Introduction

Eating disorders (ED) are often associated with other mental disorders, most frequently depression [5]. In their review, O’Brian and Vincent [5] cite studies that found rates of about 50 % for a comorbid depression in ED patients.

Depression and anxiety in anorexia nervosa (AN) patients are often attributed to the effects of reduced caloric intake. However, studies that address the link between depression and body weight in anorectic patients are contradictory. Kawai et al. [4] did not find an association between BMI and depression in a sample of 24 patients, whereas, e.g., Pollice et al. [6] had investigated 48 AN patients and had found that depressive symptoms were more intense in patients with lower body weight. Calugi et al. [3] found that the presence of a comorbid major depression did not predict treatment outcome in AN. Their sample consisted of 63 patients. Methodological issues, such as different sample compositions and a wide range of different assessment methods complicate the interpretation of findings. In this study, we examined the relation between weight status and depression levels in a very large sample of AN inpatients. Given the existing contradictory evidence, our first research question was whether the degree to which patients were underweight and level of depression were associated. Secondly, we examined whether the level of depression before treatment as well as changes in depressive symptoms, among other possible variables, predicted treatment outcome in AN.

Method

Participants

418 female patients who sought treatment at the Schoen Clinic Roseneck between January 2006 and December 2008 participated in this study. Subjects with missing data on BMI or BDI-I at admission or discharge were excluded from further analysis. Figure 1 depicts the flow chart of the patient sample.

Fig. 1
figure 1

Flow chart of patient sample

Full size image

There was no difference in BMI-I at admission (t = 1.095, p = 0.290) and discharge (t = 0.889, p = 0.374) between patients that were excluded because of missing BDI-I data and patients included in this study.

The responsible ethics committee approved the study and all participants gave informed consent.

Treatment

The Schoen Clinic Roseneck is specialized in the treatment of eating disorders. It provides intensive inpatient treatment for adults with anxiety, mood and eating disorders. Patients with AN come from all over Germany, but in general from the proximal region. They are mostly referred by their general practitioner or a medical specialist.

The treatment procedure is based on cognitive-behavioral therapy principles but new evidence-based treatments are integrated in the therapy protocol. The specific needs of each patient are met by individualized treatment. Nevertheless, it is mandatory for patients with AN to attend a disorder-specific group therapy, general group therapy, individual therapy, social skills training, art therapy, the teaching kitchen and exercise therapy. The eating disorder-specific group therapy consists of nine therapy sessions with 100 min each. The main elements of the program are psychoeducation, behavior and functional analysis, acceptance of their own body, dealing with emotions and needs and relapse prevention. The general group psychotherapy takes place up to three times a week and lasts 90–100 min. There are approximately 12 women in one group and topics not related to eating are discussed. Patients receive individual therapy once or twice for 1 h each week. During course of treatment patients participate in seven sessions at the teaching kitchen. There they learn facts about healthy nutrition, especially about adequate amounts of food. The women also have cooking sessions together. The social skills training takes place once a week for 100 min. The patients learn mostly by role-playing how to be self-confident and assertive in close relationships. Patients are required to gain 700 g each week. Co-therapists weigh the patients twice a week in the morning. Weight gain is visualized on charts. If anorectic patients do not gain weight, consecutive steps are taken: increase of food intake and monitoring during meal times, administration of high caloric fluids and feeding through a nasal tube.

Measures

The German version of the Beck Depression Inventory (BDI-I) was applied at admission and discharge. The Beck Depression Inventory is a self-report questionnaire widely used to assess the severity of depression symptoms. Each of the 21 items maps on to a particular symptom included in the DSM-III diagnostic criteria. Symptoms are rated on a 4-point scale ranging from 0 to 3. Anchors for each value are provided (e.g., 0 “I do not feel sad.”). The total score ranges between 0 and 63 with higher scores indicating more severe symptoms of depression. Cronbach’s Alpha within the presented sample was 0.88 at admission and 0.91 at discharge. The BDI-I has proven to have good-to-excellent validity and reliability as well as sensitivity to change [1].

Body mass index Weight is assessed by co-therapist twice weekly in the clinic, height at admission. The BMI was calculated using the following formula: weight in kilograms/(height in meters)2. The BMI is a widely used and currently most accepted method to classify medical risk according to weight status. BMIs at admission and discharge are reported.

Sociodemographic (age, marital status) and treatment information (length of treatment, medication, completer yes/no) were obtained from standard clinical records and entered into an electronic data base for research and evaluation. Patients were considered to have completed the program if they achieved the target weight (BMI > 18) or, in case this was not achieved, if follow-up treatment was assured (e.g., re-admission planned). Non-completers were patients who had been discharged due to lack of therapy motivation or for disciplinary reasons.

