Introduction

Anorexia nervosa (AN) is a severe mental illness associated with serious medical symptoms, difficulties in various cognitive and socio-emotional tasks, high psychiatric comorbidity rates, and lastly the highest mortality rate among all psychiatric disorders [1–4]. Adult AN patients in need of inpatient treatment have been identified as an particularly vulnerable subgroup [5], for which treatment outcomes have been repetitively shown to be unsatisfactory [6–8]. Despite the general effectiveness of inpatient treatment, naturalistic studies demonstrated that only approximately one-half of the patients successfully restore their weight during therapy (body mass index, BMI > 17.5) and only slightly more than one-third exhibit a clinically significant change in eating disorder pathology upon discharge [6–8]. Also, long-term remission rates have been shown to be lower for the more severely affected AN patients undergoing inpatient treatment [5].

Research during the past decades has mainly focused on the efficacy and effectiveness of AN treatments (e.g. [9–11]). According to the findings, we know that treatments are generally effective and that different psychotherapeutic approaches achieve similar results [10, 12]. In contrast, we still know little about how the treatments work. An understanding of the therapeutic change process is, however, an essential foundation for the improvement of treatment [13]. It enables us to explain why some patients benefit while others do not and also to direct strategies that trigger the critical change processes.

A great number of potentially relevant change process factors in the treatment of AN were proposed [14]. However, the evidence base on these factors is up until now inconsistent or sparse in the field of eating disorders in general and even more so for AN in specific [14]. To our knowledge, the most widely studied change process factor is therapeutic alliance. However, the study results on alliance are highly inconclusive. Regarding the influence of AN symptoms on the therapeutic alliance, one study detected a positive effect of more AN-specific cognitions and behaviours on the later therapeutic alliance [15], whereas another study did not find such an influence [16] and a third study showed that more AN-specific cognitions and behaviours negatively influenced the subsequent therapeutic alliance [17]. Furthermore, the widely assumed pathway from a greater therapeutic alliance to improved outcomes [18] seems to be questionable in the treatment of AN. Instead, the exact opposite direction of the alliance–outcome association (from a greater early symptomatic improvement to a stronger subsequent alliance) was demonstrated in adult cognitive–behavioural outpatient therapies [19]. A similar conclusion might be drawn with respect to motivational change factors. Single studies point to a positive effect of greater motivation on outcome [20] and therapeutic alliance [21], but in general the evidence on the effects of motivational interventions or changes in motivation during treatment is judged to be mixed [22]. Other change process factors in AN, like the level of restraint over eating, depressive symptoms, self-esteem, self-efficacy, and parental self-efficacy, showed no influence as mediators in a study investigating outpatient treatment of adolescent AN [23].

In summary, it is difficult to draw decisive conclusions or to formulate specific hypotheses based on the current evidence base regarding the process of change in AN. The lack of studies might be explained by the various difficulties associated with investigating the change process in AN [24]. One of these difficulties is the selection of variables out of the great number of potentially relevant factors. However, a reduction in the number of factors is usually necessary to perform statistically valid analyses given the typically small sample sizes due to the disorders’ low prevalence and attrition in longitudinal studies [3, 25]. At the same time, it is important to include and control for as many variables as possible. In this context, we decided to use empirically aggregated higher-rank change process factors [26]. These higher-rank factors aggregate a number of potentially relevant change process factors in the treatment of AN. The selection of the single change process factors was based on empirical and theoretical literature and expert opinion. Prominent change process factors like the therapeutic alliance as well as generic theories on psychological change were considered. Specifically, the self-determination theory and the consistency theory are represented. The self-determination theory is a well-researched theory on the motivation of behaviour and its perceived internal or external locus of causality. A key idea of the theory states that a more intrinsic and autonomous regulation make behaviour change and learning more likely [27]. The consistency theory focuses on the relevance of the compatibility of different simultaneous psychological processes within an individual [28]. Both theories share the emphasis on the fulfilment of basic psychological needs for health, well-being and growth.

