The therapeutic alliance in the early part of cognitive-behavioral therapy for the eating disorders


  • Glenn Waller DPhil,

    Corresponding author
    1. Vincent Square Eating Disorders Clinic, Central and North West London NHS Foundation Trust, London, United Kingdom
    2. Eating Disorders Section, Institute of Psychiatry, King's College London, London, United Kingdom
    • Vincent Square Eating Disorders Clinic, Central and North West London NHS Foundation Trust, London, United Kingdom
    Search for more papers by this author
  • Jane Evans DClinPsy,

    1. Vincent Square Eating Disorders Clinic, Central and North West London NHS Foundation Trust, London, United Kingdom
    Search for more papers by this author
  • Hannah Stringer BSc

    1. Vincent Square Eating Disorders Clinic, Central and North West London NHS Foundation Trust, London, United Kingdom
    2. Loughborough University, School of Sport, Exercise and Health Science, Loughborough, United Kingdom
    Search for more papers by this author



This study examined the strength of the therapeutic alliance in the early stages of cognitive-behavioral therapy (CBT) for the eating disorders, and whether the strength of that allianceis associated with early eating characteristics, comorbid Axis 1 and 2 features.


Forty-four eating-disordered patients completed measures of eating and Axis 1 and 2 characteristics at the start of therapy, and measures of the therapeutic alliance and eating characteristics at the sixth session of CBT.


The therapeutic alliance was strong, including in the domain of attachment. It was unrelated to initial eating pathology and early changes in eating cognitions and behaviors. However, there were links between initial emotional and interpersonal features and therapeutic alliance by the sixth session.


The findings counter suggestions that CBT for eating disorders is characterized by a poor therapeutic relationship. The therapeutic alliance is likely to be enhanced by addressing high levels of emotional distress and difficulties in interpersonal function where appropriate. This research needs to be extended to other therapies, other domains of function and different time points in therapy, to build a fuller picture of the role of the therapeutic relationship in working with the eating disorders. © 2011 by Wiley Periodicals, Inc. (Int J Eat Disord 2012)


It is commonly held that the therapeutic alliance is the most potent factor underlying the impact of psychotherapies, despite there being only weak to moderate associations between the therapeutic relationship and outcomes.1, 2 However, the evidence suggests that the impact of the therapeutic relationship on clinical outcomes varies across therapies. A meta-analysis3 has shown that the therapeutic alliance is relevant to the outcome of therapy only where that therapy is relatively unstructured. In more structured therapies, such as cognitive-behavioral therapy (CBT), the therapeutic relationship is necessary for therapy to have an effect4 but not sufficient, and it appears to be the more specific techniques that have the major therapeutic impact. Indeed, in CBT, there is evidence5 in support of the suggestion6 that it is successful change during therapy that drives a better therapeutic relationship, rather than the other way round, as is commonly assumed.

A pantheoretical approach7 describes the working alliance as having three elements—the establishment of shared goals between patient and clinician; acceptance of the tasks that each needs to perform; and the attachment bond between the two individuals. Such a definition is useful when considering the impact of CBT, as it encompasses both the specific elements that are commonly described in this approach (shared tasks and goals) and the element that is less commonly discussed (the therapeutic bond). The effective therapeutic relationship within CBT in has been described in similar terms,8 focusing on both the tasks/goals and the clinical alliance (“a judicious blend of empathy and firmness”). Such an alliance makes change possible, even if it does not bring that change about.

CBT is the most effective treatment for the majority of adult cases of the eating disorders, although its effects are moderate rather than strong.9, 10 As with other disorders and other therapies,4, 11, 12 it is clear that the therapeutic alliance is a necessary element of CBT for the eating disorders. Therefore, it is important to understand what generates a positive or negative therapeutic relationship in CBT for the eating disorders, in order to identify when the treatment is likely to fail. That understanding is particularly important in the early stages of CBT, where there is some evidence that behavioral change is important.13, 14 Different mechanisms can be hypothesized. First, it is possible that CBT's early stress on behavioral change (one of the “firm” parts of firm empathy) makes for a weak therapeutic alliance in CBT. The second possibility is that the therapeutic alliance is strengthened by early changes in symptoms.5, 6 Finally, it can be hypothesized that the therapeutic relationship is influenced by the individual patient's characteristics, such as pre-existing relationship patterns,15–17 predisposition to trust,18 and symptom severity.19

Therefore, the aim of this study is to examine factors that might influence the therapeutic alliance in during the early part of CBT for the eating disorders, when behavioral change is being stressed. It is hypothesized that CBT will have an impact on eating symptoms, and that the working alliance will be associated with eating pathology, changes in eating pathology, and other characteristics at the start of therapy (particularly Axis 1 and 2 symptoms).



