Chapter  135:  Management of Eating Disorders

Treatment Studies

Anorexia Nervosa. We divided the treatment literature into medication-only (generally in the context of clinical management or hospitalization), medication plus behavioral intervention, and behavioral intervention only for either adults or adolescents. The literature regarding medication treatments for AN is sparse and inconclusive. The vast majority of studies had small sample sizes and rarely had adequate statistical power to allow for definitive conclusions. Although studies did include medication administered during or after inpatient intervention, no AN studies that systematically combined medication with behavioral interventions met our inclusion criteria, revealing a substantial gap in the literature.

In the behavioral intervention literature, preliminary evidence suggests that cognitive behavioral therapy (CBT) may reduce relapse risk for adults with AN after weight restoration. Sufficient evidence does not exist to determine whether CBT has any effect during the acute phase of the illness, and one study, also requiring replication, showed that a manual-based treatment combining elements of sound clinical management and supportive psychotherapy by a specialist was more effective than CBT during the acute phase. Family therapy as currently conceptualized does not appear to be effective with adults with AN with longer duration of illness. Specific forms of family therapy initially focusing on parental control of renutrition is efficacious in treating AN in adolescents and leads to clinically meaningful weight gain and psychological change. The lack of follow-up data compromises our ability to determine the extent to which treatment gains are maintained.

Bulimia Nervosa. In medication trials, fluoxetine (60 mg/day) administered for 6 weeks to 18 weeks reduced the core bulimia symptoms of binge eating and purging and associated psychological features in the short term. The 60 mg dose performs better than lower doses and is associated with prevention of relapse at 1 year. Evidence for the long-term effectiveness of relatively brief medication treatment does not exist. The optimal duration of treatment and the optimal strategy for maintenance of treatment gains are unknown.

Studies that combine drugs and behavioral interventions provide only preliminary evidence regarding the optimal combination of medication and psychotherapy or self-help. How best to treat individuals who do not respond to CBT or fluoxetine remains a major shortcoming of the literature. For behavioral interventions for BN, CBT administered individually or in group format is effective in reducing the core behavioral symptoms of binge eating and purging and psychological features in both the short and long term. Further evidence is required to establish the role for self-help in reducing bulimic behaviors.

Binge Eating Disorder. For BED, we addressed two critical outcomes—decrease in binge eating and decrease in weight in overweight individuals. Various medications were studied, including selective serotonin reuptake inhibitors (SSRIs); a combined serotonin, dopamine, and norepinephrine uptake inhibitor; tricyclic antidepressants; an anticonvulsant; and one appetite suppressant. In short-term trials, SSRIs led to greater rates of reduction in target eating, psychiatric and weight symptoms, and severity of illness than placebo controls. However, in the absence of clear endpoint data, and in the absence of data regarding abstinence from binge eating, we cannot judge the magnitude of the clinical impact of these interventions. Moreover, in the absence of follow-up data after drug discontinuation, we do not know whether observed changes in binge eating, depression, and weight persist.

The combination of CBT plus medication may improve both binge eating and weight loss, although sufficient trials have not been done to determine definitively which medications are best at producing and maintaining weight loss. Moreover, the optimal duration of medication treatment for sustained weight loss has not yet been addressed empirically.

Collectively, clinical trials incorporating CBT for BED indicated that CBT decreases either the number of binge days or the actual number of reported binge episodes. CBT leads to greater rates of abstinence than does a waiting list control approach when administered either individually or in group format, and this abstinence persists for up to 4 months posttreatment. CBT also improves the psychological aspects of BED, such as ratings of restraint, hunger, and disinhibition. Results are mixed as to whether CBT improves self-rated depression in this population. Finally, CBT does not appear to produce decreases in weight.

Various forms of self-help were efficacious in decreasing binge days, binge eating episodes, and psychological features associated with BED. Self-help also led to greater abstinence from binge eating than waiting list; short-term abstinence rates approximate those seen in face-to-face psychotherapy trials.

Strength of Evidence in Treatment Literature. We graded the strength of the body of evidence for each question separately. For efficacy of treatment (KQ 1), we graded evidence for AN treatment as weak, that for BN medication and behavioral interventions as strong, and that for BED therapies as moderate. For harms associated with treatment (KQ 2), we graded medication interventions for BN and BED as consistently strong; the literatures for all AN interventions and all other BN and BED interventions were graded as weak to nonexistent because many studies failed to address harms associated with treatment. For factors associated with efficacy of treatment (KQ 3), with the exception of behavioral interventions for BN, which we graded as moderate, we graded the literature uniformly as weak. No published literature provided evidence on whether the efficacy of treatment for these conditions differs by sociodemographic factors (KQ 4). Overall, the literature on the treatment of AN in particular was deficient.

Outcomes Literature

Outcomes of Eating Disorders. One prospective cohort study, conducted in Sweden, followed individuals with AN in the community. Over a 10-year period, approximately half of the group had fully recovered; a small percentage continued to suffer from AN, and the remainder still had other eating disorders. Members of the AN group no longer differed from those in the comparison group in terms of weight, but they continued to be more depressed and to suffer from a variety of personality disorders, obsessive-compulsive disorder, Asperger syndrome, and autism spectrum disorders.

The remaining AN studies followed patient populations. Typically, at least one-half of the patients no longer suffered from AN at followup. However, many continued to have other eating disorders such as BN or EDNOS, and mortality was significantly higher than would be expected in the population matched by sex and age. Factors associated with recovery or good outcomes included lower levels of depression and compulsivity. Factors associated with increased mortality included concurrent alcohol and substance use disorders.

All of the BN outcomes studies followed patient populations. This literature emphasizes comparisons of various definitions of disease outcomes and diagnostic subtypes. Generally, more than one-half of the patients followed no longer had a BN diagnosis at the end of the study. A substantial percentage continued to suffer from other eating disorders, but BN was not associated with an increased mortality risk. A limited number of analyses uncovered factors significantly associated with outcomes of this disease, but only depression was consistently associated with worse outcomes.

Only sparse evidence addresses factors associated with BED outcomes. The three included studies have vastly different designs and research questions; more importantly, they do not converge on any systematic findings. Recalling that no studies of EDNOS outcomes exist, we conclude that the literature regarding outcomes of both EDNOS in general and BED in particular is seriously lacking; we believe that no conclusions can be drawn about factors influencing outcomes of these disorders.

Age of AN disease onset was examined in several AN outcomes studies. However, the relation between this variable and outcomes was mixed. No additional differences by participant sex, gender, age, race, ethnicity, or cultural group emerged from the AN, BN, or BED outcomes literature.

Strength of Evidence in Outcomes Literature. The strength of the evidence addressing factors associated with outcomes among individuals with AN and BN is moderate. In contrast, given the limited information about factors related to outcomes among individuals with BED (KQ 5), we rated BED evidence as weak. We used the body of literature concerning KQ 5 to examine differences in outcomes by sociodemographic factors (KQ 6). We graded the AN literature as weak and the BN and BED literature as nonexistent.

Clinical Characteristics

AN is a serious psychiatric illness marked by an inability to maintain a normal healthy body weight, often dropping well below 85 percent of ideal body weight. Patients who are still growing fail to make expected increases in weight (and often height) and bone density. Despite increasing weight loss, individuals with AN continue to obsess about weight, remain dissatisfied with the perceived size of their bodies, and engage in an array of unhealthy behaviors to perpetuate weight loss (e.g., purging, dieting, excessive exercise, fasting). Individuals with AN place central importance on their shape and weight as a marker of self-worth and self-esteem. Although amenorrhea is a diagnostic criterion, it is of questionable relevance. There do not appear to be meaningful differences between individuals with AN who do and do not menstruate.^{4, 5} Typical personality features of individuals with AN include perfectionism, obsessionality, anxiety, harm avoidance, and low self-esteem.⁶

The most common comorbid psychiatric conditions include major depression^{7, 8} and anxiety disorders.^{9, 10} Anxiety disorders often predate the onset of the eating disorder,^{9, 10} and depression often persists post-recovery.¹¹

Diagnostic Criteria

Table 1 presents the diagnostic criteria that authors of articles reviewed in this report use. They include Russell criteria,¹² Feighner criteria,¹³ Diagnostic and Statistical Manual for Mental Disorders III, III-R and IV (DSM III, III-R, and IV),¹⁴–¹⁶ and the International Classification of Diseases-Versions 9 and 10 (ICD-9 and ICD-10).¹⁷

Epidemiology

The mean prevalence of AN in young females in Western Europe and the United States is 0.3 percent.¹ The prevalence of subthreshold AN, defined as one criterion short of threshold, is greater—ranging from 0.37 percent to 1.3 percent.^{18, 19}

Although awareness of the disorder has increased, the data on changing incidence are conflicting. Some studies suggest that the incidence is increasing,²⁰–²⁶ and others report stable rates.²⁷–³¹ Epidemiological studies indicate that the peak age of onset is between 15 and 19 years.³² Anecdotal reports suggest increasing presentations in prepubertal children³³ and new onset cases in mid- and late-life.^{34, 35} The gender ratio for AN is approximately 9:1, women to men.¹⁶

Etiology

The etiology of AN remains incompletely understood. Although numerous psychological, social, and biological factors have been implicated as potentially causal, few specific risk factors have been consistently replicated in studies of the etiology of the disorder.^{36, 37} Although not disorder-specific, common risk factors across eating disorders include sex, race or ethnicity, childhood eating and gastrointestinal problems, elevated shape and weight concerns, negative self-evaluation, sexual abuse and other adverse events, and general psychiatric comorbidity.³⁶ In addition, prematurity, smallness for gestational age, and cephalohematoma have been identified as risk factors for AN.³⁸

The preponderance of reports from western cultures fueled early conceptualizations of AN as a culturally determined disorder, but the past decade of biological and genetic research has revealed that AN is familial³⁹ and that the observed familial aggregation is attributable primarily to genetic factors.⁴⁰–⁴² Moreover, molecular genetic studies have identified areas of the human genome that may harbor susceptibility loci for AN^{43, 44} and specific genes that may influence risk.^{45, 46}

In addition, an array of pharmacologic, genetic, and neuroimaging studies have identified fundamental disturbances in serotonergic function in individuals with AN even after recovery.⁴⁷ Although serotonin has received considerable research attention, given the interrelatedness of neurotransmitter function, other neurotransmitter systems, most notably dopamine, are also implicated in these disorders.⁴⁸ The ultimate understanding of AN etiology will likely include main effects of both biological and environmental factors as well as their interactions and correlations.

