möglicherweise durch einen noch stärkeren Fokus auf die Reduktion komorbider Merkmale der Borderline-Persönlichkeitsstörung verbessert werden.

目標：本研究針對嚴重神經性暴食症患者，實施以心理治療為主要焦點的密集式治療方案，並檢視這些患者症狀改變的統計和臨床顯著程度。方法：針對295位連續住院的暴食症患者，在治療初談和離院時實施厭食與暴食症狀的結構性訪談—自我評量（SIAB-S），飲食偏差量表-2（EDI-2），簡式症狀量表（BSI）、以及貝克憂鬱量表--二版 （BDI-II）。結果：結果顯示症狀的減輕達到統計顯著性，並在狂吃和補償行為（SIAB-S）、整體飲食偏差症狀嚴重性（EDI-2）、整體心理病理（BSI）、以及憂鬱症狀嚴重性（BDI-II）等分量表上呈現極大的效果量，即使控制使用抗憂鬱藥物。大部分的患者呈現既可靠（EDI-2: 33.7%, BSI: 34.8%, BDI-II: 18.1%），又達到臨床顯著性的症狀改變（EDI-2: 43.2%, BSI: 33.9%, BDI-II: 56.9%）。當與進步未達到臨床顯著性的患者比較，臨床上顯著進步的患者在初談時沮喪程度較低，也較不可能同時患有邊緣性人格疾患的併發症。結論：研究發現密集性心理治療式住院治療可能對大約75%的嚴重暴食症患者具有效果。對於其他反應不佳的患者，透過更加聚焦於邊緣性人格特質併發症的降低，住院治療可能有所進展。

Obiettivo: questo studio esamina la significatività statistica e clinica dei cambiamenti sintomatologici nel corso di un programma di trattamento intensivo di pazienti ricoverati, con una forte attenzione psicoterapeutica per soggetti affetti da bulimia nervosa grave. Metodo: 295 pazienti bulimici ricoverati consecutivamente sono stati sottoposti alla Structured Interview for Anorexic and Bulimic Syndromes- Self-Rating (SIAB-S), all'Eating Disorder Inventory-2 (EDI-2), al Brief Symptom Inventory (BSI) e al Beck Depression Inventory-II (BDI-II) al momento all'inizio del trattamento e alla dimissione. Risultati: i risultati hanno indicato una riduzione dei sintomi statisticamente significativa con ampio effect size per quanto riguarda la gravità delle abbuffate e del comportamento compensatorio (SIAB-S), la gravità dei sintomi del disturbo alimentare (EDI-2), la psicopatologia generale (BSI) e la gravità dei sintomi depressivi (BDI- II) anche quando i dati sono stati controllati per il trattamento antidepressivo. La maggior parte dei pazienti mostrava cambiamenti sintomatologici affidabili (EDI-2: 33,7%, BSI: 34,8%, BDI-II: 18,1%) o persino clinicamente significativi (EDI-2: 43,2%, BSI: 33,9%, BDI-II: 56,9%). I pazienti con miglioramento clinicamente significativo mostravano un disagio inferiore in fase iniziale e minore probabilità di soffrire di un disturbo di personalità borderline in comorbilità rispetto a quelli che non miglioravano in modo clinicamente significativo. Conclusioni: i risultati indicano che il trattamento psicoterapeutico intensivo di pazienti ricoverati può essere efficace in circa il 75% di pazienti affetti da bulimia grave. Per i restanti pazienti che non rispondono, il trattamento ospedaliero potrebbe essere migliorato attraverso un focus ancora più forte sulla riduzione dei tratti di personalità borderline comorbilità.

Keywords: bulimia nervosa; intensive inpatient treatment; cognitive-behavioral therapy; statistical and clinical significance; symptom severity; borderline personality disorder; tratamento intensivo hospitalar; terapia cognitiva comportamental; significância estatística e clínica; gravidade dos sintomas; transtorno de personalidade Borderline; Bulimia nervosa; intensive stationäre Behandlung; kognitive Verhaltenstherapie; statistische und klinische Signifikanz; Symptomschwere; Borderline-Persönlichkeitsstörung; 神經性暴食症; 密集式住院治療; 認知行為治療; 統計和臨床顯著性; 症狀嚴重性; 邊緣性人格疾患; trattamento intensivo ospedaliero; terapia cognitivo-comportamentale; significatività statistica e clinica; gravità dei sintomi; disturbo borderline di personalità

Bulimia nervosa (BN) is a severe eating disorder characterized by an inordinate preoccupation with weight and recurrent episodes of binge eating followed by compensatory behaviors, such as self-induced vomiting, abuse of laxatives, and excessive exercise (American Psychiatric Association [APA], [2]). Prevalence rates among the female population range between 0.9% and 1.5% and are estimated to be even higher under the new DSM-5 criteria (Smink, van Hoeken, & Hoek, [83]; Stice, Marti, & Rohde, [86]). BN is frequently accompanied by comorbid disorders (Keski-Rahkonen et al., [51]; Swanson, Crow, Le Grange, Swendson, & Merikangas, [87]) and associated with significant functional impairment (Kessler et al., [52]; Stice et al., [86]) and poor quality of life (DeJong et al., [19]). Course of illness is often recurrent or even chronic (Keel & Brown, [48]; Steinhausen & Weber, [85]; Stice et al., [86]).