We aimed to investigate the interrelations between the higher-rank change process factors, the BMI and AN-specific cognitions and behaviours in an exploratory fashion over the course of adult cognitive–behavioural inpatients treatments for AN.

Methods

Design and recruitment

We conducted a naturalistic prospective longitudinal multi-centre study. Three participating hospitals (Schön Clinics: Bad Bramstedt BB; Prien P; Hamburg Eilbek HH) included patients who fulfilled the following inclusion criteria: primary diagnosis of AN, female gender, minimum age of 16 years and the ability to speak German with adequate fluency. Exclusion criteria were acute drug abuse, acute suicidality, existence of a psychotic or bipolar disorder or a severe life-threatening somatic disorder. Diagnoses were validated using the German version of the Structured Clinical Interview for DSM-IV Axis I Disorders [29].

The recruitment took place within a period of 27 months between April 2010 and July 2012. Eligible patients were asked whether they wanted to participate by the responsible study psychologist or physician shortly after admission. Written informed consent was obtained from all participants prior to the inclusion into the study. The study was approved by all respective local research ethics committees.

Inpatient treatment programme

The eating disorder specialist inpatient units were composed of multidisciplinary teams that worked according to a cognitive–behavioural approach. All units belong to the same hospital network and share jointly developed best practice recommendations including clinical pathways for the treatment of AN. The treatment was structured in a symptom-oriented, a problem-oriented and a stabilizing phase. The therapeutic components consisted of nutritional counselling, eating diaries, meal supervision and support, therapy contracts containing a weekly weight gain of 600–1000 g, eating disorder group therapy, individual therapy, psycho-education, emotional competency training, and relaxation, art and music therapy. Importantly, the discharge decision was not based on achieving a clinical cutoff criterion, but rather on the clinical judgement of the responsible therapist and team.

Assessments

To assess various change process factors, we used the Change Process Questionnaire for patients with AN (CPQ-AN) [26]. The questionnaire is a self-report measure, which uses a four-point Likert scale with the response options ranging from “disagree” to “agree”. It was developed with the aim of considering as many as possible AN-related change process factors described in the literature and by experts in the field. To enable an easy utilization for research purposes, the factors were empirically aggregated into a manageable number of higher-rank dimensions. The factor structure and the psychometric properties of the scale have been shown [26]. The CPQ-AN consists of four subscales, which represent higher-rank change process dimensions: 1—AN-specific cognitions and behaviours (AN-C-B), 2—basic need satisfaction (BSN), 3—emotional involvement and commitment (I-C) and 4—alliance and treatment confidence (A-TC).

The first subscale measures AN-specific cognitions and behaviours such as weight and shape concerns, restraint, dieting and body-checking behaviour. This subscale can be understood as a proximal outcome. Dysfunctional cognitions, in particular negative self-evaluations, are considered to be central to the maintenance of eating disorders [30] (sample item: “I felt a strong desire to reduce my weight”, 15 items). The second subscale, basic need satisfaction, assesses whether different psychological needs are fulfilled within the current therapeutic environment. The subscale measures needs which are considered to be essential to form new schemas and develop health and growth. Among them are the need for competence, relatedness (attachment-supportive experiences), pleasurable experiences and the need for self-esteem and self-efficacy (sample item: “I felt alive and full of energy”, 46 items). The third subscale, emotional involvement and commitment to treatment, captures the extent to which the patient engages with confronting personal topics and to which she actively participates in the treatment process (sample item: “I worked hard on very personal matters.”, 22 items). The last subscale, alliance and treatment confidence, subsumes the quality of the alliance with the treatment providers, the belief in the treatment suitability and the feeling of autonomy within the therapeutic context (sample item: “I do not have the feeling of being fully understood by the therapists”, 21 items). All subscales had an acceptable to excellent reliability (Cronbach’s α 0.73–0.91) [26].

Patients were asked to complete the CPQ-AN upon admission, discharge and on a weekly basis during their inpatient stay. Additionally, a member of staff measured the patients’ height at admission and weight on a weekly basis to enable the calculation of the patients’ BMI. Due to the naturalistic design and individual therapy durations for each patient, the number of measurements per patient varied. We selected three time points of the longitudinal courses of each patient who completed the therapy: the first assessment (t0), the mid-treatment assessmentFootnote 1 (t1) and the last assessment of the inpatient treatment (t2).