Course of Illness

AN has serious medical and psychological consequences that can persist even after recovery. Features associated with the eating disorder including depression, anxiety, social withdrawal, heightened self-consciousness, fatigue, and multiple medical complications.^{7, 49}–⁵¹ The social toll of AN interferes with normal adolescent development.⁵² Across psychiatric disorders, the highest risks of premature death, from both natural and unnatural causes, are from substance abuse and eating disorders.⁵³

A history of AN is associated with greater problems with reproduction,⁵⁴ osteoporosis,⁵⁵–⁵⁷ continued low body mass index (BMI, a commonly used measure of normal weight, overweight, or obesity calculated as weight in kilograms divided by height in meters squared [kg/m²]), and major depression.¹¹ Chapter 6 reviews eating-related, psychological, and biomarker-measured outcomes of AN in detail.

Treatment

Given the high morbidity and mortality associated with AN, developing effective treatments for AN is critical. Because of the frequent medical complications and nutritional compromise, clinical practice typically includes a comprehensive medical evaluation and nutritional counseling. Typically, less medically compromised cases of AN are treated on an outpatient basis by psychiatrists, psychologists, and other therapists with primary care providers managing medical care. Professional organizations have developed several English-language treatment guidelines or position papers for the treatment of AN; these include the American Psychiatric Association,⁵⁸ the National Institute for Clinical Excellence,⁵⁹ the Society for Adolescent Medicine,⁶⁰ the American Academy of Pediatrics,⁶¹ and the Royal Australian and New Zealand College of Psychiatrists.⁶²

Psychotherapeutic approaches include individual psychotherapy (cognitive-behavioral, interpersonal, behavioral, and psychodynamic), family therapy (especially for younger patients), and group therapy. The American Psychiatric Association Working Group on Eating Disorders concluded that hospitalization is appropriate for individuals below 75 percent of ideal body weight.⁵⁸ Weight is not the only parameter to be considered in level of care decisions. Other considerations include medical complications, suicide attempt or plan, failure of outpatient or partial hospitalization treatment, psychiatric comorbidity, role impairment, poor psychosocial support, compromised pregnancy, and lack of availability of less intensive treatment options.⁵⁸ Such treatment commonly involves highly specialized multidisciplinary teams including psychologists, psychiatrists, internists or pediatricians, nutritionists, social workers, and nurse specialists.

Striegel-Moore et al. reported the average length of stay to be 26 days using an insurance database of approximately 4 million individuals in the United States;⁶³ this is substantially shorter than the lengths of stay in other countries, including New Zealand (72 days)⁶⁴ and Europe, which ranges from 40.6 days (Finland) to 135.8 days (Switzerland).⁶⁵ They found that, per patient, AN treatment costs in the United States were higher than those for obsessive-compulsive disorder and comparable to those for schizophrenia, both of which have prevalences similar to those of AN.⁶³

A workshop sponsored by the National Institute of Mental Health (NIMH) examined problems in conducting research on AN treatment.⁶⁶ It highlighted obstacles such as relatively low incidence and prevalence, lack of consensus on best treatments, variable presentation within the patient population based on age and illness factors, high costs of providing treatment, and the complex interaction of medical and psychiatric problems associated with the illness. This report also highlighted the importance of improving and expanding the workforce in the eating disorders research field.

Clinical Characteristics

BN is characterized by recurrent episodes of binge eating in combination with some form of inappropriate compensatory behavior. Binge eating is the consumption of an abnormally large amount of food coupled with a feeling of being out of control. Compensatory behaviors (aimed at preventing weight gain) include self-induced vomiting; the misuse of laxatives, diuretics, or other agents; fasting; and excessive exercise.

The onset of BN usually occurs in adolescence or early adulthood and is most frequently seen in women who are of normal body weight.¹⁶ Although the gender ratio is approximately 9:1, women to men, the diagnostic criteria themselves are gender-biased. In contrast to women, men tend to present with a greater reliance on nonpurging forms of compensatory behavior such as excessive exercise.^{67, 68} Considerations of differences in the clinical presentation of BN in men may lead to revised estimates.^{67, 69}

Approximately 80 percent of patients with BN are diagnosed with another psychiatric disorder at some time in their life.⁷⁰ Commonly comorbid psychiatric conditions include anxiety disorders, major depression, dysthymia, substance use, and personality disorders.^{9, 71}–⁷⁷ Personality features of individuals with BN include some features shared with AN such as high harm avoidance, perfectionism, and low self-esteem. Features more specific to BN include higher novelty seeking, higher impulsivity, lower self-directedness, and lower cooperativeness.⁷⁸–⁸⁰

Diagnostic Criteria

Table 2 presents DSM III, III-R, and IV and ICD-10 diagnostic criteria for BN. According to DSM IV criteria, a diagnosis of BN requires a minimum of 3 months of binge eating and compensatory behavior occurring twice a week or more. Similar to AN, individuals have to report the undue influence of weight and shape on their self-esteem. In addition, BN is diagnosed secondary to AN (i.e., the illness is diagnosed as BN only if the criteria for AN are not met). Thus, to be diagnosed with BN, individuals should have a BMI greater than 17.5 or the equivalent in children and adolescents. The DSM distinguishes two subtypes of BN based on the individualapos;s compensatory behavior: purging (including vomiting and misuse of laxatives, diuretics, or enemas) and nonpurging (restricted eating and exercise). The ICD-10¹⁷describes only the compensatory mechanisms of vomiting and use of purgatives for BN, because of societal pathologizing of vomiting and laxative misuse when compared with exercise or restrictive eating. ICD-10 does acknowledge alternate periods of starvation in BN.

Epidemiology

A recent review estimated the prevalence of BN to be 1 percent for women and 0.1 percent for men across Western Europe and the United States.¹ The prevalence of subthreshold BN was considerably higher: 1.5 percent for full syndrome and 5.4 percent for partial syndrome. Because of the late introduction of BN into psychiatric nomenclature, few studies have explored temporal changes in the incidence of the disorder. Moreover, few studies have estimated the prevalence of BN among children and adolescents.

Etiology

Historically, like AN, BN has been conceptualized as having sociocultural origins. Substantial familial aggregation of BN has been reported.³⁹ Twin studies reveal a moderate to substantial contribution of additive genetic factors (between 54 percent and 83 percent) and unique environmental factors to BN.^{81, 82} Linkage analyses have identified areas on chromosome 10p that may be implicated in BN.⁸³ Numerous candidate genes have been studied for their role in risk for the disorder.⁴⁶

Ongoing biological studies suggest fundamental disturbances in serotonergic function in individuals with BN.^{80, 84} The ultimate understanding of the etiology of BN and of other disturbances that contribute to the development of inappropriate responses to satiety clues⁸⁵ will most likely include main effects of both biological and environmental factors as well as their interactions and correlations.

Course of Illness

Although BN is not typically associated with the serious physical complications normally associated with AN, patients commonly report physical symptoms such as fatigue, lethargy, bloating, and gastrointestinal problems. Individuals with BN who engage in frequent vomiting may experience electrolyte abnormalities, metabolic alkalosis, erosion of dental enamel, swelling of the parotid glands, and scars and calluses on the backs of their hands.⁸⁶ Those who frequently misuse laxatives can have edema, fluid loss and subsequent dehydration, electrolyte abnormalities, metabolic acidosis, and potentially permanent loss of normal bowel function.⁸⁶ Chapter 6 reviews eating-related, psychological, and biomarker-measured outcomes of BN in detail.

Treatment

In the United States, most treatment for BN is conducted on an outpatient basis. Given the frequency of medical⁸⁷ and nutritional complications, a comprehensive medical evaluation is the typical first step in treatment. Thereafter, psychotherapy, delivered either individually or in group format, is usually the cornerstone of BN interventions. Common approaches include cognitive-behavioral therapy and interpersonal psychotherapy. In cases in which the individual is experiencing medical complications of BN, is pregnant, or is unable to bring an entrenched binge-purge cycle under control on an outpatient basis, partial hospitalization or inpatient treatment is often warranted.

In 1996, the Food and Drug Administration (FDA) approved fluoxetine for the treatment of BN. Currently, this is the only FDA-approved medication for the treatment of any eating disorder.