Cognitive-behavioral therapy (CBT) and selective serotonin reuptake inhibitors (SSRIs) have been recommended as the treatment of choice for this debilitating mental health problem (APA, [1]; National Institute for Health and Care Excellence, [66]). Fortunately, empirical evidence for the statistical significance of symptom changes during CBT and medical treatment with SSRIs in outpatient settings is strong (see for an overview, Krüger & Kennedy, [53]; Lewandowski, Gebing, Anthony, & O'Brien, [58]; and Mitchell, Peterson, Myers, & Wonderlich, [64]; for further single studies, see Goldstein, Wilson, Ascroft, & Al-Banna, [36]; Jones & Clausen, [47]; Milano, Petrella, Sabatino, & Capasso, [62]; Poulsen et al., [72]; Romano, Halmi, Sarkar, Koke, & Lee, [76]; Shapiro et al., [82]; Thackwray, Smith, Bodfish, & Meyers, [88]; Wilson & Zandberg, [95]; Wonderlich et al., [96]). However, analyses taking the variability of treatment response within a sample into account show that up to 80% of patients do not reach remission during outpatient treatment (Chen et al., [15]; Mitchell et al., [63]; Schnicker, Legenbauer, & Hiller, [81]; Thompson-Brenner, Glass, & Westen, [90]) and that up to 49% do not respond to any treatment (Ghaderi, [35]; Schnicker et al., [81]).

According to German treatment guidelines, bulimic patients should be treated in specialized inpatient units (i) if they are severely disturbed regarding eating and compensatory behavior; (ii) if they are non-responding to previous (outpatient) treatment(s); (iii) if they are suffering from severe comorbid conditions; and (iv) if they live in a treatment-impeding environment (DGPM, DKPM, DÄVT, DGKJP, DGPPN, DGPs, DGVM, [22]; only a German version of the guidelines exists). Unfortunately, evidence for the statistical significance of symptom changes due to CBT and SSRIs in inpatient settings is scant with only four studies indicating its efficacy in uncontrolled pre-post designs (Bossert, Schnabel, & Krieg, [10]; Fichter et al., [31]; Fichter & Quadflieg, [25], [30]). Furthermore, only one study has investigated the effectiveness of inpatient treatment for BN on an individual level. This study showed that 48% of patients were in partial remission and 33% in full remission at discharge (Zeeck et al., [97]). However, in that study, symptom changes during treatment were not taken into account when estimating treatment outcome.

Thus, in the present study, we aimed to investigate the statistical and clinical significance of symptom changes during an intensive inpatient treatment for severely disturbed bulimic patients according to the criteria of Jacobson and Truax ([45]). When estimating treatment outcome according to these criteria, both treatment response and shifts into the functional population range are taken into account. Thereby, patients are classified into four specific treatment outcome groups: patients who have deteriorated, remained unchanged, made a reliable improvement and made a clinically significant improvement. Finally, we sought to identify demographic (age, education), clinical (motivation, symptom load, comorbid disorders, previous inpatient, and outpatient treatments), and treatment-specific variables (anti-depressive medication, treatment duration) that distinguish significantly between bulimic patients who deteriorate/remain unchanged, who show a reliable improvement and those who change to a clinically significant extent.

