Factors associated with treatment attrition in anorexia nervosa : a systematic review

Introduction The reasons for attrition in the therapy of anorexia nervosa are not fully understood. This systematic review provides a summary of trials focusing on factors associated with attrition; it was aimed to compare and contrast findings between different treatment settings. Materials and methods Data were extracted from published reports sourced from online searches (till February 2013) of the following databases: SCOPUS, PubMed and PsycINFO. The search included French and English language papers, and the following search terms were used: ‘anorexia nervosa’ and ‘attrition/drop-out/premature termination of treatment/outcome’. Results Four hundred and twenty-one papers were identified; 34 met the inclusion criteria, four were excluded as they were reviews and three investigated outcome and not attrition. Two papers of the 27 included were qualitative studies. Factors mostly studied were the subtype of anorexia nervosa (n = 8), the personality characteristics or disorder (n = 11) and eating psychopathology (n = 10). Six studies could not find an association with attrition for demographic or other clinical characteristics. The factors associated with attrition in most studies were the subtype of anorexia nervosa, where the purging type was associated with a higher attrition rate than the restrictive type (n = 7 studies), and more frequent personality features and other psychiatric comorbidities. The majority of trials were conducted with adults and inpatient samples. Conclusion More studies with regard to attrition in younger people and outpatient settings, and a more consistent and standardised assessment of attrition in anorexia nervosa research, are needed. Addressing the use of purging behaviours and psychological comorbidities in anorexia nervosa is likely to enhance treatment retention, but further research is required in this regard. Introduction Anorexia nervosa (AN) is defined in the current American Psychiatric Association and other international diagnostic schemes as an eating disorder characterised by dietary restriction leading to low body weight, with psychological and/or behavioural resistance to gaining weight, and overvaluation of shape and weight. There are two subtypes of AN, restricting (AN-R) and binge eating/purging (AN-P)1. AN is a common disorder with a lifetime prevalence of around half to 1% in the general population, this percentage being higher in women (up to 3.5%)2. The predominant treatment modality is a psychological therapy, but the evidence available is limited; attrition is often high and motivation poor3. The high rate of drop-out in AN poses a serious obstacle to successful treatment and is likely to promote poorer outcomes and subsequent chronicity with an increased severity in the course of illness as well as personal and community burden3. The most recent systematic review of attrition studies that we identified included a review of outpatient trials3. This review reported 19 studies with attrition rates of 4.8% for family therapy to 100% for dietary advice, but most attrition rates ranged between 20% and 40%. Seven studies were of individual psychotherapy for adults with AN. The review focused on effects of attrition on outcomes and not on factors that may predict attrition. This review highlighted the inconsistency in results, variations in the definition of drop-out and the proposed new definitions for the types of attrition, e.g., clinical withdrawal versus patient-initiated withdrawal as well as the timing of attrition, e.g., after assessment or at time points during and up to the end of treatment. Prior to this review3, Fassino et al.4 conducted a systematic review of the factors associated with attrition in eating disorders including AN and inpatient and outpatient treatments. They found that AN-P type was the most consistent predictor of attrition. In two inpatient studies and one outpatient study, borderline traits predicted attrition. No other consistent predictors of attrition specific to AN psychotherapy were found in more than one trial. In addition, other reviews have not included non-English language reports and have not discussed the differences or similarities in attrition predictors between inpatient and outpatient Tr ea tm en t * Corresponding author Emails: ghada.el-baky@sswahs.nsw.gov.au; p.hay@uws.edu.au 1 NSW Institute of Psychiatry Fellow, School of Medicine, University of Western Sydney, Sydney, Australia 2 Centre for Health Research School of Medicine, University of Western Sydney, Sydney, Australia 3 School of Medicine, James Cook University, Townsville, Australia 4 School of Psychology and Centre for Eating and Dieting Disorders (Boden Institute), University of Sydney, Sydney, Australia


Introduction
The reasons for attrition in the therapy of anorexia nervosa are not fully understood.This systematic review provides a summary of trials focusing on factors associated with attrition; it was aimed to compare and contrast findings between different treatment settings.

Materials and methods
Data were extracted from published reports sourced from online searches (till February 2013) of the following databases: SCOPUS, PubMed and PsycINFO.The search included French and English language papers, and the following search terms were used: 'anorexia nervosa' and 'attrition/drop-out/premature termination of treatment/outcome'.