Clinical Characteristics

Eating disorders not otherwise specified (EDNOS) is a diagnostic category that captures those individuals with eating disorders who do not meet criteria for AN or BN. The DSM IV lists six different examples of presentations of EDNOS:

• 1

  all features of AN except amenorrhea;

• 2

  all features of AN except remaining in a normal weight range;

• 3

  all criteria for BN except frequency of binge eating or purging or duration of 3 months;

• 4

  regular inappropriate compensatory behavior after eating small amounts of food;

• 5

  chewing and spitting out food; and

• 6

  binge eating disorder (BED).

Clinical reports suggest that individuals with EDNOS constitute the majority of individuals seeking professional help for an eating disorder.^{88, 89} This suggests that the nomenclature for eating disorders is imperfect. Moreover, our attempts to address the key questions of this evidence report for the global category of EDNOS indicated a paucity of investigations on the nature of the highly heterogeneous category of EDNOS and on the treatment and outcome of specific presentations of EDNOS. We redirected the task to focus on BED, the one category of EDNOS that has a corpus of research.

Diagnostic Criteria

The symptom of binge eating was first recognized in a subset of obese individuals by Stunkard in 1959.⁹⁰ BED has had a slow and controversial evolution in the psychiatric nosology for eating disorders.⁹¹–⁹⁴ DSM IV currently includes BED as a disorder requiring further study.

The DSM IV criteria appear in Table 3. Individuals with BED engage in regular binge eating behavior. A binge eating episode is determined in the same manner as in BN; it requires consumption of an unusually large amount of food and a sense of being out of control. The frequency criterion of twice per week is the same as in BN, although this criterion is not well supported by the literature.^{95, 96} Unlike BN, individuals with BED do not regularly engage in compensatory behaviors. Several other criteria in the provisional BED diagnosis require further empirical support.

Epidemiology

Population-based studies suggest that between 0.7 percent and 3 percent of individuals in community samples meet criteria for BED.^{92, 97}–⁹⁹ Community studies of obese individuals have found a prevalence of BED between 5 percent and 8 percent.^{100, 101} Population-based studies of BED and the component behavior of binge eating report a relatively equal gender distribution,^{92, 99} few differences in prevalence across races or ethnic groups,¹⁰² and possibly increased risk associated with lower socioeconomic status.^{103, 104} In a population-based study of female twins, 37 percent of obese women (BMI ≥ 30) endorsed the symptom of binge eating,¹⁰⁵ representing 2.7 percent of the female population studied.

Etiology

In a community-based case-control study, Fairburn et al.¹⁰⁶ found significant differences in exposure to risk factors between women with BED and healthy controls, but surprisingly few differences between women with BED and BN. In comparison to healthy controls, women with BED reported greater adverse childhood experiences, parental depression, personal vulnerability to depression, and exposure to negative comments about weight, shape, and eating.

BED has been shown to aggregate in families.¹⁰⁷ Although heritability estimates for frank BED are not yet available, the heritability of binge eating in the absence of compensatory behaviors has been estimated to be 41 percent.¹⁰⁸ In addition, binge eating has been explored as a potential intermediate behavioral phenotype in understanding the genetics of obesity. It has also been preliminarily identified in some studies as an important phenotypic characteristic of individuals with a mutation in the melanocortin 4 receptor (MC4R), a candidate gene that influences eating behavior,¹⁰⁹ although this finding has not been replicated.¹¹⁰

Course of Illness

Given that BED has only recently entered the psychiatric nomenclature, we have minimal population-based data on morbidity and mortality. The presence of binge eating or BED in obese individuals carries substantial risk. Obese individuals with binge eating or BED in clinical and community studies report earlier onsets of obesity and dieting,^{92, 111, 112} greater weight fluctuations,¹¹² more cognitive features of disordered eating,¹¹³ lower self-esteem and self-efficacy,¹¹⁴ and higher scores on depression indices.¹¹⁴–¹¹⁷ Chapter 6 reviews eating-related, psychological, and biomarker-measured outcomes of BED in detail.

Treatment

In the United States, treatment for BED is typically conducted on an outpatient basis. Psychological and dietary interventions aim to reduce binge eating and control weight.¹¹⁸ Common psychotherapeutic approaches include cognitive-behavioral and interpersonal psychotherapy; nutritional approaches include very low calorie diets and behavioral self-management strategies.¹¹⁸ Pharmacotherapy targeting both the core symptoms of binge eating and weight loss are also available as off-label interventions.¹¹⁹

Organization

Given that eating disorders are an important public health problem, the Agency for Healthcare Research and Quality (AHRQ), the National Institutes of Health's Office of Research on Women's Health, together with the Health Resources and Services Administration (HRSA), and in consultation with National Institute of Mental Health (NIMH), commissioned an evidence report through its Evidence Based Practice Program and assigned it to the RTI International-University of North Carolina Evidence-Based Practice Center (RTI-UNC EPC). The issue is also of particular concern to the American Psychiatric Association and the Laureate Psychiatric Clinic and Hospital, which nominated the topic.

Chapter 2 describes our methodological approach, including the development of key questions and their analytic framework, our search strategies, and inclusion/exclusion criteria. In Chapters 3 through 5, we separately present the results of our literature search and synthesis on the treatment of each disease (respectively, AN, BN, and BED). Chapter 6 documents our findings about outcomes associated with each disease. Chapter 7 further discusses our findings, grades the strength of the bodies of literature, highlights methodological shortcomings of the extant research, and offers recommendations for future research. Appendixes (available electronically at http://www.ahrq.gov provide a detailed description of our search strings (Appendix A ^*), our quality rating forms (Appendix B), detailed evidence tables (Appendix C), list of excluded studies (Appendix D), and acknowledgments including our Technical Expert Panel and peer reviewers (Appendix E).

Technical Expert Panel

We identified experts in the field of eating disorders to provide assistance throughout the project. The Technical Expert Panel (TEP) (seeAppendix E) contributes to AHRQ's broader goals of (1) creating and maintaining science partnerships as well as public-private partnerships and (2) meeting the needs of an array of potential customers and users of this product. The TEP served as both a resource and sounding board during the project. Our TEP comprised 10 individuals: three psychiatrists and two psychologists with eating disorder expertise; two nurses; one pediatric/adolescent medicine physician; one nutritionist; and one patient advocate.

To ensure accountability and scientifically relevant work, the TEP was called upon to provide guidance at all stages of the project. TEP members participated in conference calls and e-mail exchanges to

• refine the analytic framework and key questions at the beginning of the project;

• refine the scope of the project; and

• discuss inclusion and exclusion criteria.

Because of their extensive knowledge of the literature on eating disorders, including numerous articles authored by TEP members, and their active involvement in professional organizations and as practitioners in the field, we also asked TEP members to participate in external peer review of the draft report.

Uses of This Report

We anticipate this report will be of value to members of the various professional organizations who treat eating disorders. These include the Academy for Eating Disorders, American Academy of Pediatrics, American Academy of Family Practice, American College of Obstetricians and Gynecologists, American Dietetics Association, American Psychiatric Association, American Psychological Association, International Association of Eating Disorders Professionals, National Association of Social Workers, and Society for Adolescent Medicine.

More generally, the report will assist these organizations in their mission to inform and educate practitioners. From this review, the National Institutes of Health can identify serious gaps in the research on eating disorders to guide funding policy. It can inform practitioners on the current evidence about outcomes associated with having these eating disorders and treating patients with them. Researchers will benefit from the concise analysis of the current status of the field, which will enable them to design future studies to address deficiencies in the field. Health educators can use this report to improve health communication. Finally, policymakers can use this report to allocate resources toward future research and initiatives that are likely to be successful.

Key Questions

• 1. What is the evidence for the efficacy of treatments or combination of treatments for each of the following eating disorders: AN, BN, and BED?

• 2. What is the evidence of harms associated with the treatment or combination of treatments for each of the following eating disorders: AN, BN, and BED?

• 3. What factors are associated with the efficacy of treatment among patients with the following eating disorders: AN, BN, and BED?

• 4. Does the efficacy of treatment for AN, BN, and BED differ by sex, gender, age, race, ethnicity, or cultural group?

• 5. What factors are associated with outcomes among individuals with the following eating disorders: AN, BN, and BED?

• 6. Do outcomes for AN, BN, and BED differ by sex, gender, age, race, ethnicity, or cultural group?

In the analytic framework for these questions (Figure 1), we depict the partially overlapping syndromes of AN, BN and BED, the two types of studies included in this review (treatment and outcome analyses), and factors that influence both treatment response and disorder outcome. We do not include in our figure influencing factors, such as physical and sexual abuse, that are not discussed in the literature meeting our inclusion criteria.

Also depicted on the framework are the six KQs discussed in this report. KQ 1 addresses the efficacy of available treatments for the three disorders; we categorize outcomes as eating-related outcomes that deal with the core behavioral and psychological pathology of the disorders, psychiatric or psychological outcomes that focus on the presence of comorbid depression and anxiety, and biomarker outcomes that reflect weight, body mass index (BMI), and other biological indices of the disorders. Treatment may include relapse, diagnostic crossover, and symptomatic change. KQ 2 explores the harms associated with both medication and psychological treatments for these disorders. KQs 3 and 4 highlight the roles of illness-related factors (e.g., comorbid depression, subtype of the eating disorders, early onset of illness) and illness-independent factors (e.g., sex, gender, race or ethnicity, age) in influencing the outcomes of treating these conditions.