Participants were included in the study if they met the following criteria: (i) inpatient treatment at a psychosomatic hospital in Germany between January 2009 and October 2014; (ii) ICD-10 diagnosis of BN; (iii) admission body mass index (BMI) of at least 17.5; (iv) at least 18 years of age; and (v) female gender. As illustrated in the flow chart (Figure 1), an initial sample of N = 481 met these criteria. One hundred and 86 patients were excluded due to missing EDI-2 data at intake. Thus, our final sample comprised N = 295 individuals (see Table I for sample details). Patients with and without EDI-2 data at intake did not differ significantly in terms of various demographic (age, education, relationship), clinical (type and duration of BN, BMI at intake, comorbidities, severity of binge eating and compensatory behavior at intake, overall and depressive symptom distress at intake, previous treatments), or treatment-specific (motivation at intake, duration) variables. Out of the final sample of N = 295 bulimic patients, 85.8% were discharged regularly and 14.2% prematurely (see Figure 1). Patients were discharged "regularly" if they met treatment targets; they were discharged "prematurely" for a variety of reasons (for further details, see Figure 1). Those who were discharged regularly and those who were discharged prematurely did not differ in terms of any demographic, clinical, or treatment-specific variable except for treatment duration and motivation (see Table I). Specifically, those who were discharged regularly were treated significantly longer and were significantly more motivated than those who were discharged prematurely. However, treatment duration and motivation were not correlated with each other (r = .06, p = .24). About 89.3% of the final sample had been treated at least once prior to the present treatment, 85.7% as outpatients, and 51.9% as inpatients. Diagnoses were given by the treating therapist who was either an experienced clinician (psychologist or psychiatrist) or a therapist in training under the supervision of an experienced clinician. Moreover, all therapists were specialized in eating disorders. All patients (including those who were discharged prematurely) were administered the Structured Interview for Anorexic and Bulimic Syndromes-Self-Rating (SIAB-S), the Eating Disorder Inventory-2 (EDI-2), the Brief Symptom Inventory (BSI), and the Beck Depression Inventory-II (BDI-II) at treatment intake and discharge. Written informed consent was obtained from all participants prior to the assessments.

Graph: Figure 1. Flow chart.

Table I. Socio-demographic, clinical and treatment-specific characteristics of subsamples of patients who were discharged regularly and prematurely, and of the overall sample (N = 295).

Notes. BN, bulimia nervosa; SIAB-S, Structured Interview for Anorexic and Bulimic Syndromes-Self-Rating; EDI-2, Eating Disorder Inventory-2; BSI, Brief Symptom Inventory; GSI, Global Severity Index; BDI-II, Beck Depression Inventory-II; N/n, frequencies; M, mean; SD, standard deviation; Test statistic, χ2 for categorical variables and t for dimensional variables; p, level of significance. In order to correct for alpha error accumulation due to multiple testing (24 tests), a Bonferroni-corrected alpha level was used (p < .002, two-tailed).

The patients received an intense, multimodal inpatient treatment program comprising both individual and group psychotherapy. Individual therapy took place once a week for those on statutory health insurance and twice a week for those on private insurance. Each session lasted 50 minutes. Group treatments included art therapy, sports therapy, social cooking, social skills training (e.g., practices to express feelings and wishes, to show assertive behavior and to set limits), and a manualized, disorder-specific group. Further on, patients received mealtime support. Despite social cooking and mealtime support, patients themselves were responsible for their eating and compensatory behavior. They had free access to a cafeteria, a supermarket, and lavatories. Individual therapy and the disorder-specific group were both based on the cognitive-behavioral model and included treatment elements described in published cognitive-behavioral treatment programs for patients with eating disorders (see, e.g., Fairburn, [24]; Jacobi, Thiel, & Beintner, [43]; Legenbauer & Vocks, [57]). The individual therapy was not manualized but consisted of the following elements in general: psycho-education on eating and eating disorders, individualized case formulation, fostering changes in diet, body exposure and fostering body acceptance, restructuring of cognitive biases, and relapse prevention. The eating disorder group which was designed internally by clinic employees was manualized and included modules as described in Table II. Contents of group therapy sessions were debriefed in individual therapy. Non-specialized inpatient treatment programs usually do not include social cooking, mealtime support, body exposure, and eating-disorder-specific groups. Cognitive-behavioral interventions were complemented with further strategies such as mindfulness- and acceptance-based interventions if indicated. For example, patients with comorbid borderline personality disorders were usually registered to take part in a skills group as well. Treatments were carried out by clinical psychologists and/or psychiatrists who were all trained in CBT and supervised by experienced therapists. The majority of patients (64.1%) were taking at least one antidepressant during treatment. SSRIs were prescribed most often (48.5%), followed by selective serotonin noradrenaline reuptake inhibitors (SNRIs; 8.5%) and tricyclic antidepressants (7.8%). In 27.6% of patients the antidepressant medication which they were taking at treatment intake was not changed during treatment, but in 36.5% of patients it was changed (continued, discontinued or adjusted). Mean treatment duration was 75.34 days (SD = 36.88).

Table II. Contents of eating disorder group therapy.