The participants were 42 women and two men undertaking a course of individual cognitive behavioral therapy for adults with eating disorders. A further six patients (two with diagnoses of atypical anorexia nervosa, three with bulimia nervosa and one with atypical bulimia nervosa) commenced the treatment but dropped out (N = 4) or were transferred to another treatment modality (N= 2) over the course of the six sessions. There were no differences in initial levels of body mass index or scores on the measures of eating attitudes and Axis 1 and 2 pathology (below) between those who remained in treatment over the six sessions (t < 1.0 in every case). As therapeutic alliance was not measured until session six, it was not possible to determine whether this characteristic differed between those who did and did not remain in treatment.

Atypical cases of anorexia nervosa and bulimia nervosa were grouped with those with the relevant full disorder. The final sample included 14 patients with a diagnosis of anorexia nervosa (N = 11) or atypical anorexia nervosa (N = 3—all lacking the single criterion of amenorrhea, but no other criteria). The mean BMI of the anorexic group was 15.5 (SD = 1.14). The remaining 30 patients met criteria for a diagnosis of bulimia nervosa (N = 25) or atypical bulimia nervosa (N = 5 — all binging at below the level of two per week, but with other criteria present). The mean BMI of these women was 24.1 (SD = 4.96). The mean age of the 44 patients was 27.2 years (SD = 6.41).

Therapeutic Intervention

The data for this study were collected routinely from individuals who were recruited at the beginning of a course of cognitive behavioral therapy (CBT) for their eating disorder. The clinical approach used20 was based on a transdiagnostic approach to the eating disorders. This was not part of a clinical trial, but part of routine care in this clinical setting. The content of the first six sessions was delivered in an individualized, flexible manner, but adhering to the existing principles of CBT,20, 21 which is an approach that has been supported by clinical research.22 The core tasks of this time period are: engagement; understanding the patient's symptoms, and developing a preliminary formulation; explaining the CBT model; outlining the expectations of the therapy; setting patient goals for the therapy; psychoeducation; the completion of other homework (e.g., food diaries), and addressing any therapy-interfering behaviors (such as non-completion); weekly weighing; and preliminary dietary change.23

Measures and Procedure

Each participant completed three questionnaire measures at the outset of treatment (eating pathology; Axis 1 pathology; Axis 2 pathology), and then two measures following the sixth weekly session (eating pathology; therapeutic relationship). Because the dataset was based on those who completed the measures at Session 6, those who dropped out were not included in the analyses. However, initial levels of eating attitudes, Axis 1 and Axis 2 pathology are broadly unrelated to drop-out from this form of CBT over this early part of treatment.24

Eating Disorders Examination-Questionnaire, Version 6.0

The Eating Disorders Examination-Questionnaire (EDE-Q; 21) was used at Sessions 1 and 6 to measure eating-disordered attitudes (dietary restraint; eating concern; body shape concern; weight concern) and behaviors. Higher scores (range = 0–6) indicate higher levels of eating pathology. Changes in pathology over the first six sessions were determined by subtracting the scores at Session 1 from the scores at Session 6 (i.e., a negative score shows a reduction in symptoms). The original EDE-Q has been validated against the interview form.25–27

Brief Symptom Inventory

The Brief Symptom Inventory (BSI)28 was administered at Session 1 to address Axis 1 psychological symptoms (e.g., anxiety, depression, somatization). It is a 53-item self-report measure, which has been validated widely and which has good psychometric properties.29 Higher scores indicate greater levels of the relevant pathology.