KQ 5 addresses short- and long-term outcomes of the disorders. We apply information from observational, cohort, and case series investigations and focus on eating-related, psychiatric or psychological, and biological indices. Finally, KQ 6 highlights whether these outcomes differ by sex, gender, age, race or ethnicity, or cultural groups.

Inclusion and Exclusion Criteria

After discussions with our TEP, we generated a list of article inclusion and exclusion criteria (Table 4) for these KQs. We limited our review to human studies, including participants ages 10 years and older. Although interest is growing in developing appropriate nomenclature and interventions for young children with eating disorders,¹²⁰ we judged this literature to be beyond the scope of this review. We considered studies published in all languages from 1980 to September 2005. We included studies conducted with participants of both sexes, in all nations. The study population must be primarily diagnosed with AN, BN or BED.

We excluded data that combined diseases because such mixed information would preclude us from separately examining evidence on any one of the three conditions. We also excluded editorials, letters, and commentaries; articles that did not report outcomes related to our key questions; and studies that did not provide sufficient information to be abstracted. Studies were required to report on at least one of our outcomes categories of interest: eating, psychiatric and psychological, or biomarker measures.

We defined individuals as having one of the three disorders of interest according to specific diagnostic criteria. We examined the impact of treatment through a review of the RCT efficacy of treatment literature.

To address a TEP concern that the size of the available AN and BED literature was too limited to permit us to constrain this review based on sample size or followup duration, we included very small AN and BED RCT treatment studies in our review (10 or more participants) and did not require specified followup durations for a study to be included. The BN literature, however, is much more voluminous, which allowed us to limit the treatment studies to larger ones (i.e., those with 30 or more participants).

To help ensure that we were not measuring short-term fluctuations in disease symptoms, we required BN efficacy of treatment studies to follow patients for a minimum of 3 months. The decision to place more stringent requirements on the BN literature was made in consultation with our TEP. Because of financial and time considerations, we used a recently completed EPC report entitled Drug Class Review on Second Generation Antidepressants ¹²¹ as a starting point for our discussion of harms or side effects related to receiving treatment for AN, BN, and BED; we then supplemented this information with harms reported in the RCT studies meeting our inclusion criteria.

We examined outcomes related to having one of the three eating disorders through a review of observational studies; outcomes included eating, psychiatric or psychological, and biomarker variables and death. Although many participants followed in these studies have received treatment, the outcomes of interest relate not to efficacy of treatments but rather to disease levels and other problems that persist over time. To avoid reporting short-term fluctuations among the disease populations and to have sufficient sample sizes to observe changes over time, we limited our review to studies of 50 or more individuals, followed for a minimum of 1year, with or without comparison groups. Our TEP concurred with this plan.

For both the RCT and outcome literatures, we were unable to perform pooled meta-analyses. Given the absence of consensus definitions of remission, recovery, and relapse for eating disorders, as well as the overabundance of outcome measures, we judged meta-analysis to be both inadvisable and infeasible.

Literature Search and Retrieval Process

Databases and search terms. To identify the relevant literature for our review, we conducted systematic searches based on search terms, reviewed included studies by our TEP, and hand searched reference lists. We searched standard electronic databases such as MEDLINE®, the Cumulative Index to Nursing and Applied Health (CINAHL), PsycINFO, the Educational Resources Information Center (ERIC), the National AGRICultural OnLine Access (AGRICOLA), and Cochrane Collaboration libraries.

Based on inclusion/exclusion criteria specified above, we generated a list of Medical Subject Heading (MeSH) search terms, supplemented by key word searches of MEDLINE®. Comparable terms were used to search other databases. MeSH terms included anorexia, anorexia nervosa, and bulimia. Text terms included binge eating disorder. We limited our searches by type of study, including RCT, single-blind method, double-blind method, random allocation, longitudinal studies, and observational studies. For interventions, we used therapeutics or cognitive therapy or family therapy or drug therapy or therapy, computer-assisted. For outcomes of disease, we used outcome assessment (health care), treatment outcome, outcome and process assessment (health care), and recurrence. Finally, we asked our external peer reviewers for titles of articles that we may have missed.

Figure 2 presents the yield and results from our searches. We conducted our initial search in late 2004 and updated it in August 2005 (treatment studies) and September 2005 (outcome studies). Beginning with a yield of 2,188 titles and abstracts, we reviewed and further narrowed this pool to 478 articles.

We retained the following for our review to answer KQs about treatment efficacy: 35 articles on AN, 58 articles on BN, and 26 articles on BED. To answer KQs about disease outcomes, we retained 38 articles on AN, 14 articles on BN, 7 articles on both AN and BN, and 3 articles on BED.

Article selection process. Once we had identified articles through the electronic database search, review articles, and bibliographies, we examined titles and abstracts to determine whether the studies met our inclusion criteria. One reviewer initially evaluated abstracts for inclusion or exclusion. If one reviewer concluded that the article should be included, it was retained. Abstracts initially excluded from the study by one reviewer received a second review by senior project staff—Nancy Berkman, PhD, MLIR (Project Director), Cynthia Bulik, PhD (Scientific Director), or Gerald Gartlehner, MD, MPH (UNC Project Manager).

In all, 478 articles appeared to meet our inclusion criteria through abstract review, so we obtained the full articles. For the full article review, one senior reviewer read each article and determined if it met our eligibility criteria. Those articles that the reviewer determined did not meet our criteria were re-reviewed by a second senior reviewer to ensure agreement that the article should be excluded. We assigned each of these articles one or more reasons for exclusion.

Development of Evidence Tables and Data Abstraction Process

The senior staff members for this systematic review jointly developed the evidence tables. We created two designs for the evidence tables, one for KQs 1 to 4 (treatment studies) and one for KQs 5 and 6 (outcome studies). They are intended to provide sufficient information for readers to understand the study and determine its quality; we emphasized presenting information essential to answering the main questions. The formats of the two sets of evidence tables were based on successful designs used for prior systematic reviews.

Columns in the evidence tables for treatment studies report baseline and outcome measures for eating-related, psychological or psychiatric, and biomarker variables. For each outcome measured, the tables present data in a consistent format. Given the large number of outcomes that these studies typically report, our evidence table entries are relatively long. In contrast, the outcome studies evidence tables are shorter. However, because of the appreciable variety of study approaches and outcomes reported in this literature the presentation of outcome data is, by necessity, less consistent than that for the treatment studies.

For this work, the RTI-UNC EPC team decided to abstract data from included articles directly into evidence tables; this system has worked effectively in many of our past reviews. Because we bypassed the use of data abstraction forms, we had significant efficiencies in production.

We trained data abstractors intensively, thoroughly familiarizing them with table designs, required information and formats, and examples of abstracted articles. As the work progressed, we shared various reporting requirements with abstractors to ensure that information appeared in a consistent and easily understandable manner.

For both the treatment and the outcomes literatures, the first reviewer (UNC faculty, postdoctoral psychology fellow, or psychology graduate student) initially entered data from the article into the evidence table. The second reviewer (Drs. Berkman, Bulik, Brownley, Carey, or Gartlehner) read the article and edited the initial table entry for accuracy, completeness, and consistency. All disagreements concerning the information reported in the evidence tables were reconciled by the two abstractors.

The final evidence tables are presented in their entirety in Appendix C ^*. Separate tables are included for treatment studies by disease and type of treatment intervention:

• AN: Evidence Table 1, medication trials; Evidence Table 2, medication plus behavioral intervention trials; Evidence Table 3, behavioral intervention trials (adults); and Evidence Table 4, behavioral intervention trials (adolescents ages 10 and older);

• BN: Evidence Table 5, medication trials; Evidence Table 6, medication plus behavioral intervention trials; Evidence Table 7, behavior intervention with no medications trials; Evidence Table 8, self-help interventions trials; and Evidence Table 9, other interventions trials;

• BED: Evidence Table 10, medication trials; Evidence Table 11, medication plus behavioral interventions trials; Evidence Table 12, behavioral intervention with no medications trials; Evidence Table 13, self-help intervention trials; and Evidence Table 14, other interventions trials.

Appendix C also presents three evidence tables for outcome studies organized only by disease:

• AN outcome studies, Evidence Table 15;

• BN outcome studies, Evidence Table 16; and

• BED outcome studies, Evidence Table 17.

Within each evidence table, entries are listed alphabetically by the last name of the first author. Abbreviations and acronyms used in the tables appear in a glossary at the beginning of the appendix.

Finally, as noted earlier, the number of assessment instruments that investigators used for both diagnosis and outcome measurement in the studies reviewed here was extremely large. To help readers identify these, we created Table 5 (found at the end of this chapter) to briefly identify all measures, their acronyms or abbreviations, and their subscales, with a citation to a definitive source for the instrument.

Quality and Strength of Evidence Evaluation

Rating the quality of individual articles. For this systematic review, we developed our approach to assessing the quality of individual articles using domains and elements recommended in the evidence report by West and colleagues, Systems to Rate the Strength of Scientific Evidence.¹²² We developed two quality-rating forms, one for the treatment literature and the other for the outcomes literature. Quality rating forms did not differ by disease. We tested several drafts of these forms, revising them as needed to ensure that they efficiently captured the desired information. The final grading forms can be found in Appendix B.