The SIAB-S (Fichter & Quadflieg, [26]) is designed to comprehensively assess symptoms of eating disorders and related symptoms. It comprises 87 items belonging to seven subscales. Items are to be rated on a 5-point Likert scale (0 = no, 4 = very strongly). The SIAB-S shows at least acceptable internal consistencies (Fichter & Quadflieg, [29]), sensitivity and specificity (Fichter & Quadflieg, [28]), as well as good convergent, discriminant, and construct validity (Fichter & Quadflieg, [29]). The self-report version is largely equivalent to the expert rating, with similar psychometric properties and good kappa values when comparing both versions (Fichter & Quadflieg, [28]). In the present study, the SIAB-S was used to calculate outcomes measuring the severity of binge eating and compensatory behavior in accordance to the recommendations of Fichter and Quadflieg ([27]). Specifically, severity of binge eating was calculated using the highest score on any of the three items "episodes of binge eating," "average frequency of a binge eating episode," and "loss of control with regard to binge eating." Severity of compensatory behavior was calculated as the highest score on any of the six items "vomiting," "laxative abuse," "abuse of diuretics," "abuse of appetite suppressants," "fasting," and "excessive physical exercise."

The EDI-2 (Garner, [33]) is a multidimensional self-report measure that was used to assess overall eating disorder pathology. It consists of 11 subscales comprising 91 items in total that are rated on a 6-point Likert scale (1 = never, 6 = always). The German version (Paul & Thiel, [71]) has shown satisfactory internal consistency for all subscales (Paul & Thiel, [71]), good test–retest reliability (Thiel & Paul, [89]) and criterion validity (Paul & Thiel, [71]) as well as adequate discriminant and convergent validity (Salbach-Andrae et al., [79]).

The Brief Symptom Inventory (BSI; Derogatis, [20]; Derogatis & Melisaratos, [21]) is a 53-item questionnaire that is designed to assess nine dimensions of psychological symptoms within the past week. Answers are rated on a 5-point Likert scale (0 = not at all, 4 = extremely). In a clinical sample, the German version (Franke, [32]) has exhibited excellent internal consistency for the Global Index of Severity (GSI), good convergent and construct validity as well as adequate sensitivity to change (Geisheim et al., [34]). Test–retest reliability has been found to range between acceptable and excellent depending on the subscale (Franke, [32]).

The BDI-II (Beck, Steer, & Brown, [7]) is a widely used questionnaire to measure self-reported depressive symptomatology within the past seven days. For each of the 21 items, participants choose one statement with an assigned value between 0 and 3. These values can be summed up to a total score (0–63), with higher scores indicating greater depression severity. The German translation (Hautzinger, Keller, & Kühner, [40]) exhibits satisfactory internal consistency, acceptable test–retest reliability, good convergent and discriminant validity, as well as sensitivity to change (Kühner, Bürger, Keller, & Hautzinger, [54]). The factorial structure from the original was replicated in the German version (Keller, Hautzinger, & Kühner, [50]).

Socio-demographic, clinical, and treatment-specific characteristics were retrieved from patients' records. Treatment motivation at admission was rated by the treating therapist on a Likert scale ranging from 0 (not motivated at all) to 4 (highly motivated).

Our analyses were based on a sample of patients including both those who were discharged regularly and prematurely. For percentages of missing data in this sample, see Figure 1. Statistical significance of treatment outcome was analyzed with a multilevel analysis (Raudenbush & Bryk, [74]) with repeated measures of symptom severity (i.e., pre-treatment and post-treatment scores; level 1) nested within participants (level 2) for each outcome variable. Measuring point "0" was centered to the mean of pre- and post-treatment scores. Thus, the intercept now represents the mean pooled over pre- and post-treatment scores. To control for medication effects, we included the variable antidepressant medication (no antidepressant medication, changing medication, or stable medication) in our analyses. Specifically, participants without any antidepressant medication were compared to (a) those with a changing medication and (b) those with a stable medication. We used the software R (R Development Team, [73]) with packages lme4 (Bates et al., [6]) and lmerTest (Kuznetsova, Brockhoff, & Christensen, [55]) for these analyses. Due to multiple testing (5 tests), a Bonferroni-corrected alpha was used (p < .01, two-tailed). In order to quantify the effects of treatment on all outcome measures, the effect size partial eta squared was computed as part of repeated measures ANOVAs with the within-subjects factor of time and the between-subjects factor of antidepressant medication for each outcome. In accordance with Cohen's ([17]) convention, partial η2 ≥ 0.01 was regarded as small, ≥0.06 as medium, and ≥0.14 as large effect.