Statistical analysis

Firstly, we calculated the descriptive statistics of the baseline information. Secondly, we adjusted the BMI and the CPQ-AN subscales (AN-C-B, BSN, I-C, A-TC) for different treatment durations because of the well-known effect of the treatment duration on inpatient BMI gain [7] and potential effects on the change process variables such as alliance [31]. We calculated correlations between the adjusted variables over time (t0, t1, t2). We inspected the histograms of the variables for skewness and kurtosis. Thirdly, a path model was computed to analyse the interdependencies between BMI and the CPQ-AN subscales over the course of treatment (t0, t1, t2) We started from a model with paths between all variables and time points (t0 → t1; t0 → t2; t1 → t2) and paths for the associations between all variables at baseline (t0). Based on this original model, we performed backward selection of non-significant paths. The significance level was set at p < 0.05 (two sided). We used the root mean square error of approximation (RMSEA) as the main measure of model fit. Values of <0.08 or <0.05 indicate acceptable or good model fit, respectively [32]. As additional measures of fit, we used the Chi-squared values divided by the degrees of freedom (χ 2/df), the Bentler’s Comparative Fit Index (CFI) and the Tucker–Lewis’ Index (TLI). Missing values were imputed automatically by AMOS using a full information maximum likelihood (FIML) algorithm. Finally, we decided to perform a bootstrap procedure as secondary analysis in case of non-normal data.

The statistical analyses were performed in cooperation with the Department of Medical Biometry and Epidemiology of the University Medical Center Hamburg-Eppendorf. We used AMOS 22 for the path model and Stata 13 for all prior analyses.

Results

Patient flow and characteristics

From the 233 patients recruited, data from 176 patients were included in the analyses. Twenty-five patients were removed due to missing data or failing plausibility checks (e.g. missing admission BMI, implausible admission BMI, very early study dropouts). Of the remaining 208 patients, 176 completed the therapy and were subsequently analysed (completion rate 84.7%; dropout rate 15.3%).

The patients had a mean age of 27.1 years (SD = 8.9 years) and a mean BMI of 15.0 kg/m2 (SD = 1.6 kg/m2) upon admission. The average age of illness onset was at 17.8 years (SD = 6.0 years) and the majority of the patients (n = 102 of 170 with non-missing information on previous hospitalizations, 60.0%) had at least one previous inpatient stay associated with a mental disorder. Eighty-eight patients had a restrictive AN-subtype (54.0%) and 75 patients had a purging subtype (46.0%; n = 163 patients with non-missing information on AN-subtype). On average, the patients stayed 11.8 weeks (SD = 5.2 weeks) in the hospital. The sample comprised 29 (16.5%) patients from centre HH, 46 (26.1%) from centre P and 101 (57.4%) from centre BB. A significant difference between therapy completers and patients who dropped out from treatment could only be found for the treatment duration (t = 7.01, p < 0.001).

Path analysis

The correlations between the adjusted BMI and the CPQ-AN subscales over time (t0, t1, t2) are shown in Table 1 of Supplement 1 (S1). The histograms of the vast majority of the variables showed a normal distribution. The variable A-TC at t2 was relatively left skewed and the variable AN-C-B at t2 was to a lesser degree right skewed. The estimates for all paths of the original model are presented in S1 Table 2. The original model already reached a good model fit with an RMSEA coefficient lower than 0.001 (90% CI 0.000; 0.083). Starting from this model, the backward selection took 64 steps. Figure 1 shows the remaining significant interdependencies of the path analysis. A higher BSN and a higher I-C to treatment at the beginning of therapy (t0) positively influenced the BMI at the mid-treatment assessment (t1). BMI at the end of treatment (t2) was related to previous BMI assessments, but not to any of the change process factors. In addition, a greater BSN at t0 led to less AN-C-B at the end of treatment (t2). All of the mentioned interdependencies represent direct predictions.

