This study aimed to first to investigate associations between health related quality of life, stage of change and other previously found factors such as illness sub-type and treatment attrition in women with SE-AN, and the second aim was to explore the strength of association between adaptive function and stage of change and other more established predictors of attrition. In this study, as in Halmi et al.  there were few factors associated with treatment dropout; one of which was however not previously reported namely, poor eating disorder quality of life. The previously consistently identified factor of AN-purging subtype [1–3] was as well associated with treatment attrition in this study. There were no significant differences between attrition and therapy randomized to, educational level, motivational stage of change, BMI, global eating disorder psychopathology, general health related quality of life, social adjustment, duration of illness or level of depression. Furthermore, the strongest predictors on multivariable analysis were eating disorder quality of life and AN-purging subtype.
Our findings are consistent with many others that have reported AN-purging subtype to predict treatment attrition [1, 2] in both inpatients and outpatient setting. We similarly found the EDE global and subscale scores to be associated with higher a level of attrition which supports the majority of previous studies. Thus, the more severely ill participants, in terms of behavioural, social and psychological parameters, the more likely they are to drop out of treatment, and this was unaffected by the specific psychological therapy they received. Clinicians should take particular effort to engage and retain these patients in care, as they have more severe illness and whilst in most need of care are more likely to terminate prematurely.
To our knowledge, this is the first study to find that overall eating disorder specific quality of life predicted attrition. In addition overall specific quality of life was the strongest predictor in the multivariable model in this study. As persistent anorexia nervosa most often over time is associated with an increase in social and interpersonal deficits eating disorder quality of life may be an even stronger predictor in studies where there is more variability in illness duration and adaptive function. Our findings highlight the importance of further investigation of the relationship between this and attrition, and interventions to prevent and/or break a cycle of functional decline and premature termination of treatment. The generic measure of quality of life and eating disorder specific quality of life in the physical and cognitive domains (as compared to financial, psychological, work and schooling) did not however predict attrition. This may be because the ego-syntonic nature of anorexia nervosa is associated with denial or minimisation of the physical and cognitive effects of illness.
Whilst depression was associated with higher level of attrition this was not statistically significant, and depression has not consistently been associated with attrition in other studies. For example, Masson et al.  reported that co-morbidity with depression in fact enhanced treatment completion in the inpatient setting. However, it is possible the study lacked statistical power to detect a difference. This may also apply to the findings for global eating disorder psychopathology, social adjustment and duration of illness where there appeared to be associations that however failed to reach significance. Caution should thus be applied to these negative findings. The lack of statistical association between preadmission BMI and premature treatment termination that we found may have been because this was an outpatient trial and BMI was not as low or as variable as BMI as in other studies. However, most other studies have also not found an association with BMI and attrition.
Our findings did not support those of Dalle Grave et al.  and Huas et al.  who found an association between level of education and attrition. This may have been because of the relatively high level of graduate education (67%) and low level of variability in education level in the present study participants.
Findings continue to be inconsistent with regard to the few studies investigating level of motivation and treatment attrition. In our study we did not confirm our hypothesis that stage of change would predict treatment attrition. This may have been because the motivation to participate in this study was for the stated primary aim of improved quality of life rather than weight gain (although weight gain was clearly an important secondary aim). However, the ANSOCQ questionnaire measures motivation related to reduction in eating disorder behaviours and increase in body weight, not change in quality of life. The relationship between stage of change and various measure of treatment outcome is also very complex. For example, whilst finding no relationship between stage of change and BMI outcome, Mander at al.  found being at the stage of contemplation predicted treatment alliance.
The strengths of this study are the comprehensive range of predictors and validated assessment instruments. The major limitations are the small sample and unequal groups (n = 15 non completers vs. 48 completers) which increased the likelihood of Type II statistical error, precluded separate analyses of participants with early and later attrition and examination of bidrectionality of relationships using more sophisticated analytic methods such as structural equation modelling. Unequal group sizes and small numbers in the attrition arm were the results of the low rates of dropout in the present study. This contrasts with other outpatient studies, where completion rates can be as low as 27% . A more recent outpatient trial of Zipfel et al.  also reported a treatment completion rate of 74% but Schmidt et al.  reported that only 25% participants completed the majority (>61%) of sessions. The low rates in the present may have been because of the modification of treatment goals and tailoring of treatment to the participant’s stage of change. Finally, co-morbidity more broadly, i.e. personality characteristics, could not be examined in the present study. The findings as well may not generalise to participants with shorter illness duration.