As indicated previously, clinical significance of treatment outcome was determined using the criteria proposed by Jacobson and Truax ([45]). To classify patients in the outcome groups, an index was calculated to determine whether a symptom change is reliable or only due to measurement errors or chance, as well as a cut-off point that distinguishes between a dysfunctional and a functional norm population. The index that was used to measure whether symptom change was reliable is the reliable change index (RCI; Jacobson, Follette, & Revenstorf, [44]). It is calculated with the following formula:

Graph

X pre represents the patient's score at intake and Xpost the patient's score at discharge (each on the respective measure). SE is the standard error of the clinical sample at intake and rxx represents the reliability (Cronbach's alpha) of the corresponding measure (see the appendix). As recommended by Lambert and Ogles ([56]), we used the internal consistency and not the test–retest reliability. Test–retest reliabilities are supposed to be deflated by real individual differences in patients' symptom changes and by erratic behavior patterns of this population. Non-patient samples were not used to reduce test–retest deflation as reliability of scales varies across populations (for more details, see Martinovich, Saunders, & Howard, [61] and Tingey, Lambert, Burlingame, & Hansen, [92]). Still, it has to be acknowledged that Cronbach's alpha leads to higher estimations of reliable change than using the test–retest reliability. An RCI above the 95% confidence limits ±1.96 constitutes evidence for a reliable change (p ≤ .05).

A criterion C was calculated to distinguish between the dysfunctional and functional population using the following formula:

Graph

M 0 is the mean of the normative sample and SD0 the standard deviation thereof. Norms for the healthy population were obtained from the manuals of the outcome measures applied. M1 is the mean of the clinical sample at intake and SD1 the standard deviation thereof (see the appendix).

Patients were then classified into their respective outcome group (which were all based on the EDI-2 global score) if they met the following criteria:

• Normative patients: patients with scores within the norm both at intake and discharge
• Clinically significant improvement: patients improved reliably (RCI ≥ 1.96) and symptoms moved into the range of the functional population
• Reliable improvement: patients improved reliably but symptoms remained within the range of the dysfunctional population
• No change: patients did not change reliably in either direction
• Reliable deterioration: patients worsened reliably (RCI ≤ 1.96).

In the present study, we only included patients from Group 2 to 5 in the analyses and excluded patients with normative scores at intake and discharge (Group 1).

Finally, to identify potential differences between patients in these outcome groups in terms of various socio-demographic, clinical, and treatment-specific variables, we computed χ2-tests for categorical variables and univariate ANOVAs for continuous variables. Given that the number of deteriorated patients on the global EDI-2 score was very low with less than 5%, we built a single group with both patients who deteriorated and those who remained unchanged. We set alpha at p < .002 (two-tailed) as we had conducted 25 tests. SPSS 23 was used for all analyses except for the multilevel analyses.

Means and standard deviations of eating disorder, general, and depressive symptoms in the three different medication groups at treatment intake and discharge are presented in Table III. As can be seen in Table IV, the multilevel analyses revealed significant time effects for all outcome measures. These findings indicate significant reductions in terms of severity of binge eating and compensatory behavior, overall eating disorder symptoms, general psychopathology, and depressive symptom severity during inpatient treatment while controlling for antidepressant medication. Further on, findings showed that patients with both a changing and a stable medication suffered from significantly more mean eating disorder, general, and depressive symptomatology than those without a medication. Those medication effects were not found for the mean intensity of binging and compensating behavior except for a significantly higher mean binging intensity in those who were taking a stable medication than in those without a medication. Additionally, one single significant time × medication effect showed that those patients who received a stable medication experienced fewer reductions in binging behavior than those without a medication. These findings are consistent with those from the repeated measures ANOVAs. Effect sizes (partial η2) for time and medication effects were large on all outcome measures (time effects: SIAB-binge = 0.81, SIAB-compensate = 0.56; EDI-2 = 0.62, BSI = 0.49, BDI-II = 0.70; medication effects: EDI-2 = 0.13; BSI = 0.13; BDI-II = 0.17) except for medication effect sizes of binging (SIAB-binge = 0.02) and compensating behavior (SIAB-compensate = 0.01) which were small. Effect sizes for interaction effects were at most small in the different outcome measures (all partial η2 < 0.04). All findings remain unchanged when controlling for baseline symptom severity in univariate analyses of covariance (ANCOVAs) with symptom change scores as dependent variable and medication as group factor. The only different result is that patients without a medication show greater reductions in general psychopathology than those with a stable medication when controlling for general psychopathology at intake.

Table III. Means and standard deviations of outcome measures in three different medication groups (N = 295).

Note. AD, Antidepressant Medication; binging, binging behavior; compensating, compensating behavior; SIAB-S, Structured Interview for Anorexic and Bulimic Syndromes-Self-Rating; EDI-2, Eating Disorder Inventory-2; BSI, Brief Symptom Inventory; GSI, Global Severity Index; BDI-II, Beck Depression Inventory-II; M, Mean; SD, standard deviation.

Table IV. Statistical significance of symptoms changes (N = 295).