Results
Four hundred and twenty-one papers were identified; 34 met the inclusion criteria, four were excluded as they were reviews and three investigated outcome and not attrition.Two papers of the 27 included were qualitative studies.Factors mostly studied were the subtype of anorexia nervosa (n = 8), the personality characteristics or disorder (n = 11) and eating psychopathology (n = 10).Six studies could not find an association with attrition for demographic or other clinical characteristics.The factors associated with attrition in most studies were the subtype of anorexia nervosa, where the purging type was associated with a higher attrition rate than the restrictive type (n = 7 studies), and more frequent personality features and other psychiatric comorbidities.The majority of trials were conducted with adults and inpatient samples.

Conclusion
More studies with regard to attrition in younger people and outpatient settings, and a more consistent and standardised assessment of attrition in anorexia nervosa research, are needed.Addressing the use of purging behaviours and psychological comorbidities in anorexia nervosa is likely to enhance treatment retention, but further research is required in this regard.

Introduction
Anorexia nervosa (AN) is defined in the current American Psychiatric Association and other international diagnostic schemes as an eating disorder characterised by dietary restriction leading to low body weight, with psychological and/or behavioural resistance to gaining weight, and overvaluation of shape and weight.There are two subtypes of AN, restricting (AN-R) and binge eating/purging (AN-P) 1 .AN is a common disorder with a lifetime prevalence of around half to 1% in the general population, this percentage being higher in women (up to 3.5%) 2 .The predominant treatment modality is a psychological therapy, but the evidence available is limited; attrition is often high and motivation poor 3 .
The high rate of drop-out in AN poses a serious obstacle to successful treatment and is likely to promote poorer outcomes and subsequent chronicity with an increased severity in the course of illness as well as personal and community burden 3 .The most recent systematic review of attrition studies that we identified included a review of outpatient trials 3 .This review reported 19 studies with attrition rates of 4.8% for family therapy to 100% for dietary advice, but most attrition rates ranged between 20% and 40%.Seven studies were of individual psychotherapy for adults with AN.The review focused on effects of attrition on outcomes and not on factors that may predict attrition.This review highlighted the inconsistency in results, variations in the definition of drop-out and the proposed new definitions for the types of attrition, e.g., clinical withdrawal versus patient-initiated withdrawal as well as the timing of attrition, e.g., after assessment or at time points during and up to the end of treatment.Prior to this review 3 , Fassino et al. 4 conducted a systematic review of the factors associated with attrition in eating disorders including AN and inpatient and outpatient treatments.They found that AN-P type was the most consistent predictor of attrition.In two inpatient studies and one outpatient study, borderline traits predicted attrition.No other consistent predictors of attrition specific to AN psychotherapy were found in more than one trial.In addition, other reviews have not included non-English language reports and have not discussed the differences or similarities in attrition predictors between inpatient and outpatient Treatment treatment settings.Our aims were therefore to conduct a systematic review including studies since the Fassino et al. 4 review of factors associated with attrition in AN psychological therapy and discuss results across treatment settings, inpatient and outpatient, and across age groups.Our focus was on studies where a psychological therapy was employed as this is the core to all treatment approaches in AN.

Materials and methods
The following inclusion criteria were employed: studies (i) of attrition in AN, or mixed eating disorder diagnostic types where results on AN were reported, (ii) of all age groups, (iii) of psychological therapies, (iii) in either an inpatient or outpatient setting, (iv) written in the French, English or Arabic languages and (v) of original reports published in full.Studies were identified with a systematic online search of SCOPUS, PubMed and PsycINFO data-bases up to the date of 11 February 2013 using the search terms: 'anorexia nervosa' and 'attrition' , or 'drop-out' , or 'premature termination of treatment' or 'outcome' .In addition, a manual search was performed from the references sections of the identified studies.There were no specific exclusion criteria, and qualitative as well quantitative studies were evaluated.
Data extraction and identification of studies were done by the first author and consensus ratings made with the second author.Data extraction included sample and inclusion criteria, definition of attrition, context and type of study, outcome measure, treatment used and predictors of drop-out.