We assessed the treatment literature through 25 items in 11 categories: (1) research aim/study question, (2) study population, (3) randomization, (4) blinding, (5) interventions, (6) outcomes, (7) statistical analysis, (8) results, (9) discussion, (10) external validity, and (11) funding/sponsorship. We did not exclude any studies with so-called fatal flaws, such as the approach to randomization. Rather, we reduced the study's overall score if a category was flawed or inadequate. Because patients and those administering interventions in the psychological treatment studies could not be blinded, we did not evaluate these items when studies included these interventions. However, we always evaluated whether the outcome assessor was blinded. Studies that were reported in more than one article were given the same quality grade.

We weighted each item equally and calculated a score out of 100 percent. We then collapsed those scores into three categories: poor, 0 percent to 59 percent; fair, 60 percent to 74 percent; and good, 75 percent or better.

For the outcomes literature, we used 17 items in 8 categories: (1) research aim/study question, (2) study population, (3) eating disorder diagnosis method, (4) study design, (5) statistical analysis, (6) results/outcome measurement, (7) external validity, and (8) discussion. As with the RCTs, we weighted each item equally. Rather than calculating a score out of 100 percent, however, we converted ratings for each item into numeric values of 0, 1, or 2, in which 0 = poor, 1 = fair, and 2 = good. Studies without comparison groups were not evaluated by items addressing this aspect of design. However, studies that included comparison groups were scored as “good” on one item, whereas those without were scored as “poor” on that item. We calculated the mean score for all graded items and we concluded that, overall, an article should be graded as poor with a rating < 1, fair with a rating ≥ 1 and < 1.5, and good with a rating of ≥ 1.5.

Each quality grade was the composite (averaged) rating of two independent evaluators. The only items reconciled between the evaluators were those in which one rater provided a score for the item and the other said the item was not applicable. In assessing quality of the treatment studies, we asked the two evaluators to discuss their results if the difference in their total scores was 20 points or greater, but we did not require them to come to agreement.

Rating the strength of the available evidence. We rated the strength of the evidence base for both interventions and disease outcomes separately for the three diseases, using a single scheme for all bodies of evidence. Starting with the West et al. report that compared various schemes for grading bodies of evidence,¹²² we based our evaluation on criteria developed by Greer et al.,¹²³ which we deemed most applicable to the study designs in this review. It includes three domains: quality of the research, quantity of studies (including number of studies and adequacy of the sample size), and consistency of findings.

We graded the body of literature applicable to each of the six KQs separately. For the treatment literature, we further divided studies by whether the intervention was pharmaceutical, behavioral, or a combination. Three senior staff defined by consensus four strength-of-evidence categories, as follows:

• I. Strong evidence base. The evidence is from studies of strong design; results are both clinically important and consistent with minor exceptions at most; results are free from serious doubts about generalizability, bias, or flaws in research design. Studies with negative results have sufficiently large samples to have adequate statistical power.

• II. Moderate evidence base. The evidence is from studies of strong design, but some uncertainty remains because of inconsistencies or concern about generalizability, bias, research design flaws, or adequate sample size. Alternatively, the evidence is consistent but derives from studies of weaker design.

• III. Weak evidence base. The evidence is from a limited number of studies of weaker design. Studies with strong design either have not been done or are inconclusive.

• IV. No evidence base. No published literature.

Participants

Of the 19 studies rated fair or good, 10 were conducted in the United States, six in the United Kingdom, two in Canada, and one in New Zealand. A total of 891 individuals participated in fair or good clinical trials for AN. One study failed to report sex. From those studies that reported sex, 861 women and 23 men participated. Seventeen studies failed to report ethnicity for participants. Of those that did, 123 participants were identified as white, eight as Asian and three as other ethnicity.

Medication Trials

Table 8 presents results from medication treatment trials for AN, including treatment aims, setting (inpatient or outpatient), and a summary of outcomes. Similar to text, it is organized by medication class. Of the identified AN trials, eight were randomized controlled double-blind medication trials. Medication trials for AN were most commonly conducted in the context of clinical management or during or following inpatient refeeding. Of these, none reported race or ethnicity of participants, while all but one reported sex of participants; six were conducted in the United States. One study explicitly reported intention-to-treat analyses.²¹² The number of participants in the medication trials ranged from 15 to 72, with the total enrollment for all medication trials being 345. Thus, the average number of patients per study was 23. Based on those studies that reported sex, this includes 319 women and 1 man.

Weight gain is the primary outcome variable in the treatment of AN. Secondary outcomes in this population include reduction of the psychological features of AN (e.g., body dissatisfaction and drive for thinness), reduction of associated behaviors such as overexercising, resumption of menses, and, in the bingeing and purging subtype, decreased binge eating and purging behaviors. Additional psychiatric outcomes include reduction in depression and anxiety.

Second-generation antidepressants. The term “second-generation antidepressants” is commonly used in the psychiatric and pharmacological literature to distinguish newer antidepressants such as selective norepinephrine reuptake inhibitors (SNRIs), selective serotonin reuptake inhibitors (SSRIs), bupropion, nefazodone, and trazodone from traditional or first-generation antidepressants such as tricyclic antidepressants and monoamine oxidase inhibitors. We adopted this term to be consistent in terminology with other research conducted in the area of psychopharmacology.

Fluoxetine. Two trials used fluoxetine at different stages of refeeding in AN patients. In an inpatient study, Attia et al.²⁰⁶ randomized 31 females between 16 and 45 years who had achieved weight restoration of at least 65 percent of ideal body weight (IBW) to fluoxetine (60 mg/day) or placebo. The mean BMI at randomization was 15 kg/m². Patients continued to receive psychotherapy. No significant differences emerged between fluoxetine and placebo on weight gain (16 versus 13 pounds), psychological features of eating disorders, or depression or anxiety measures. Three percent of participants dropped out of fluoxetine treatment.

In the second study, patients were randomly assigned to either initiation on fluoxetine or placebo before inpatient discharge with a beginning dosage of 20 mg/day adjusted over 52 weeks to a maximum of 60 mg/day.²⁰⁷ The range of weight for all participants at randomization was 76 percent to 100 percent average body weight (ABW) with the majority above 90 percent. Outpatient psychotherapy was permitted. Dropout was considerable. Of 39 individuals randomized, only 13 remained at the 52-week endpoint (47 percent of fluoxetine and 85 percent of placebo). In this small group of completers, fluoxetine was associated with significantly greater weight gain, reduced anxiety, depression, obsessive-compulsive features, and eating-disorder-related symptoms.

Tricyclic antidepressants. Two trials of fair or good quality investigated tricyclic antidepressant medication use. Neither provided strong data supporting the use of these medications in treating AN patients.

Amitriptyline in doses up to 175 mg/day in 25 youth ages 11 to 17 years led to no significant differences in eating, mood, or weight outcomes in comparison to placebo.²⁰⁹ No patients dropped out in this trial. Halmi et al. compared amitriptyline (160 mg/day) versus cyproheptadine (32 mg/day) versus placebo in 72 females 13 to 36 years, determined to have AN according to the Diagnostic and Statistical Manual, third edition (DSM III).²⁰⁸ Daily caloric intake was significantly higher in cyproheptadine than placebo and significantly fewer days were needed to achieve target weight (in those who did) in both the amitriptyline and cyproheptadine groups, compared with placebo. Drop out was thirty percent in the amitriptyline group, 25 percent in the cyproheptadine group, and 20 percent in the placebo group.

Hormones. Investigators have studied three hormones in the treatment of AN: growth hormone (rGH), testosterone, and estrogen. Three weeks of transdermal testosterone (150 mg or 330 mg) administered to 38 patients with AN ages 18 to 50 led to greater decreases in depression in patients who were depressed at baseline, but differences in weight were not interpretable.²¹¹ Dropout was 13 percent overall.

Growth hormone (15 mg/kg/day) administered to 14 female and 1 male patient receiving inpatient care for AN led to fewer days to display normal orthostatic heart rate response to a standing challenge among the treatment group than among placebo group.²¹² No patient dropped out of this study.

Klibanski et al. compared estrogen/progesterone (0.625 mg Premarin® or 5 mg Provera® per day) versus nonmedication control in 48 females 16 to 43 years and found no differences between groups on bone density at 6 months.²¹⁰ Dropout was 14 percent in hormone group and 4 percent in the nonmedication group.

Hormone treatment during the acute phase of AN illness does not appear to improve bone density.²¹⁰ Scant, preliminary evidence suggests that rGH leads to faster normalization of orthostatic changes seen in AN²¹²and that testosterone improves depression in individuals with AN and depressed mood.²¹¹

Nutritional supplements. The one study of nutritional supplements was performed in 54 female inpatients older than 15 years with 14 mg/day zinc. It provides preliminary evidence that zinc may increase the rate of increase in BMI.²¹³ Dropout was 39 percent in zinc and 32 percent in placebo, suggesting that conclusions from this study must be viewed with great caution.

Summary of drug trials. All eight studies assessing the efficacy of medication interventions on AN examined weight gain; most reported on eating outcomes and some reported on additional symptom change.

Overall, none of the pharmacological interventions for AN had a significant impact on weight gain. Although tricyclic antidepressants may be associated with greater improvement in secondary mood outcomes, this outcome does not appear to be associated with improved weight gain. No trial has been adequately replicated.