Note. Binging, binging behavior; compensating, compensating behavior; SIAB-S, Structured Interview for Anorexic and Bulimic Syndromes-Self-Rating; EDI-2, Eating Disorder Inventory-2; BSI, Brief Symptom Inventory; GSI, Global Severity Index; BDI-II, Beck Depression Inventory-II; ß, unstandardized regression coefficient; SE, standard error; df, degrees of freedom. In order to correct for alpha error accumulation due to multiple testing (five tests), a Bonferroni-corrected alpha level was used (p < .01, two-tailed). **p < .01. ***p < .001.

Results on the clinical significance of symptom changes during treatment can be seen in Figure 2. The majority of patients showed reliable (EDI-2: 33.7%, BSI: 34.8%, BDI-II: 18.1%) or even clinically significant symptom changes (EDI-2: 43.2%, BSI: 33.9%, BDI-II: 56.9%). Approximately one quarter of patients experienced no symptom changes (EDI-2: 21.1%, BSI: 26.9%, BDI-II: 24.6%) and less than 5% deteriorated (EDI-2: 2.0%, BSI: 4.4%, BDI-II: 0.4%).

Graph: Figure 2. Clinical significance of symptoms changes as measured with the Eating Disorder Inventory-2 (EDI-2), the Global Severity Index of the Brief Symptom Inventory (BSI-GSI) and the Beck Depression Inventory-II (BDI-II).

Univariate ANOVAs and χ2-tests revealed significant overall differences between the three outcome groups in terms of symptom load at intake (EDI-2, BDI-II, and BSI), and comorbid personality disorders (see Table V). These findings indicate that patients who improve to a clinically significant extent, who change reliably, and who remain unchanged/deteriorate in terms of the EDI-2 global score differ significantly with respect to their symptom load at intake, and the presence or absence of a comorbid personality disorder. Bonferroni-corrected post-hoc tests showed that patients who changed to a clinically significant extent were suffering from less overall (BSI) and depressive (BDI-II) symptomatology at intake, and they were less likely to suffer from a comorbid personality disorder than patients from the other outcome groups. Moreover, post-hoc tests showed that those who improved reliably had a lower eating disorder symptom load at intake (EDI-2) than those who deteriorated/remained unchanged and even those who improved to a clinically significant extent.

Table V. Differences between outcome groups (EDI-2 Global Score) (N = 295).

Notes. BN, bulimia nervosa; SIAB-S, Structured Interview for Anorexic and Bulimic Syndromes-Self-Rating; EDI-2, Eating Disorder Inventory-2; BSI, Brief Symptom Inventory; GSI, Global Severity Index; BDI-II, Beck Depression Inventory-II; OCD, Obsessive-compulsive disorder; M, mean; SD, standard deviation; N/n, frequencies; G1, Group 1; G2, Group 2; G3, Group 3; Test statistic, χ2 for categorical variables and F for dimensional variables; p, level of significance. In order to correct for alpha error accumulation due to multiple testing (25 tests), a Bonferroni-corrected alpha level was used (p < .002, two-tailed).

Given that our findings indicated differences between outcome groups regarding symptom severity at intake and comorbid personality disorders, we examined in post-hoc analyses whether the comorbid personality disorders would moderate the differences in symptom severity at intake among outcome groups. However, results did not support this assumption (EDI-2: F(2,193) = 1.80, p = .17, partial η2 = 0.02; BSI: F(2,188)  = 0.41, p = .67, partial η2 = 0.004; BDI-II: F(2,186) = 2.43, p = .09, partial η2 = 0.03). This indicates that both baseline symptom severity and comorbid personality disorders contribute independently to treatment outcome.

Moreover, descriptive statistics revealed that borderline personality disorder was diagnosed in 24.4% of cases (n = 72); other personality disorders were diagnosed in less than 3% of the present sample. Thus, we tested post-hoc whether patients from the three outcome groups differ significantly in terms of the presence or absence of a borderline personality disorder. A χ2-test to test for differences in the distribution of borderline personality disorders across the three outcome groups revealed a significant overall effect (χ2(2) = 12.19, p = .002). Post-hoc tests indicated that patients who improved to a clinically significant extent were less likely to be suffering from a comorbid borderline personality disorder than patients who deteriorated/remained unchanged or improved reliably.

The present study showed that intensive inpatient treatment is highly effective in a sample of multimorbid patients with severe BN. Interestingly, our data also show that in such intensive treatment programs, antidepressant medication does not seem to enhance symptom decreases during treatment. As such, our findings provide strong support for an intense psychotherapeutic focus in inpatient treatments for individuals with BN. Furthermore, findings on clinical significance indicate that about 30% of patients responded reliably and about 45% to a clinically significant extent in terms of overall eating disorder, general and depressive symptom load. Finally, our findings demonstrate that patients who improved to a clinically significant extent were less distressed at intake and less likely to suffer from a comorbid borderline personality disorder than those who did not.