Personality Beliefs Questionnaire-Short Form

This questionnaire was administered at session 1.30 It measures the cognitive component of axis 2 personality pathology, addressing ten personality disorders (although the measure itself is not a diagnostic tool for Axis 2 disorders). The Personality Beliefs Questionnaire-Short Form (PBQ-SF) is a short form of the original measure.31 Both the long and short versions of the PBQ have been shown to have clinical utility and validity with eating disordered populations,32, 33 as well as with other groups.34–36 Higher scores on a scale indicate a higher level of beliefs that underpin that personality disorder.

Working Alliance Inventory - Revised - Short Form

The Working Alliance Inventory Short Form (WAI-SR; 37) was used to assess the therapeutic alliance by the sixth session of CBT. It is a short form of the Working Alliance Inventory.38 This measure is based on the three-element model (37) of the therapeutic relationship—common therapeutic tasks (e.g., “I believe the way we are working with my problem is correct”), shared goals (e.g., “We agree on what is important for me to work on”), and the attachment bond (e.g., “My therapist and I respect each other”). Patients' opinions were measured in preference to clinicians' opinions, as the former have been shown to have better clinical validity. Higher scores (range, 1–7) indicate a better perceived alliance. There were no differences in WAI-SR scores between patients with anorexia nervosa or bulimia nervosa (independent samples t-test: t < 1.38, p > 0.15 for all three WAI-SR scales). Therefore, this eating-disordered group was treated as a single transdiagnostic sample for all remaining analyses.

Data Analysis

Kolmogorov-Smirnov tests demonstrated that the distribution of the eating-related variables met criteria for parametric analyses. However, three BSI scales, one PBQ-SF scale and all three WAI-SR scales were not normally distributed. Therefore, any analyses using those measures were conducted using nonparametric analyses. Paired t-tests were used to test the first hypothesis (change in eating pathology across the first six sessions). The second hypothesis was tested using nonparametric correlations (Spearman's rho), examining the association of the patients' therapeutic relationship scores (WAI-SR) with the measures of eating (at Session 1, Session 6, and the change between the two) and Axis 1 and 2 pathology.


Preliminary Analyses

Table1 shows the group's mean scores on the BSI and PBQ-SF scales at the outset of therapy. Those scores are similar to those of comparable groups.28, 30Table2 shows the eating pathology levels of the 44 patients at the outset of treatment and at Session 6, and the results of paired t-tests used to determine if eating pathology levels altered during this early part of CBT for the eating disorders. There was an overall early change in such symptoms, with the patients showing substantial and statistically significant reductions in EDE-Q scores and levels of objective binging. While the level of vomiting also fell, this change was not statistically significant.

Table 1. Levels of axis 1 characteristics (Brief Symptom Inventory) and axis 2 characteristics (Personality Beliefs Questionnaire) at the start of therapy
Brief symptom inventory  
 Obsessive compulsive1.70(1.00)
 Phobic anxiety0.84(0.97)
 Interpersonal sensitivity1.64(1.10)
Personality Beliefs Questionnaire–short form  
Table 2. Change in eating symptoms across the first six sessions of cognitive behavioral therapy for the eating disorder
 Start of TreatmentSession 6Paired t-Tests
EDE-Q total scores3.75(1.27)2.77(1.29)5.210.001
Objective binges (per 28 days)11.8(12.4)4.75(6.76)4.500.001
Vomiting episodes (per 28 days)11.0(17.2)7.53(17.4)1.780.09

Level of Therapeutic Alliance at the Sixth Session

The patients' mean scores on the WAI-SR at Session 6 were: goals = 6.61 (SD = 0.63); tasks = 6.34 (SD = 0.81); and bond = 6.14 (SD = 1.18). CBT resulted in a therapeutic alliance that was equal or superior across all three domains to that achieved in psychotherapy for a general range of disorders,35 indicating that the patients rated the therapeutic alliance in CBT relatively highly.