Dropout rates for medication studies for AN are substantial, especially in outpatient trials. Conclusions drawn from studies with such high attrition must be reviewed with extreme caution.

Taken together, the literature regarding medication treatments for AN is sparse and inconclusive. The vast majority of studies had small sample sizes and rarely had adequate statistical power to allow for definitive conclusions. Many studies examined patients who were receiving additional treatments in conjunction with the study medication, including psychological interventions and concurrent pharmacological treatments. Some of these studies examined patients who were in inpatient settings, thus limiting generalizability to outpatient treatment. Only one conducted intention-to-treat analyses; the remaining studies reported completer analyses only. With one exception,²⁰⁹ no medication trials have focused on adolescent patients. Because followup was limited, assessing longer-term impact of interventions on such outcomes as bone density was impossible. Finally, only one male participated in any of these studies, thereby making it impossible to draw any conclusions about the pharmacological treatment of AN in boys and men.

Behavioral Intervention Trials

Of the 11 behavior trials rated good or fair (Tables 9 and 10), four focused solely on adolescents (mean ages 14 to 15), six focused solely on adults (approximately 18 years and older), and one combined adolescent and adult patients. Of the 11 trials, four were conducted in the United States. We present behavioral interventions for adults with AN in Table 9.

Behavioral interventions for adults with anorexia nervosa. In the psychotherapy trials for adults only and the combined adult and adolescent trials, investigators tested CBT (three trials), various types of nonspecific therapy (three), family therapies (two), CAT (two), dietary counseling (one), interpersonal psychotherapy (IPT) (one), behavioral therapy (BT) (one), and focal analytic therapy (one).

Cognitive behavioral therapy. CBT studies generally used a form of therapy tailored to AN that focused on cognitive and behavioral features associated with the maintenance of eating pathology. Of the three CBT studies, one followed inpatient weight restoration²²³ and two were done in the underweight state.^{224, 225} CBT significantly reduced relapse risk and increased the likelihood of good outcome compared to nutritional counseling based on nutritional education and food exchanges after inpatient weight restoration.²²³ Of those receiving CBT, a greater number of individuals with good outcomes were also receiving antidepressant medication.

One study of underweight AN outpatients compared CBT with IPT and nonspecific supportive clinical management (NSCM).²²⁴ IPT in the treatment of AN is based on IPT used for the treatment of depression²⁴⁰ and BN;²⁴¹ it focuses on one of four interpersonal problem areas: interpersonal disputes, role transitions, grief, or interpersonal deficits. NSCM was designed for this study to mimic the type of treatment an individual could receive in the community from a provider familiar with the treatment of ED and incorporates elements of sound clinical management and supportive psychotherapy. In an intention-to-treat analysis, NSCM performed significantly better than IPT in producing global good outcome ratings; CBT outcomes fell in between and were not significantly different from the other two outcomes.²²⁴ The second study compared CBT with BT and a control group for 6 months.²²⁵ At 12-month followup, CBT showed no advantage over BT or control in eating, mood, or weight outcomes.

On the basis of one trial, preliminary evidence suggests that CBT delivered after weight restoration may help to decrease relapse. In contrast, when delivered during the acute phase of the illness, CBT does not appear to offer significant advantage over NSCM, which did offer advantage over IPT. No evidence suggests that nutritional counseling alone is efficacious in the treatment of AN.

Cognitive analytic therapy (CAT). The two studies that utilized CAT, a treatment which integrates psychodynamic with behavioral factors and focuses on interpersonal and transference issues, failed to find any advantage of CAT over educational behavioral therapy or focal family therapy in eating, mood, or weight outcomes.^{226, 228} Focal family therapy focused on eliminating the eating disorder from its controlling role in determining the relationship between the patient and other family members.

Family therapy. Of the three studies in this category, Dare et al. found family therapy to be superior to routine treatment but equivalent to a focal time-limited psychodynamic psychotherapy in increasing percentage of adult body weight, restoring menstruation, and decreasing bulimic symptoms; overall clinical improvement was modest, however.²²⁸

Crisp et al.²²⁷ found outpatient individual and family therapy with variable numbers of sessions to be superior to referral to a family physician for increased weight at 1- and 2-year followup.

The efficacy of family therapy in treating adults with AN has not yet been completely addressed. It may be more effective than medical management by a family physician and routine treatment; family therapy (including the family of origin) may be more effective in younger patients with shorter duration of illness. No studies have explored family therapy for adult patients that included the family of insertion (spouse and offspring of the patient) rather than the family of origin.

Behavioral interventions for adolescents with anorexia nervosa. We present behavioral interventions for adolescents with AN in Table 10.

Family therapy. Four family therapy studies focused exclusively on adolescents and one combined adolescent and adult patients.²³¹ Family therapy was more effective for younger patients with earlier onset than for older patients with a more chronic course in the United Kingdom trial performed by Russell et al.²³¹ and the followup by Eisler et al.²³⁹ These studies did not yield evidence that the specific type of family therapy administered was helpful for the older more chronic group.^{228, 231} A form of family therapy focusing initially on parental control of re-nutrition delivered in two different manners revealed a significant advantage of conjoint therapy (family treated as a unit) over separated family therapy (parents and patient seen separately) on eating and mood outcomes but not on weight outcomes.²²¹

In a second study, no differences emerged between family therapy and family psychoeducation on any outcomes at 16 weeks.²²⁹ For a specific form of family therapy, when delivered in conjunction with a common medical and dietary regimen, behavioral family systems therapy (BFST), also characterized initially by parents taking control of renutrition, Robin et al. found BFST to be superior to ego-oriented individual therapy in increasing BMI and restoring menstruation, although neither therapy was superior on eating or mood outcomes.^{230, 237} Addressing the issue of optimal duration of family therapy, Lock et al. randomized adolescents to either short (10 sessions over 6 months) or long (20 sessions over 12 months) manualized family therapy based on the initial parental control of refeeding model²⁴² and found no differences on eating, psychiatric, or biomarker outcomes.²²² Longer-term family therapy suggested that those with more severe eating-related obsessions and nonintact families did better with longer treatment. Finally, in the one study that included both adolescents and adults, family therapy was superior to individual therapy for adolescent patients with shorter duration of illness. This difference did not emerge for adult patients with longer duration of illness.²³¹ Although few differences were observed across interventions, specific forms of family interventions did consistently show improvement over time with adolescent patients.

Summary of behavioral interventions for adults and adolescents with anorexia nervosa. Overall, one study of adults provides tentative evidence that CBT may reduce relapse risk for adults with AN after weight restoration has been accomplished.²²³ Sufficient evidence does not exist to determine whether CBT is effective during the acute phase of the illness (i.e., in the underweight state before weight restoration); one study found that a manualized nonspecific supportive treatment (NCSM) was more effective than CBT or IPT in terms of global outcome during the acute phase.²²⁴ The three family therapy studies provide no support for the efficacy of the type of family therapy delivered in adults with AN with longer duration of illness; the superiority of this approach for younger patients with a shorter illness course is based on one study.²³¹ Two studies failed to find any benefit of CAT for eating, mood, or weight outcomes when compared to other treatments for this population.^{226, 228} No methodologically sound studies that systematically tested combinations of medication and psychotherapy were identified.

Serious methodological concerns arose with some of these trials. Two were very small (8 to 12 participants per group),^{225, 230} which does not provide adequate statistical power for the comparative analyses conducted. In addition, both had marked pretreatment differences between groups. Failure to control for contact time with a clinician while comparing multiple treatments, with some groups getting up to 80 percent more time in treatment than others, was another problem.²²⁸ In addition, only one group of researchers conducted a follow-up study to determine the long-term impact of their interventions.²³⁹

Five studies evaluated family therapy in adolescents with AN. Overall, family therapy based on principles of parental control of initial refeeding leads to clinically meaningful weight gain and psychological change. However, the majority of family therapy studies compares one form of family therapy to another form and were underpowered to detect significant differences between active similar treatments. One study suggested that family therapy was superior to a non-family therapy comparison intervention for adolescent patients with relatively short duration of illness.²³¹ One additional study reported significantly greater weight gain at the end of treatment in family therapy than in ego-oriented individual therapy for adolescent AN patients.²³⁰ The other three studies all involved some sort of family treatment - either comparing conjoint to separated family therapy or comparing family therapy to family psychoeducation.^{221, 229} Conjoint therapy was superior to separated family therapy for improving eating and mood but not weight outcomes.²²¹ Similarly, one study examining family therapy versus family psychoeducation found no differences between groups.²²⁹

Inadequate statistical power was a common problem among the behavioral interventions in AN, and power calculations were rarely reported. No studies had a pure no-treatment condition, which is appropriate given the gravity of the illness, although “usual” treatment took various forms. Many of these studies had adequate power to detect pre-post within-group differences or differences between no treatment and an active treatment, but few were adequately powered to detect differences across two or more treatment groups.

Participants

Of the 38 studies rated fair or good, 19 were conducted in the United States, five in Canada, four in Germany, three in the United Kingdom, two in Australia, and one each in Austria, Finland, New Zealand, and Norway. In addition, one multinational trial had US and Canadian sites; another had German and Australian sites.

Of the fair and good studies, three failed to report the age of participants; of the remainder, the age range of participants was 16 to 61 years with the majority of participants being adults. A total of 3,403 individuals participated in fair or good clinical trials for BN. From those that reported sex, 2,985 women and 23 men participated.