Our findings on the statistical significance of symptom changes add to the large body of evidence in support of the efficacy of CBT for BN in both outpatient and inpatient settings (e.g., Fichter & Quadflieg, [30]; Lewandowski et al., [58]; Poulsen et al., [72]; Shapiro et al., [82]; Wilson & Zandberg, [95]). However, to the knowledge of the authors, the present sample is the largest of routinely treated bulimic inpatients studied thus far. The effect sizes identified in our study are higher than effect sizes (partial eta squared as well) reported for CBT in an outpatient setting (Ghaderi, [35]), despite the fact that patients in the present study were more impaired and treated for shorter periods. These findings show that an intensive inpatient treatment is effective for severely disturbed patients with BN. However, in this context it should also be acknowledged that findings from reviews and meta-analyses of controlled trials indicate that no-treatment and placebo-groups yield at least small effects (in bulimic individuals) as well (Hartmann, Herzog, & Drinkmann, [39]; Nakash-Eisikovits, Dierberger, & Westen, [65]; Wampold, Minami, Tierney, Baskin, & Bhati, [93]). Moreover, results from these reviews and meta-analyses suggest that mean treatment effects are high in general (even if not as high as in our study) and roughly comparable across psychotherapeutic approaches (Hartmann et al., [39]; Wampold et al., [94]). Thus, our findings might also be interpreted as another evidence for the assumption that all bona-fide psychotherapies are roughly comparable effective due to common factors of psychotherapy such as the therapeutic alliance (Baldwin, Wampold, & Imel, [5]; Hartmann et al., [39]). They might also be interpreted as an evidence for the hypothesis that all treatments are comparable effective on a mean level but may differ on an individual level and should thus be tailored to the patients' individual needs (Hartmann et al., [39]; Norcross & Wampold, [67]).

Our findings concerning the clinical significance of treatment effects are the first that demonstrate in an inpatient sample that approximately 30% of patients improve reliably and about 45% to a clinically significant extent under the criteria of Jacobson and Truax ([45]). Rates found in our study are higher than those identified for bulimic outpatients (Openshaw, Waller, & Sperlinger, [69]), day hospital patients with eating disorders (Ben-Porath, Wisniewski, & Warren, [8]), and anorectic inpatients (Schlegl, Quadflieg, Löwe, Cuntz, & Voderholzer, [80]) in prior studies; however, rates for reliable changes found in our study were roughly comparable to those identified in the aforementioned previous studies (Ben-Porath et al., [8]; Openshaw et al., [69]; Schlegl et al., [80]). Overall, these findings are consistent with findings from prior research indicating that anorectic and day hospital patients experience worse treatment outcome than bulimic patients in general, and inpatients in particular (Byrne, Fursland, Allen, & Watson, [12]; Herzog et al., [41]; Schnicker et al., [81]; Zeeck et al., [97]), probably because anorectic patients are more disturbed and day hospital treatments are less intense.

Our findings which show that patients who improve to a clinically significant extent experience a lower symptom load at intake and are less likely to suffer from comorbid borderline personality disorders give weight to prior studies with smaller samples that have identified both variables as moderators or predictors of treatment outcome in BN (Baell & Wertheim, [4]; Castellini et al., [13]; Fahy & Russell, [23]; Johnson, Tobin, & Dennis, [46]; Rossiter, Agras, Telch, & Schneider, [77]; for contradictory findings concerning borderline personality disorder as moderator/predictor, see, e.g., Ames-Frankel et al., [3] or Steiger & Stotland, [84]). The finding that no other comorbid condition had an impact on treatment outcome is consistent with previous studies showing that comorbid Axis-I disorders and outcome are not related (Bulik, Sullivan, Joyce, Carter, & McIntosh, [11]; Fichter & Quadflieg, [25]; Keel & Mitchell, [49]; Reas, Williamson, Martin, & Zucker, [75]; for contradictory findings, see Herzog et al., [41]; Thompson-Brenner & Westen, [91]). Moreover, it shows that borderline personality disorder is not just an indicator of general psychiatric comorbidity.