Correlations of Working Alliance With Patient Eating Characteristics

Table3 shows that there were no associations (two-tailed Spearman's rho) between therapeutic alliance at Session 6 and any of the eating variables (EDE-Q; BMI; frequency of binging; frequency of vomiting) at either the beginning of treatment or Session 6. Nor was there any association between WAI-SR scores and change in those symptoms. Therefore, the working alliance was not related to levels of eating pathology at any stage or with symptom improvement.

Table 3. Association (Spearman's rho correlations) of early therapeutic alliance (at session 6 of cognitive behavioral therapy) with eating characteristics (at the start of therapy, at session 6, and change between those sessions) [p > 0.10 in all cases]
 Working Alliance Inventory-Short form revised
Eating Disorders Examination-Questionnaire (global score)
 Start of therapy−0.1010.0690.131
 Session 60.002−0.0660.207
 Change between sessions 1 and 6−0.178−0.015−0.286
Body mass index   
 Start of therapy0.1140.0880.196
 Session 60.0990.0610.149
 Change between sessions 1 and 6−0.0750.0010.031
Objective binges   
 Start of therapy−0.103−0.040−0.145
 Session 60.063−0.0260.137
 Change between sessions 1 and 6−0.1190.166−0.087
 Start of therapy−0.113−0.101−0.073
 Session 6−0.156−0.213−0.092
 Change between sessions 1 and 6−0.0350.026−0.062

A key issue is that the very high WAI-SR scores of the sample might have resulted in ceiling effects, thus reducing the power of the correlations to detect associations. An alternative strategy to this dimensional approach is to determine if there is more of a categorical difference, comparing the eating-related scores of patients who score particularly low on the WAI-SR with those of the remainder of the sample. Independent sample t-tests (using equal or unequal variance assumptions, as appropriate) were used to compare eating scores and change levels of the high vs low scorers on the WAI-SR individual scales and total score (carried out using a median split and comparing those with scores above and below the lowest quartile). Those t-tests universally failed to achieve significance (t < 1.4, p > 0.18 in all cases), suggesting that the failure to find correlations was not the result of using a dimensional approach to association.

Correlations of Working Alliance With Patient Axis 1 and 2 Characteristics

Table4 shows the correlations (two-tailed Spearman's rho) between Axis 1 pathology (BSI scales) and the WAI-SR scales. No BSI scales were associated with WAI-SR Tasks of therapy, but three emotion-based elements of Axis 1 pathology (anxiety, depression and interpersonal sensitivity) were associated with WAI-SR Goals of therapy. In each case, higher levels of the emotional state were associated feeling that the goals of therapy were less well shared. Finally, the WAI-SR Bond scale was significantly negatively associated with the BSI psychoticism, depression, and interpersonal sensitivity scales. This pattern suggests that individuals with eating disorders see their attachment relationship with their clinician as less positive if they have these Axis 1 characteristics at the start of therapy.

Table 4. Association of early therapeutic alliance (at session 6 of cognitive behavioral therapy) with axis 1 characteristics (Brief Symptom Inventory) and axis 2 characteristics (Personality Beliefs Questionnaire) at the start of therapy
 Working Alliance Inventory- Short form revised
Brief symptom inventory   
 Obsessive compulsive−0.1590.001−0.128
 Phobic anxiety−0.064−0.091−0.088
 Interpersonal sensitivity−0.370*−0.301−0.390*
Personality Beliefs Questionnaire—Short Form

Table4 also shows the correlations between the working alliance at Session 6 and Axis 2 pathology at the start of therapy. Only three PBQ-SF scales were linked to the therapeutic alliance. First, dependent personality cognitions at the start of treatment were associated with less agreement on shared goals and tasks. Second, avoidant personality cognitions were associated with lower levels of agreement on the goals and bond of therapy. Finally, paranoid personality cognitions were associated with a poorer therapeutic bond.


This study has examined whether patient factors influence the development of an early effective therapeutic alliance in CBT for the eating disorders. Considering the specific aims of the study, the first point of note is that the patients' rating of therapeutic alliance in this group was at a high level, comparable to that found (using different measures) in other studies of CBT for the eating disorders.19, 39, 40 That alliance was unrelated to eating pathology or to early changes in eating pathology. Therefore, there were other characteristics at the start of treatment that influenced that alliance, including emotional and interpersonal elements of Axis 1 pathology (anxiety, depression, social anxiety, and psychoticism) and interpersonal elements of Axis 2 pathology (cognitions that underpin avoidance, dependence, and paranoia).