Comorbid disorders (mood disorders, anxiety disorders, personality disorders) were diagnosed by trained psychologists under supervision according to ICD-10 criteria. Existence of an emotionally unstable personality disorder was considered as a potential predictor for treatment outcome, as it has previously shown to predict poor outcome in eating disorders (e.g., [7]).

Data analysis

Differences in continuous variables between admission and discharge were tested for significance with paired-sample t tests. For multiple t tests (differences in the three AN subgroups), we used Bonferroni correction to avoid Alpha-error inflation, setting the significance level at p = 0.006 (eight t tests). To determine effect sizes (ES), differences between means at discharge and admission were divided by standard deviations at admission.

An analysis of variance was conducted to compare group differences (independent variable: subtypes of AN) in BMI and BDI-changes. Treatment outcome was operationalized as the difference between BMI values at admission and discharge.

Because the data were non-normally distributed, associations between variables were measured using Spearman’s Rho (rank order) correlations.

A multivariate linear regression analysis was carried out with change in BMI as the dependent variable (criterion). Demographic, anamnestic and treatment variables served as independent variables. Regression analyses were performed with stepwise, forward, backward and enter methods to ensure the soundness of the model. The variables that entered into the analyses are depicted in Table 1. To compare the magnitude of the regression coefficients and to evaluate the relative importance of all explanatory variables, standardized regression coefficients are presented. R 2 and adjusted R 2 are reported to estimate the explained variance.

Table 1 Results of regression analyses for BMI changes (dependent variable)
Full size table

The requirements for regression analysis were checked: scatter plots were inspected for abnormal distribution or extreme values to confirm linearity between continuous variables. Scatter plots of estimated values with standardized residuals and histograms for normal distribution were used to ensure homoscedasticity. To control for collinearity, the variance inflation factor (VIF), tolerance and condition index (CI) were calculated. A VIF above 10 or tolerance under 0.10 is critical. A CI above 15 indicates moderate multi-collinearity, a CI greater than 30 signifies severe multi-collinearity [2].

All statistical analyses were performed with SPSS 23.

Results

Study sample

The average age at admission was 26.42 years (SD = 8.25) ranging from 18 to 62. Regarding the education level, 33 patients (7.9 %) were still in school, 2 (0.5 %) had no graduation, 32 (7.7 %) hold a secondary general school certificate, 123 (29.4 %) an intermediate school certificate, and 227 (54.3 %) higher education entrance qualification.

The majority of patients (N = 289; 69.1 %) were classified as the restricting type. 100 (23.9 %) met criteria for the purging subtype and 29 (6.9 %) for the atypical subtype. The average duration of the ED was 9.21 years (SD = 6.98).

At least one comorbid disorder was diagnosed in 363 patients (86.8 %), with the majority having more than one. The highest prevalence rates were found for mood disorders: 36.4 % (152) of the patients suffered from a depressive episode and 39.5 % (165) from a recurrent depressive disorder. 7.1 % (29) suffered from comorbid obsessive–compulsive disorder and 2.6 % (11) from other anxiety disorders. 12.9 % (54) were diagnosed with a personality disorder, of which 10.5 % (44) with emotionally unstable personality disorders. All other diagnosed personality disorders were found in less than five persons and are not reported here.

Medication

108 of the patients (25.8 %) were taking antidepressants, 58 (13.9 %) as long-term medication, 7 (1.4 %) intercurrently.

Treatment outcome

BMI increased significantly from 14.65 (SD = 1.73; range 10.04–17.47) to 17.27 (SD = 1.51; range 12.98–24.24; t = −37.196; df = 417; p < 0.001; ES = 1.51). Significant weight gains were observed in all diagnosis subgroups (restricting: t = −30.035; df = 288; p < 0.001; ES = 1.44; purging: t = −18.709; df = 99; p < 0.001; ES = 1.68; atypical: t = −12.012; df = 28; p < 0.001; ES = 2.11).

The mean BDI-I at admission was 26.07 (SD = 10.46; range 1–54) and 13.63 (SD = 10.60; range 0–52) at discharge. The t test revealed a significant improvement (t = 26.170; df = 417; p < 0.001) with an effect size of 1.19. All diagnosis subgroups showed substantial improvement with an effect size of 1.28 for the restricting type (t = 22.640; df = 288; p < 0.001), 1.13 for the purging type (t = 12.561; df = 99; p < 0.001) and 0.79 for the atypical subtype (t = 4.971; df = 28; p < 0.001).