Our findings have important theoretical and clinical implications. From a theoretical perspective, the results suggest that supplementing group statistics with analyses of outcome on an individual level offers a more detailed and meaningful picture of treatment outcome as the exact number of responding and non-responding patients can be determined. These findings may be more informative than those of statistical significance testing, especially for non-research clinicians. From a clinical perspective, our findings indicate that patients who suffer from a severe form of BN and who are multimorbid can be effectively treated as part of an intensive inpatient treatment program that has a strong psychotherapeutic focus and is specifically tailored to the needs of patients with eating disorders. In such an intense treatment, an antidepressant medication does not seem to impact outcome as much as usually expected. Our findings also indicate that some of the patients who do not respond to such an inpatient program suffer from comorbid borderline personality disorders. Given that changes in interpersonal or emotion regulation difficulties have repeatedly been suggested as potential working mechanisms in treatment for patients with borderline personality disorder (Clarkin & Levy, [16]; Lynch, Chapman, Rosenthal, Kuo, & Linehan, [60]; Rossiter et al., [77]), current treatments for patients with both BN and a borderline personality disorder might be improved through an even stronger focus on minimizing interpersonal and emotion regulation difficulties. Dialectical-behavior therapy directly addresses these issues. As it has already been proven to be effective in bulimic patients with borderline personality disorder (Chen, Matthews, Allen, Kuo, & Linehan, [14]; Palmer et al., [70]; Safer, Telch, & Agras, [78]), complementing traditional CBT with a full dialectical-behavior therapy program might thus have the potential to enhance treatment outcome for this non-responding sub-group (Coker, Vize, Wade, & Cooper, [18]; Linehan, [59]). Moreover it might be helpful to add trainings to treatment that directly focus on systematically enhancing emotion regulation skills such as the affect regulation training (Berking, [9]).

There are several limitations in this study. Major limitations include the exclusive use of self-report measures, the inability to calculate the clinical significance of symptom changes in binge and compensatory episodes due to missing norms and the uncontrolled, non-randomized design of the study. The naturalistic setting of our study enhances the clinical relevance of our findings and allows investigation of treatment outcome under real-world conditions which enhances the external validity of the study, especially in the context of the comparatively large sample, but to the detriment of internal validity. Given the multimodal treatment concept of our study including pharmacological, psychological and milieu-based interventions, we cannot specify which particular technique or therapeutic process helped patients (mostly). Another limitation concerns the question of generalizability of the present study's findings. Given that other countries than Germany often have less coverage from insurance companies for treating bulimic patients in inpatient settings, this kind of program is not very common in other countries than Germany. Moreover, data were collected from one single psychosomatic hospital. Thus, findings might be due to an unspecific positive hospital effect not generable to other hospitals and not to the specific treatment concept. Further on, the formulas of the method of clinical significance assume normal distribution of the variables included and we do not know whether clinical significance testing is robust against violations of this assumption (Jacobson & Truax, [45]). Since not all of our variables included in the analyses had been normally distributed, the predictive accuracy of our findings might be reduced and generalizations should be drawn with caution (see Tingey et al., [92]). Finally, it is noteworthy that the clinical significance method also has disadvantages which include the variability of the C criterion depending on symptom severity in the sample at baseline as well as regression to the mean (Hsu, [42]; Ogles, Lunnen, & Bonesteel, [68]). Moreover, it has been criticized that aggregating individual reliable changes across individuals to obtain an overall group percentage leads to a too conservative estimation of treatment outcome (Hageman & Arrindell, [37], [38]).

Due to these shortcomings, future research should utilize further measures of treatment outcome. Moreover, given the multimodal treatment concept of the present study, future studies should assess and examine specific effects of single treatment elements on outcome in dismantling studies as well as therapeutic processes in mediational models that might be responsible for outcome in inpatient treatments for bulimic individuals. Furthermore, future research should include follow-up assessments to examine the longitudinal stability of symptom changes and to identify relevant predictors of clinically significant and reliable changes in the long-term. Additionally, future research should clarify whether CBT combined with a full dialectical-behavior therapy program or with the affect regulation training increases treatment efficacy in bulimic patients. Finally, additional moderator and mediator analyses should be conducted to identify what type of processes lead to treatment response, non-response or even deteriorations and in what type of bulimic inpatients. Specifically, it might be highly relevant for clinical practice to examine the factors and processes that prevent treatment success in bulimic inpatients with comorbid borderline personality disorders and severe symptom load at intake as these factors may help further improve inpatient treatment for BN.

We thank Tatjana Michel and Stella Berboth for their assistance with data management and Philipp Sckopke for his statistical support.

Pre-treatment Scores of Patients and Norm Samples, Cronbach's Alpha, Criterion C and Percentages of Patients in the Normative Range for Each Outcome Measure.

Notes. EDI-2, Eating Disorder Inventory-2; BSI, Brief Symptom Inventor; GSI, Global Severity Index; BDI-II, Beck Depression Inventory-II; N, frequencies; M, Mean; SD, standard deviation. Data for the norm samples were taken from the manuals of the respective questionnaire.