Fig. 1
figure 1

Path model of the relationship between BMI, AN-specific cognitions and behaviours (AN-C-B), and change process factors over the course of inpatient treatment: beginning (t0), mid-treatment (t1) and end (t2) [standardized path coefficients, RMSEA: 0.000 (90% CI 0.000; 0.039)]

Full size image

With respect to the interdependencies among the change process factors, a greater early A-TC (t0) was found to be positively associated with I-Cat t1. Higher mid-treatment BSN (t1) negatively influenced I-C at t2. Greater AN-C-B at t0 and greater I-C at mid-treatment (t1) were negatively related to A-TC at t2. A-TC did not influence the outcomes.

For clarity reasons, the associations between the variables at t0, and the influences of the previous one on subsequent assessments of the same variable are not shown in Fig. 1. All associations between the variables at t0 were significant, with the exception of BMI and BSN, and BMI and I-C. In every variable, subsequent assessments were significantly influenced by the earlier assessment of the same variable (t0 → t1, t1 → t2). The first assessment (t0) had an additional influence on the last assessment (t2) for BMI and BSN.

The RMSEA coefficient of the final path model was lower than 0.001 (90% CI 0.000; 0.039) indicating a good model fit. All of the additional measures of fit confirmed this result (Χ 2/df = 1.080; TLI = 0.996; CFI = 0.997) [33].

Finally, we performed the bootstrap procedure due to the two non-normally distributed variables (A-TC t2, AN-C-B t2). The results were almost unchanged. However, two paths (BSN t0 → BSN t2 and BSN t0 → AN-C-B t2) only showed a tendency in this analysis, but were no longer significant (p = 0.073 and p = 0.069, respectively).

Discussion

The present study aimed to contribute to the scarce evidence base on the change process in the treatment of patients with AN. In an exploratory analysis, we investigated the temporal interdependencies between higher-rank change process factors, BMI and AN-specific cognitions and behaviours over the course of naturalistic inpatient treatment using a path model.

Relationships between outcomes and change process factors over the course of treatment

Two of the examined higher-rank change process factors exhibited an influence on the mid-treatment BMI (t0 → t1). Firstly, we found that a greater BSN at the beginning of treatment facilitated BMI restoration during the first half of the treatment. This result seems to be reinforced by the finding that a higher early satisfaction of needs also positively influenced the proximal outcome AN-specific cognitions and behaviours at the end of treatment, though the latter result needs to be viewed with caution because it could not be confirmed in the bootstrap model. Both findings are in line with theoretical perspectives formulated within the self-determination theory and the consistency theory [14]. According to these theories, psychopathology originates from long-lasting and frequently occurring states in which basic needs are frustrated or in conflict with each other. Following this idea, the therapeutic environment should provide contrary experiences and facilitate need satisfaction and thereby the subsequent development of new schemas and behavioural options [28]. Based on these present findings, we would recommend that the general regulations and conditions as well as the atmosphere in inpatient units should allow the staff members to be aware of and flexibly address individual needs of patients, especially at the beginning of the treatment. It might even be helpful to directly and routinely ask patients soon after admission whether they noticed modifiable factors interfering with their well-being during their inpatient stay. In the current study, we investigated the basic needs including the need for mastery or competence, relatedness (attachment-supportive experiences) and emotional acceptance, pleasurable experiences and vitality, and self-esteem and self-efficacy.

The other influential change process factor in our analysis was I-C. Higher values on this subscale early in the therapy were positively related to a higher mid-treatment BMI. This subscale captures the aspects of problem confrontation, patient engagement and affective experience, all of which were previously judged to be highly relevant active ingredients in psychotherapy [34]. Furthermore, the present finding is in accordance with the result that the motivational factor ‘readiness to change’ is an indicator of later inpatient AN treatment outcome [20]. According to this finding, it might be advantageous to explicitly encourage patients at the beginning of treatment to use the chance of the protected therapeutic environment to work on confronting personal topics. It would be interesting to investigate if such an explicit focus on emotional involvement could increase the effect of motivational interventions which usually yield only mixed results in actually enhancing motivation [22].