Description of studies
A flowchart of the identification of studies is depicted in Figure 1.Four hundred and twenty-one papers were initially identified in the computerised search; of these, 27  were included and seven were excluded. The even excluded studies were four reviews and three studies that looked at outcomes and not attrition.The 27 studies are summarised in Tables 1-3.Sample sizes in quantitative studies ranged from eight to 320.Two of the 27 studies were qualitative studies 30,31 .Sixteen [1][2][3][4][5] of the 27 studies were conducted in an inpatient setting, eight [22][23][24][25][26][27][28][29] were conducted in an outpatient setting and one was done in both settings 1 .Twelve studies 5,6,7,8 ,9,11,12,19,20,21,24,27 were of AN patients solely, and 13 9,12,13,14,15,16,17,22,24,25,26,28,29 were of mixed samples of bulimia nervosa (BN) or other eating disorders.Two 22,27 were of adolescent or young adult patient samples.In inpatient trials, attrition mostly referred to patients self-discharging prior to completion of treatment or reaching a goal body mass index (kg/m 2 ; BMI) or weight.In outpatient studies, attrition mostly referred to not completing planned numbers of treatment sessions and usually was divided into leaving at early or late stages of therapy.Two trials 17,29 examined failing to attend for followup and two were mixed 22,27 .

Predictors of attrition -inpatient studies
Six 6,7,8,11,12,19 of the inpatient studies examined the subtype of AN and five 6,7,8,12,19 reported that the purging subtype consistently predicted attrition; one study did not find it predictive 11 .Personality was also commonly studied.In seven studies 10,11,14,15,16,19,20 that looked at different aspects of personality character and/or disorder, it was found that having a borderline or antisocial personality organisation was associated with attrition 15,19 .In separate studies, the following features were found to predict attrition: an antisocial attitude 15 , emotional instability 15 , lack of ego mastery 15 , social and emotional alienation 15 , high levels of aggression 10 , extraversion and low level of inhibitedness 10 , high self-transcendence 20 , lower scores on the Temperament   First impressions of therapeuti c alliance were signifi cantly lower for non-completers Premature terminati on associated with lower discharge BMI, and poorer weight gain AD, active discharge by clinician; AN, anorexia nervosa; BMI, body mass index (kg/m 2 ); BN, bulimia nervosa; CBT, cognitive-behavioural therapy; CSA, child sex abuse; DO, discharge against advice; ED, eating disorder; EDNOS, eating disorder not unless otherwise specified, unless otherwise specified DSM-IV diagnostic criteria were used; IBW, ideal body weight; P, binge eating/purging type; R, restricting type; Rχ, therapy or treatment.and Character Inventory (TCI) persistence scale 14 and higher score on the 90-item psychiatric Symptom Checklist (SCL-90) paranoid scale 19 .
However, results were conflicting in the two studies which exam-ined the presence or absence of a comorbid personality disorder 11,20 .The two studies 11,13 that examined other psychiatric disorder comorbidities (depression specified in one study 11 ) also had inconsistent results.
Eating disorder psychopathology was associated with attrition in six studies.Specific pathologies were examined which were predictive of attrition; these included dietary restriction 7 , lower restraint 8  intense maturity fears 8,11 , higher weight concern 8 , having a less negative self-image 16 , presence of more bulimic features 10 and having a low desired weight 19 .Two studies 6,17 were unable to identify any statistically significant associations between eating disorder features and attrition.
One study looked at attachment style 9 .They found that AN-P patients with an anxious attachment were more likely to complete treatment, but AN-R patients who had an avoidant attachment were less likely to complete treatment.In this study, the authors commented that attachment avoidance may thus be a contraindication for group-based, partial, hospital therapy of AN-P, but that attachment anxiety may facilitate remaining in treatment for those with ANP.Attrition was also associated with a high prevalence of parental separation or problem 14,15 ,   Scores on precontemplati on at the beginning were predictors for drop out at the two-year follow-up Character variables such as responsibility, integrity and self acceptance were protecti ve factors at the pre-contemplati on stage-of-change AN, anorexia nervosa; BMI, body mass index (kg/m 2 ); BN, bulimia nervosa; CBT, cognitive-behavioural therapy; ED, eating disorder; EDNOS, eating disorder not unless otherwise specified, unless otherwise specified DSM-IV diagnostic criteria were used; IIP, identity intervention program; P, binge eating/purging type; R, restricting type; Rχ, therapy or treatment; SP, supportive psychotherapy.a lower level of education 14,19 , low BMI 7 and poor therapeutic alliance 21 .