The finding of a high level of early therapeutic alliance in all domains might not have been predicted for CBT, given the stress on early behavioral change. However, these patients expressed high levels of attachment bond as well as shared goals and tasks, compared with patients in more dynamically oriented therapies.37 It was also noteworthy that eating disorder symptom severity was not related to the level of therapeutic alliance, as this link to key symptoms has been considered in one study of bulimia nervosa.19 The hypothesis that the change in eating pathology would be predictive of improvements in the working alliance6 was not supported, which differs from findings when CBT is used with other disorders,5 although either diagnostic or methodological differences might account for this disparity. The link of higher levels of emotional and personality level comorbidity with a poorer working alliance seems clinically sensible and relates to the wider literature.15–17 For example, the impact of more paranoid personality cognitions upon the therapeutic bond seems to reflect the finding that low trust is a predictor of poor working alliance.18 Similarly, the relationship of anxious and depressive features with lower agreement regarding goals might indicate that the patient is more focused on those features than on the eating pathology per se, and feels that the wrong goals are being stressed by the clinician. Alternatively, the hopelessness or vulnerability associated with depression and anxiety respectively might mean that the patient sees engaging in change as futile or frightening, thus reducing agreement about whether any goals should be worked on. However, it should be remembered that this was against the background of a relatively strong overall alliance, and that the evidence regarding outcome of CBT for the eating disorders does not indicate that the core treatment should be modified for these comorbid problems, even if there needs to be some other work done on such symptoms (e.g., the use of antidepressants; behavioral activation). One important point is that the very high alliance scores in this study mean that there is a risk that the lack of association of the working alliance and other variables (whether assessed dimensionally or categorically) was a product of ceiling effects. Therefore, it would be helpful if future research was carried out using measures of the working alliance with more extensive scales, to determine if the lack of an association was an artifact or a genuine finding.

The high level of therapeutic alliance found in this study is promising and is consistent with previous findings that have used other measures to assess that alliance in CBT for the eating disorders.19, 39, 40 However, there is a clear need for replication and extension in other settings and with other treatments. For example, there is some evidence that the therapeutic alliance grows slightly more across the course of interpersonal psychotherapy than across the course of CBT for bulimia nervosa.41 The time point in therapy where the alliance is measured also has clinical relevance.19 For example, a deterioration in the working alliance is associated with drop-out from therapy for anorexia nervosa, while a longer-term increase in that alliance is associated with completing the treatment.39 That study also reinforces the value of focusing on the patient's view of the therapeutic alliance, as this was more predictive of therapy retention than the clinicians' perceptions. Furthermore, it will be important to consider how the intrapersonal influences on the working alliance studied here interact with other potential influences (e.g., the clinician's personal style; the type of therapy). All of these developments need to be conducted with larger sample sizes and with separate diagnostic subgroups, as the relatively modest number of patients and the heterogeneous nature of the sample in this study means that conclusions are necessarily limited. Such studies will need to consider the working alliance from the earliest point in therapy, as a key issue is whether a poor initial alliance was predictive of drop-out and attrition, thus enhancing the apparent level of alliance among the sample by Session 6.

In clinical terms, these findings suggest that CBT for the eating disorders is not accompanied by a poor level of therapeutic alliance, despite concerns about its focus on behavioral change20, 21 and its balance of firmness and empathy.8 Using Bordin's model and a measure derived from that model (7,37), patients undertaking CBT not only report high levels of agreement on the goals and tasks of therapy, but also a strong attachment bond. In short, clinicians need not be concerned that delivering CBT for the eating disorders makes for a poor working alliance. However, where the patient has relatively high levels of emotional distress and difficulties with interpersonal function, it might be necessary to add those elements to the case formulation and work with them from a relatively early point, using extended elements of CBT for the eating disorders.21

Earn CE credit for this article!

Visit: for additional information. There may be a delay in the posting of the article, so continue to check back and look for the section on Eating Disorders. Additional information about the program is available at