The factor AN-C-B and the factor A-TC showed no influence on BMI in our analyses. The latter result is in line with previous studies and theoretical discussions, which question the importance of the prominent and widely discussed change process factor alliance in the treatment of eating disorders [35, 36].

Due to the impact of malnutrition on cognitive, emotional and interpersonal processing [1, 37], an effect of the BMI itself on the change process factors could have been expected. Interestingly, BMI did not influence any of the change process factors. Furthermore, all of the relevant change process factors were solely direct predictors of outcome. We did not find more complex pathways from a particular change factor via another change factor and subsequently to the outcome.

Relationships among the change process factors over the course of treatment

In addition to the described relationships between outcomes and the change process factors, we found several interdependencies among the change process variables. The association between the baseline severity of AN-C-B and the later alliance was considered in various previous investigations with contradictory results [15–17]. The present result supports the notion of a negative influence of AN-specific cognitions and behaviours on the subsequent alliance [17]. In summary, this relationship remains rather unclear and is potentially dependent on the treatment, the sample and the method of assessment. Another negative influence on alliance late in the progression of treatment was a higher I-C in the middle of the therapy. This finding appears to be contrary to clinical impression and more or less contradicts a previous result according to which higher values on the motivational contemplation scale predicted later positive alliance [21]. The finding is even more surprising in light of the positive effect of the early alliance on the mid-treatment I-C. So, patients with a higher early alliance quality and a greater feeling of treatment suitability confront themselves more with emotional and personal topics and participate more actively in the therapy process at the mid-treatment time point. A possible interpretation for the negative influence of higher mid-treatment I-C on the late A-TC might also be to focus on the aspect ‘treatment confidence’ within this higher-rank factor. So, thoughts that another treatment might help better may be part of a natural re-orientation during the process of ending the treatment. The final interaction between higher BSN at mid-treatment and lower I-C at the end time point could be interpreted similarly. It could be that I-C is less relevant during the last week of therapy which is usually characterized by other topics like the evaluation of the achievement of the treatment goals or the organization of subsequent outpatient therapy.

Strengths and limitations

Our analysis has several strengths and limitations. To begin with, we were able to include a relatively large number of adult AN patients. Moreover, the multi-centre design adds to the generalizability of the results. On the one hand, using the aggregated higher-rank factors which consider a comprehensive range of single AN relevant change process factors, we could gain a broad view on the change process. On the other hand, the aggregated higher-rank factors might hold the disadvantage that small specific influences could not be detected. In general, we can conclude that it might be fruitful to further investigate BSN in the treatment of AN, yet we currently do not know if all basic needs are equally relevant or if specific needs are of higher priority.

A limitation is that our analysis was exploratory due to the lack of previous evidence. Therefore, we have to consider the detected model as a first suggestion which needs to be further validated in independent samples. Furthermore, our analysis was not conducted in the context of a randomized controlled trial. As a consequence, we cannot rule out that we measured naturally occurring rather than treatment-induced change processes. To confirm the causal relationships, an experimental manipulation would be needed [24]. Also, we did not control for differences in the treatment program between patients. In general, the treatment components were equal for all patients. However, fluctuations might have occurred due to vacation or illness. Moreover, we did not consider the effects of the hospital or the therapists in the present analysis due to a relatively small sample size within some of these clusters. Finally, it might have been interesting to use additional instruments to measure the outcome (e.g. general psychopathology) and to validate the aggregation of the change process factors using latent variables.

Conclusion

Our path analysis provides new insights into the change mechanisms underlying the inpatient treatment of adults with AN. The findings highlight the importance of satisfying the basic needs early in treatment. Accordingly, patients might benefit with respect to BMI restoration as well as in a reduction of AN-specific cognitions and behaviours if general regulations and conditions allow individual needs to be flexibly addressed at the beginning of treatment. In addition, the results suggest encouraging emotional involvement and commitment to treatment at the beginning may be beneficial due to their subsequent positive influence on the BMI. Further research on change mechanisms in the treatment of AN is strongly recommended.