Outpatient studies
There were few studies that examined the same/similar predictors.Like inpatient studies, Bandini et al. 26 found that attrition was significantly higher in patients with AN-P than AN-R, in those with use of other compensatory behaviours and in those with other psychiatric comorbidity.Waller 23 also found that 'bulimic' and borderline features were associated with attrition, but this was a study which included a majority of BN patients.Lock et al. 27 also reported comorbid psychiatric disorder; they also reported that being randomised to longer treatment predicted greater drop-out in an adolescent population.One study 29  predicted a two-year attrition and that personality features such as selfacceptance were protective factors at the pre-contemplation stage.Two studies in younger samples 22,23 examined parental expressed emotions and found over involvement and critical comments to be associated with attrition.

Qualitative studies
Both qualitative studies were conducted in adults.The first looked at the role of reluctance to recover in drop-out of treatment 30 .Seven core obstacles were described that had interfered with the informants' wish to recover.These were: perception of being judged by others, feeling 'stuck' or hopeless, feeling depressed, not acknowledging AN, not wishing to recover through eating or gaining weight and not appreciating the benefit of recovery versus having AN.The second study explored the expe-rience of eight women who dropped out of treatment 31 .In this study, the perceived loss of 'control' in conditions created by the therapeutic milieu appeared to prompt dropping out of treatment.

Discussion
The present review found a small number of consistently investigated predictors of treatment attrition.In two or more studies, AN-P, personality, some eating disorder features and psychiatric co-morbidity were consistently associated with attrition.Most findings, in particular the increased attrition in AN-P, were also consistent across treatment settings.Severity and type of eating disorder psychopathology such as having a low desired weight, bulimic symptoms and more severely restrictive eating, were also associated with attrition in either inpatient or outpa-tient treatment settings.In three of the four (two outpatient) studies investigating psychiatric disorder, such comorbidity was found to be a predictor of drop-outs.
In addition, in almost all studies, the presence of a personality disorder or other personality characteristics associated with a personality disorder, were also associated with higher attrition.The results support other findings that eating disorder patients with more complex psychopathology may require additional treatment to address interpersonal deficits or mood intolerance as in the trans-diagnostic therapy of Fairburn 32 .However, there were inconsistencies; such as, in one study, lower self esteem 24 predicted attrition and in one study, higher self-esteem predicted attrition 16 .Thus, attrition need not imply poorer pre-treatment function.
Competi ng interests: none declared.Confl ict of interests: none declared.
All authors contributed to the concepti on, design, and preparati on of the manuscript, as well as read and approved the fi nal manuscript.All authors abide by the Associati on for Medical Ethics (AME) ethical rules of disclosure.
Four of the inpatient and two of the outpatient studies found no association between demographic features and attrition, and in one study, there was lower attrition with poorer geographic accessibility.Duration of illness and previous treatment were also inconsistently predictive of attrition.These findings support the importance of not rationing care provision on the basis that those from a particular age or other demographic group, or with longer illness duration or a history of previous care may be less (or more) likely to leave treatment prematurely.
There were insufficient studies of adolescents in previous literature for comparative analyses with inpatient or older patient groups.One study examined parental expressed emotions 22 and the other looked at predictors of drop-out and remission in family therapy for adolescent AN in a randomised clinical trial 27 .Higher expression of emotions was identified as a predictor for drop-out in the adult 22 as well as in the adolescent sample 22 .Psychiatric comorbidity was identified as a predictor of dropout in the adolescent study as well as in two of the three adult studies.
Attrition rates in the present review ranged from around 20% to almost 60%, but most attrition rates were around 30%.This negate the view that patients with AN have necessarily high attrition 24,33 despite the known low levels of motivation.However, there was only one quantitative study that examined the stage-of-change 29 and one that reported results on 'cooperativeness' 26 .Themes emergent in the two qualitative studies 30,31 support the need to address the perceived benefits of AN to the patient and its likely contribution to patients leaving treatment early.
The review's strengths were in the systematic search and appraisal of studies and the consensus between two authors on the data extracted.Limitations include the reporting of findings in studies of mixed diagnostic groups where data on AN alone could not be elicited, and that no authors were approached for data and there was no 'grey' literature search.

Critical appraisal of the validity of relevant articles
Articles identified in the present review focussed largely on treatment attrition in inpatient programs.A major gap identified was the need for more research on adolescent samples and the relationship of stage-of-change to attrition.Studies should also comprehensively assess a range of factors that may be associated with attrition and employ multivariate designs to investigate the relative strength of prediction.Quantitative studies have focused on factors in the participants; however, research is needed on factors in the treatment milieu and health care providers.As in prior reviews, disparate definitions of attrition are problematic, and studies should examine trial as well as treatment attrition and the temporal stage of attrition as factors may vary.