Thirty-one studies failed to report the race or ethnicity of participants. Of those that did, 1,203 participants were identified as white, 79 as nonwhite, 27 as African American, 40 as Hispanic American, 30 as Asian or Pacific Islander, and one as Native American.

Similar to the AN studies, some BN trials also had high attrition. Table 13 documents the percentages of dropouts in total and in each arm of the study. Three studies had five study groups; those are combined with information relating to the fourth treatment group.

Medication-only Trials

We report on 12 randomized controlled double-blind medication-only trials (Table 14). The total number of individuals enrolled was 1,430. Based on studies that reported sex, 1,364 women and 21 men participated in medication-only trials. The number of participants ranged from 26 to 398. The age of participants ranged from 16 to 55. Two trials reported the race of participants; in these, 521 individuals were reported as white and 27 as nonwhite. Seven trials were conducted in the United States, two in Canada, and one each in Australia, Germany, and Finland.

The medication-only trials used the following two designs: medication versus placebo (10) and medication (dose a) versus medication (dose b) versus placebo (1). The results of these studies are presented below by drug class.

Second-generation antidepressants. Fluoxetine. Six trials compared fluoxetine to placebo in outpatient and inpatient settings. The mean age of participants was mid-twenties; no studies of fluoxetine focused exclusively on adolescents.

Overall, fluoxetine (60 mg/day) administered for between 8 weeks and 16 weeks led to significant reductions in binge eating in most^{244, 249, 250, 254} but not all studies.^{248, 252} Fluoxetine (60 mg/day) also performed significantly better than fluoxetine (20 mg/day) in decreasing binge eating.²⁴⁹ No effect of fluoxetine (60 mg/day) compared with placebo was observed in the one study in which patients were already receiving intensive inpatient psychotherapy.²⁴⁸

Fluoxetine (60 mg/day) was superior to placebo in decreasing purging behavior,^{244, 249, 250, 254} although not in the inpatient setting.²⁴⁸

All six fluoxetine trials either failed to report abstinence rates (absence of binge eating and purging behaviors) or did not report whether abstinence rates differed significantly between drug and placebo groups.

With reference to eating-related attitudes, fluoxetine (60 mg/day) was associated with significant improvements in measures of restraint, weight concern, and food preoccupation and with Eating Disorders Inventory (EDI) subscale scores of bulimia, drive for thinness, and body dissatisfaction.^{244, 249, 250, 254} Again, the exception was the inpatient study.²⁴⁸

Fluoxetine had mixed results on depression and anxiety scores. Some studies showed greater efficacy than placebo in decreasing depression scores,^{249, 252} but others showed no advantage of fluoxetine.^{244, 248, 250, 254}

One study explored the efficacy of fluoxetine (60 mg/day) versus placebo in preventing relapse of BN over 52 weeks.²⁵⁴ Relapse rates were significantly lower for those receiving fluoxetine (33 percent) than for those receiving placebo (51 percent). However, dropout was substantial during the observation period (83 percent in the fluoxetine group and 92 percent in the placebo group).

Drop-out rates in fluoxetine arms of these trials ranged from zero (in an inpatient study) to 50 percent (three studies had greater than 40 percent dropout). In one study, dropout was greater in the fluoxetine than in the placebo group,²⁴⁴ in three studies placebo had greater attrition,^{249, 250, 254} and one inpatient study reported no dropout in either group.²⁴⁸

Fluvoxamine. To compare maintenance of therapeutic gains and prevention of relapse of BN after inpatient treatment, Fichter et al. compared fluvoxamine (average dose 182 mg/day) with placebo for 19 weeks.²⁴⁷ Patients treated with fluvoxamine reported fewer urges to binge, lower frequency of vomiting, and lower depression scores than those receiving placebo. Both groups gained weight, with no differences between groups. Fluovoxamine was associated with a lower relapse rate. However, attrition was high (51 percent for those on fluovoxamine and 14 percent for those on placebo).

Trazodone. In a 6-week trial of trazodone (400 mg) versus placebo, trazodone led to significantly greater decreases in the frequency of binge eating and vomiting and decreased fear of eating.²⁵⁵ No differences in depression or anxiety were observed, although baseline levels were not indicative of severe depression.

Tricyclic antidepressants. One 6-week trial of desipramine (200–300 mg/day) versus placebo found the active drug to be significantly more effective than placebo in decreasing binge eating, vomiting, and scores on the Eating Attitudes Test (EAT) and Body Shape Questionnaire (BSQ).²⁵⁷ Abstinence rates from binge eating and purging did not differ between active drug and placebo. Both self-reported depression and anxiety were significantly decreased in the desipramine group compared with the placebo group; clinician-rated depression did not differ significantly. Patients in the desipramine group lost significantly more weight than those in the placebo group, who tended to gain weight. Dropout was 23 percent in the desipramine group and 16 percent in the placebo group.

Anticonvulsants. The single 10-week trial of the anticonvulsant topiramate (mean dose 100 mg/day) led to significantly greater reductions than placebo in the number of binge/purge days reported and in body dissatisfaction, drive for thinness, and EAT scores.^{251, 259} Abstinence rates from binge eating and purging were 22.6 percent for topiramate and 6 percent for placebo (not significantly different). Topiramate was associated with significant reductions in anxiety but not depression, and the topiramate group lost significantly more weight than the placebo group, who tended to gain weight. Dropout from topiramate treatment was 34 percent and 47 percent for placebo.

MAOI. One 8-week trial of brofaromine (mean dose 175 mg/day) revealed no differences between the active drug and placebo on binge eating or psychological features of the eating disorder.²⁵³ Brofaromine did lead to significant reductions in vomiting. Abstinence from binge eating and from vomiting were measured independently and did not differ between groups; no differences were observed on depression or anxiety scores, weight change, or drop-out rates (21 percent brofaromine and 24 percent placebo).

5HT3 antagonist. In a small 4-week trial of ondansetron versus placebo—self-administered when patients had an urge to binge or vomit—the active drug led to significantly greater decreases than placebo in binge and vomit frequencies and time spent in bulimic behavior, and to significant increases in normal meals.²⁴⁶ The investigators did not measure depression or anxiety, and they found no differences in weight change. One patient dropped out from ondansetron, none from placebo.

Summary of medication-only trials. Fluoxetine (60 mg/day) administered for 6 to 18 weeks has been shown in several fair- to good-rated trials to reduce the core bulimia symptoms of binge eating and purging and associated psychological features of the eating disorder in the short term. The 60 mg dose performs better than the 20 mg dose;²⁴⁹ it was also associated with prevention of relapse at 1 year in a study with considerable dropout.²⁵⁴ Considerable evidence exists for the use of 60 mg/day of fluoxetine to treat BN in the short term. Evidence for the long-term effectiveness of relatively brief medication treatment does not exist. The optimal duration of treatment and the optimal strategy for maintenance of treatment gains are unknown.

Single studies provide preliminary evidence of the efficacy of two other second-generation antidepressants, namely trazodone²⁵⁵ and fluvoxamine.²⁴⁷ Likewise, evidence from single studies provides preliminary evidence of the efficacy of desipramine²⁵⁷ and topiramate.²⁵¹ One preliminary trial of ondansetron, a 5HT3 antagonist and antiemetic, led to an intriguing decrease in binge eating and vomiting when patients could self-administer when they had urges to binge or purge.²⁴⁶ This innovative study requires replication. One trial of brofaromine, an MAOI, showed a significantly greater effect on reducing vomiting than placebo.²⁵³

When reported, abstinence rates in medication-only trials suggest that medication treatment leads to abstinence in a minority of individuals. This finding indicates that although bulimia symptoms improved, they nonetheless persisted.

Drop-out rates in medication trials ranged from zero to 51 percent. No drug showed substantially greater attrition than others.

Behavioral Intervention Trials

We report 13 psychotherapy-only trials, four self-help trials, one trial of light therapy, one of guided imagery, and one of crisis prevention. Summary outcomes data for the psychotherapy trials appear in Table 16. The total number of individuals enrolled in psychotherapy, self-help, and other trials was 1,462. From the studies that reported sex of participants, 1,064 women and two men participated. Across these 20 trials, participants ranged in age from 17 to 64 years. Six trials reported race and ethnicity of participants: in all, 410 patients were white; 22 nonwhite; 28 Hispanic American; 26 Asian, Maori, or Pacific Islander; 10 African American; and 1 Native American. In no instance were results analyzed specifically by race or ethnicity group. Of the 20 trials, seven were conducted in the United States, three each in Canada and the United Kingdom, one each in Australia, Austria, Germany, New Zealand, and Norway, and one two-site study in Germany and Australia, and one did not report location.

Psychotherapy trials for bulimia nervosa. Cognitive Behavior Therapy. CBT focusing on cognitive and behavioral factors that maintain bulimic behaviors is the most widely studied intervention for BN. Eleven trials of various designs delivered CBT either individually or in group format. CBT was compared with interpersonal psychotherapy (IPT),^{269, 276, 287, 288} with supportive expressive therapy,²⁷⁸ with nutritional counseling,^{280, 283} and with exercise.²⁸³ One study compared individually with group-administered CBT.²⁷³ Several studies dismantled CBT by comparing complete CBT with behavioral therapy (BT) in the absence of a cognitive component,²⁷⁶ by comparing cognitive therapy only with exposure with response prevention only,²⁷⁴ and by exploring the additive efficacy of exposure with response prevention grafted onto a basis of cognitive therapy.²⁷¹ Exposure with response prevention is defined as exposing individuals to their high-risk cues (e.g., prebinge cues or prepurge cues) and then preventing the response (e.g., binge eating or purging) until the urge to engage in the behavior subsides.