A multivariate analysis of variance was conducted to find out whether there were differences in BDI and BMI changes between the groups. No differences in BDI and BMI results were found (BDI t = 0.226, p = 0.798, BMI t = 0.650, p = 0.523).

Correlations between BMI and BDI

There was no substantial correlation between BMI and BDI-I at admission (r = −0.003; p = 0.959; see Fig. 2) or at discharge (r = −0.023; p = 0.634). Considering each diagnosis subtype separately these results remained stable.

Fig. 2
figure 2

Association between BMI and BDI-I at admission

Full size image

Multivariate linear regression

Predictors (Table 1) were entered into the regression analysis with change in BMI as the criterion. Standardized beta-coefficients for all variables are reported in Table 1 (Enter method). The explained variance as measured by R 2 was 0.43 (R 2adjusted  = 0.42).

Regression analyses were conducted using backward, forward and stepwise procedures. All methods yielded the same set of significant predictors: BMI at admission, decrease in BDI-I, duration of treatment, being a completer and change of antidepressant status during treatment. Restrictive subtype and BMI at admission were negative predictors.

Inspection of scatter plots as well as normal distribution of residuals confirmed the linearity of relationships. Normal distribution could not be confirmed for duration of treatment, age and age of onset. Thus, homoscedasticity cannot be assumed for these variables. A restriction of variance was detected in almost all continuous variables. Tolerance scores ranged between 0.723 and 0.975, VIF scores varied between 1.025 and 1.383 while the CI was 34.319. This indicates mild-to-severe multi-collinearity.

Discussion

The findings suggest that depressive symptoms and BMI are not associated in anorectic female patients, neither at admission to an intensive inpatient treatment, nor at discharge.

This finding is consistent with results by Kawai et al. [4], indicating that the self-rated level of depression varies independently of patients’ initial or post-treatment BMI. Both AN patients with very low as well as those with higher BMI can have high or low depression level.

Similar to the results of Calugi et al. [3], we found that pre-treatment depression level did not predict outcome. This implies that AN patients benefit from intensive multimodal treatment independent of their initial depression level. In other words, the pre-treatment depression level is neither associated with BMI, nor with treatment success in AN.

A lower BMI at admission was associated with a higher weight gain during treatment. Patients with very low BMI may receive more attention from therapists. Also, stricter therapeutic consequences might be enforced.

Moreover, decrease in depressive symptoms was positively associated with weight gain. This result might hint at an important role of depressive symptoms in the treatment of AN. However, it is also plausible that both variables changed because of a third independent variable, e.g., the intensive multimodal treatment. Multimodal treatment addresses both depressive symptoms and weight thus affecting them in parallel. To assess possible causal relationships between the level of depression and the BMI, a different design would be necessary, e.g., timeline data for both variables that could reveal whether one of these factors changes first.

Duration of treatment was a significant positive predictor of weight gain. Although usually not studied in isolation, treatment duration is known to predict therapy outcome. This association is particularly pronounced in patients with AN, as longer duration of hospitalization is probably linked to a higher weight gain.

It is noteworthy that being a completer was still a significant predictor when duration of treatment was taken into account. Lower motivation and reduced expectations of success might lead to a lower weight gain in non-completers.

Antidepressant medication did not predict improvement in BMI, but patients who had to change their antidepressant medication scheme were less likely to gain weight during inpatient treatment. This can be attributed to the fact that patients whose medication scheme needs to be changed may show more complex symptoms. Since we did not assess dosages and types of antidepressant medication, these results should be interpreted cautiously.

Restrictive AN subtype predicted worse outcome. Patients with this subtype may suffer from more obsessive–compulsive symptoms that complicate treatment.

Emotionally instable personality disorder (PD) did not significantly affect weight gain. Possibly, this PD is not directly linked to BMI. Cognitive and emotional variables like drive for thinness and fear of maturity might be important mediators. Thus, when food intake is monitored continuously, PDs might no longer be associated with the amount of weight gain.

Limitations of this study include that there was no control group, no follow-up measurement, and that the correlational design of this study does not allow conclusions about causal relationships between BMI and depression level. Future studies should include long-term outcome data to draw deeper conclusions on the role of depressive symptoms in AN. Another focus for further research would be longitudinal studies with several assessments, e.g., weekly measurements of both BDI and BMI, to discover the timelines and interconnections of these variables. Finally, this study included only female inpatients, so that it is not necessarily representative of all anorectic patients. Replication studies with different samples (e.g., outpatients, male patients) are warranted.

Strengths of this study are the very large sample size and the longitudinal design.