This study is a secondary analysis of data from a randomized controlled trial conducted at two intervention sites (i. University of Sydney and ii. St George’s Hospital at the University of London) and a data and coordinating centre (the University of Chicago) [19]. Sixty-three female participants were randomized to Cognitive Behaviour Therapy for Anorexia Nervosa (CBT-AN) or Specialist Supportive Clinical Management (SSCM). All participants met DSM-IV [20] criteria for AN, excluding criterion D (amenorrhea), for more than 7 years (thus also meeting DSM-5 criteria [21]). Participants received 30 50-minute individual treatment sessions provided over a period of eight months in an outpatient setting.
Participants
Participants were recruited from July 2007 to November 2010 by advertising to clinicians, clinics treating eating disorders (EDs), and on generic websites. After telephone screening (n = 159) to determine eligibility, 73 (46%) were invited for in-person assessment (see Figure 1). Respective site study coordinators described the protocol in detail to participants before written informed consent was obtained and the assessments conducted. Participants were eligible if they were female (males were excluded as we estimated that the number of such cases would be negligible), aged 18 years or more, met DSM-IV [20] criteria for AN, excluding criterion D (amenorrhea), and had an illness duration of at least 7 years. Participants were excluded from the study if they presented with a current manic episode or psychosis, current alcohol or substance abuse or dependence, significant current medical or neurological illness (including seizure disorder), with the exception of nutrition-related alterations that impact on weight, were engaged in psychotherapy and not willing to suspend treatment for the duration of their participation in the study, had plans to move beyond commuting distance from the study site in the following 12 months, or did not live within commuting distance to the study site. Eighty-six per cent (n = 63) of eligible participants agreed to randomization. The majority of those ineligible did not meet DSM weight loss [20, 21] or trial illness duration criteria.
Treatments
Participants were randomly assigned to either CBT-AN or SSCM by a biostatistician in the Data and Coordinating Centre at the University of Chicago, independent from either intervention sites. Randomization was conducted using Ephron’s biased coin approach, stratified within sites by subtype of illness (restrictive and binge-purge) and psychiatric medication status. To minimize therapist effect, three clinical psychologists with prior experience treating eating disorders in adults acted as therapists for both the CBT and SSCM conditions. All therapists attended two 2-day in-person workshops for training in the manualised treatments. All therapists treated pilot cases and were supervised by senior clinicians throughout the duration of the trial. Treatment occurred in outpatient settings at the University of London, and St George’s Hospital, University of London. Both treatments involved 30 individual treatment sessions provided over eight months.
SSCM includes education, care and support, while fostering a therapeutic relationship that promotes adherence to treatment. Supportive psychotherapy aims to assist the patient through use of praise, reassurance and advice. The rationale for this emphasis in treatment is that improvement in domains outside the core pathology can significantly affect patient well-being and disease burden, and research suggests that treatments that target psychosocial functioning are especially appropriate when there has been repeated relapse or a long duration of illness. Ultimately, SSCM aims to help individuals to improve their quality of life, which will further motivate and enable them to make progress on their core ED pathology.
CBT-AN and SSCM were both modified to prioritize quality of life and harm minimization associated with SE-AN, but in both weight gain was an important secondary aim. The primary outcome measures were selected to assess the extent to which individuals were better able to find satisfaction in their lives and engage meaningfully with significant others as a result of treatment. The treatments were distinct in that CBT-AN made use of specific cognitive and behavioural strategies whereas SSCM made use of a more collaborative and supportive therapeutic style.
Assessment
Participants were assessed at pre-treatment and end of treatment by independent assessors blind to treatment assignment. Socio-demographic data were obtained by interview. Eating disorder features were assessed with the Eating Disorder Examination [22]. The EDE is a standardized investigator-based interview that measures the severity of the characteristic psychopathology of eating disorders. Studies consistently support its use, sensitivity, reliability and validity, making it the gold standard for assessing eating disorders. It includes investigator measurement of weight and height to calculate BMI (kg/m2).
Motivation to change was assessed with the Anorexia Nervosa Stages of Change Questionnaire (ANSOCQ) [23]. The ANSOCQ is a 20-item multiple choice questionnaire that assesses a patient’s readiness to recover from AN. It has demonstrated good validity and has high levels of inter-rater and test–retest reliability.
Eating disorder related quality of life was assessed with the Eating Disorder Quality of Life questionnaire (EDQoL) [24]. The EDQoL is a standardized and validated 25-item instrument assessing quality of life in eating disorder populations across four subscales: psychological, physical and cognitive, financial, and work or school. This is the first instrument designed to assess quality of life in eating disorder patients. Because of an inherent bias in existing instruments towards assessing occupational attainment, it is often a challenge to accurately evaluate quality of life in this population. This instrument is uniquely designed to assess relevant components of quality of life for individuals with anorexia nervosa and has demonstrated reliability and validity.
General health related quality of life was assessed with the Short Form-12 Health Status Questionnaire (SF-12) [25]. The well validated SF-12 measures dimensions of health and role limitations due to physical and mental health, for which Mental (MCS) and Physical Component Summary (PCS) scales can be derived.
Depression was assessed with the Beck Depression Inventory (BDI) [26]. The BDI is a 21–question scale with each answer rated 0–3, this scale is widely used across the entire spectrum of studies of psychopathology and psychotherapy.
Social adjustment was assessed with the Weissman Social Adjustment Scale (WSAS) [27] The WSAS assess social adjustment in multiple areas of functioning, including marital, family, work, economic and leisure. The scale has well established reliability and validity and has been used in a wide variety of populations.
Attrition
Previous reviews e.g. [1, 3] have not identified an agreed definition of treatment completion. Furthermore, the small sample size of the present study precluded separate analyses of treatment non-engagement (randomised but did not engage in treatment), early attrition (in this study defined as less than 15 weeks of treatment competed, the definition used in the main report of the trial [19] (p.6) and late attrition (in this study defined as completing 15–29 weeks of treatment). As shown on Figure 1, two participants were randomised but failed to engage in treatment, 55 participants completed at least 15 weeks of treatment (as reported previously [19]), and 48 participants completed 30 or more weeks of treatment.
Statistical analyses
Data were analysed using SPSS v 10. Data were inspected for normality. Univariate analyses were conducted on categorical data using Chi-square test (with Yates correction for small cells), and on continuous data using t-test for parametric data and Mann–Whitney U test for non-parametric data. The alpha level was corrected for multiple testing to ≤ 0.01. Binary logistic regression analysis was used to model the predictors of treatment dropout that were identified in the univariate analyses, namely duration of anorexia nervosa, AN-subtype, global EDE score, EDQoL total and WSAS scores. (Logistic regression was dictated by the dichotomous outcome. More sophisticated multivariable analyses, such as structural equation modelling, were precluded by the insufficient sample size). The model was run with forward entry where the best variable is entered, then the next, etc., until no more variables significantly contribute to the prediction. The model was then run with backward entry whereby all variables are forced in, and one-by-one non-significant variables are removed. As the final results did not differ, only the forward entry model is reported in this paper.