Conclusion
Overall, attrition is not necessarily 'high' in AN patients, amounting to usually around one third of patients.Clinicians should not prejudge the likelihood of attrition, even in patients with long standing illness.However, the use of purging behaviours is an important indicator of the increased rate of attrition.Leaving treatment early may also have as much, if not more, to do with therapeutic engagement than with the age of the patient, duration and severity of illness or prior failed therapy.Such engagement rests on factors in the clinic, the clinicians, as well as the readiness for change in the patient.

Clinical applicability
In clinical practice addressing psychiatric comorbidity, in particular mood disorder and borderline personality traits, is important to reduce dropout and/or the need to terminate treatment early.This may include the adjunctive use of psychotropic medications as well as specific psychological therapies such as dialectical behaviour therapy and interpersonal psychotherapy.

Figure 1 :
Figure 1: Flow chart of study selection.
examined the stage-of-change and found that being pre-contemplative Licensee OA Publishing London 2013.Creative Commons Attribution Licence (CC-BY) FOR CITATION PURPOSES: Abdelbaky G, Hay P, Touyz S. Factors associated with treatment attrition in anorexia nervosa: a systematic review.OA Evidence-Based Medicine 2013 Jun 01;1(1):8.

Table 1 Features of inpati ent studies investi gati ng factors associated with att riti on in anorexia nervosa parti cipants. Study Sample and inclusion criteria Att riti on rate and defi niti on Type of study and context Treatment Predictors of att riti on
FOR CITATION PURPOSES: Abdelbaky G, Hay P, Touyz S. Factors associated with treatment attrition in anorexia nervosa: a systematic review.OA Evidence-Based Medicine 2013 Jun 01;1(1):8.Licensee OA Publishing London 2013.Creative Commons Attribution Licence (CC-BY) FOR CITATION PURPOSES: Abdelbaky G, Hay P, Touyz S. Factors associated with treatment attrition in anorexia nervosa: a systematic review.OA Evidence-Based Medicine 2013 Jun 01;1(1):8.

Table Conti nued Study Sample and inclusion criteria Att riti on rate and defi niti on Type of study and context Treatment Predictors of att riti on
Licensee OA Publishing London 2013.Creative Commons Attribution Licence (CC-BY) FOR CITATION PURPOSES: Abdelbaky G, Hay P, Touyz S. Factors associated with treatment attrition in anorexia nervosa: a systematic review.OA Evidence-Based Medicine 2013 Jun 01;1(1):8.

Table 1 Conti nued Study Sample and inclusion criteria Att riti on rate and defi niti on Type of study and context Treatment Predictors of att riti on
Licensee OA Publishing London 2013.Creative Commons Attribution Licence (CC-BY) FOR CITATION PURPOSES: Abdelbaky G, Hay P, Touyz S. Factors associated with treatment attrition in anorexia nervosa: a systematic review.OA Evidence-Based Medicine 2013 Jun 01;1(1):8.

Table 2 Features of outpati ent studies investi gati ng factors associated with att riti on in anorexia nervosa parti cipants.
Licensee OA Publishing London 2013.Creative Commons Attribution Licence (CC-BY) FOR CITATION PURPOSES: Abdelbaky G, Hay P, Touyz S. Factors associated with treatment attrition in anorexia nervosa: a systematic review.OA Evidence-Based Medicine 2013 Jun 01;1(1):8.

Table Conti nued Study Sample and inclusion criteria Att riti on rate and defi niti on Type of study and context Treatment Predictors of att riti on
Licensee OA Publishing London 2013.Creative Commons Attribution Licence (CC-BY) FOR CITATION PURPOSES: Abdelbaky G, Hay P, Touyz S. Factors associated with treatment attrition in anorexia nervosa: a systematic review.OA Evidence-Based Medicine 2013 Jun 01;1(1):8.

Table 3 Qualitati ve studies of att riti on in anorexia nervosa parti cipants. Study Sample and inclusion criteria Defi niti on of att riti on Therapy, methods Findings
Licensee OA Publishing London 2013.Creative Commons Attribution Licence (CC-BY) FOR CITATION PURPOSES: Abdelbaky G, Hay P, Touyz S. Factors associated with treatment attrition in anorexia nervosa: a systematic review.OA Evidence-Based Medicine 2013 Jun 01;1(1):8.