In comparisons of individually administered CBT and IPT tailored for BN, CBT was associated with a significantly greater probability of remission than IPT²⁶⁹ and with greater decreases in vomiting and restraint^{269, 276} and binge eating²⁶⁹ at the end of treatment. In one study at 1-year followup, these differences were no longer apparent.²⁷⁶ Neither CBT nor IPT led to greater improvements in mood or changes in weight. Changes in dietary restraint and in eating self-efficacy mediated change in binge and purge frequency.²⁸⁸ Being in the precontemplation stage of change was associated with failure to achieve remission at the end of treatment.²⁶⁸

When administered in group format, differences between CBT and IPT were less clear. Both group-administered treatments led to significantly greater decreases than waiting list on days binged, psychological features of the eating disorder, disinhibition, and restraint, with no differences observed between the active therapies.²⁸⁷

When compared directly, few differences emerged between group and individual administration of CBT. Both showed decreases in objective and subjective binge episodes, vomiting, laxative use, overexercise and EDI bulimia, drive for thinness, and body dissatisfaction subscale scores.²⁷³ Group CBT was associated with greater decreases in anxiety; individual CBT was associated with significantly higher rates of abstinence. From a cost-effectiveness perspective, the study concluded that group CBT was more economical, given the similarity of outcomes.

In the dismantling studies, which attempted to parse out the effects of various components of CBT, the cognitive component emerged as critical to therapeutic outcome. Complete CBT led to better eating-related outcomes than BT,²⁷⁶ to lower relapse than exposure with response prevention only,²⁷⁴ and to greater abstinence than a self-monitoring only intervention.²⁶⁶

Two studies examined the additive efficacy of exposure with response prevention. Agras and colleagues found no additive benefit of exposure to CBT.²⁶⁶ Bulik et al. first treated all patients with a core of cognitive therapy and then explored the added efficacy of three augmentation strategies: exposure with response prevention to prebinge cues, exposure with response prevention to prepurge cues, and a relaxation therapy control.²⁷⁰ They found no evidence that either exposure treatment led to greater improvement in binge eating and vomiting than the relaxation control.

In other comparisons, cognitive therapy performed better than support only; adding a cognitive component to nutritional counseling led to a significantly greater decrease in drive for thinness than nutritional therapy alone.²⁸⁰ CBT was superior to nutritional counseling alone in improving core binge eating, vomiting, laxative use, and body dissatisfaction. CBT also led to significantly greater decreases than supportive-expressive therapy (a nondirective psychodynamically oriented treatment) in EDI bulimia, EAT scores, food preoccupation, eating concerns, and depression.²⁷⁸ Exercise therapy was superior to CBT at 18-month followup in improving drive for thinness, laxative abuse, and binge eating.²⁸³

Overall, dropout from CBT delivered individually or in group format ranged from 6 percent to 37 percent. Typical rates were about one-quarter of individuals randomized.

Other behavioral interventions. A single study compared nutritional management (focusing on decreasing restraint, detailed nutritional self-monitoring, and stimulus control) to stress management (focusing on decreasing stressors that may trigger binge eating). Both treatments led to significant decreases in binge eating and vomiting; abstinence from binge eating was greater in nutritional management than stress management, although abstinence from vomiting did not differ. Stress management was associated with greater reductions in trait anxiety.²⁸¹

Dialectical behavioral therapy (DBT). DBT focuses on emotional dysregulation as the core problem in BN with symptoms viewed as attempts to manage unpleasant emotional states. A small study showed that patients receiving DBT had significantly greater decreases in binge eating and purging than did those on a waiting list and that abstinence was greater at the end of treatment in the DBT than in the waiting list group.²⁸²

Self-help trials. We present self-help trials for BN in Table 17. In a direct 18-week comparison of guided self-help (manual including visits with nonspecialists in eating disorders to check on progress) with group CBT, both treatments significantly decreased binge eating, vomiting, laxative use, EDI bulimia, drive for thinness and body dissatisfaction.²⁹⁰ At 1-year followup, individuals in the self-help group showed greater reductions in vomiting and EDI bulimia. CBT was associated with greater reductions in drive for thinness over the treatment period and at followup. Both treatments significantly improved depression, with no differences between groups at the end of treatment; however, at followup, individuals in the self-help group had lower depression scores. Of those who completed treatment, a significantly greater number of individuals in the self-help group than in the CBT group were in remission for more than 2 weeks at the end of treatment (74 percent versus 44 percent). No significant change was seen in weight, although those in the self-help condition weighed significantly more at 1 year.

Carter et al. compared CBT-based self-help³⁰²with nonspecific self-help, focusing on self-assertion for women, with a waiting list control group in a 2-month trial.²⁹¹ Both self-help approaches led to significant decreases in objective binge episodes and purging; the waiting list did not. CBT-based self-help was associated with greater reductions in reducing intense exercise than nonspecific self-help or waiting list. No change in depression was observed. Abstinence and weight values were not reported.

To understand the feasibility and efficacy of self-help delivered in general practitioner (GP) offices, Durand and King compared GP-supported CBT-based self-help³⁰³ with specialist outpatient treatment.²⁹² The duration of treatment was at the clinician's discretion. Patients in both groups reported significant decreases in scores on the Bulimic Investigation Test Edinburgh (BITE) and Eating Disorders Examination (EDE) total; however, binge eating and vomiting did not drop significantly. Both groups reported significant decreases in depression, but no treatment was superior. Weight change was not reported. Drop-out rates were similar across groups (24 percent in the GP group and 18 percent in specialist care).

A German study by Thiels et al. compared 16 weeks of CBT with guided self-change using a manual.²⁹³ Guided self-change included 16 sessions with a therapist encouraging use of the manual and addressing motivation, obstacles, and emergent crises. Significant decreases occurred in overeating, vomiting, BITE scores, and EAT scores for both groups combined. Only on BITE scores did the CBT group perform significantly better than the guided self-change group. Depression dropped in both treatment groups with no significant differences between groups. Dropout was 13 percent in CBT and 29 percent in guided self-change.

Additional interventions for bulimia nervosa. We present other interventions for BN in Table 18. Three studies explored interventions that did not fit into our classification scheme: active light (such as that used to treat seasonal affective disorder), crisis prevention, and guided imagery.

Light therapy. In a small 8-week trial of 10,000 lux white light (active light) versus 50 lux red light (control), individuals in the active light group showed significantly greater decreases in binge eating than individuals in the control group.²⁹⁵ Mood improved in both groups but no additional differences were observed for any other eating disorder, psychological, or biomarker outcome. The investigators did not provide long-term follow-up data. Given the size of this trial and the absence of followup, results should be viewed as preliminary.

Crisis prevention. Individuals who were abstinent after a trial of CBT were randomized to either a crisis prevention group in which they were able to contact their clinician to receive up to eight additional visits over 17 months if they felt their condition was deteriorating or a control follow-up-only group.²⁹⁷ The percentage of individuals who resumed binge eating and purging did not differ over the 17-month interval; however, none of the individuals in the crisis prevention group used any of their available calls despite the reappearance of bulimic symptoms.

Guided imagery. Esplen et al. conducted a 6-week trial of patients in a guided imagery group and a control journaling group.²⁹⁶ Guided imagery was based on developing self-comforting in BN.³⁰⁴ Guided imagery led to a significantly greater decrease in measures of binge eating, purging, EDI bulimia, drive for thinness, and body dissatisfaction. At the end of treatment, 21 percent of individuals in guided imagery and no individuals in the control condition were abstinent. Drop-out rates were comparable across groups.

Summary of behavioral interventions for bulimia nervosa. A large number of fair- to good-rated trials provide evidence that CBT administered individually or in group format is effective in reducing the core behavioral symptoms of binge eating and purging and psychological features of BN in both the short and the long term. One study suggests that CBT leads to more rapid reduction of symptoms than IPT.²⁷⁶ Another suggests that individual CBT confers no advantage over the more economical group CBT approach;²⁷³ although this finding is important for service delivery, it requires replication. The cognitive component of CBT appears to be the active ingredient for change, as behavioral interventions alone are not as effective.^{274, 276} Exposure with response prevention, either alone or as an added component to a core of cognitive therapy, offers no additional therapeutic advantage to basic CBT.^{270, 272, 274}

Adding a cognitive component to nutritional intervention led to greater effectiveness in one study,²⁸⁰ and CBT led to better outcomes than a psychodynamically oriented supportive-expressive therapy.²⁷⁸ Preliminary evidence suggests that DBT is effective and worth additional study for the treatment of BN.²⁸²

Four studies provided mixed evidence regarding the efficacy of self-help methods for BN. One German and one Austrian study provide support for guided self-help in comparison to group CBT²⁹⁰ and individually administered CBT.²⁹³ The nature of the self-help approach (CBT oriented versus nonspecific) did not lead to different outcomes.²⁹¹ Preliminary evidence from the United Kingdom indicates that GPs can successfully deliver self-help.²⁹² No self-help trials conducted in the United States met our inclusion criteria. Overall, especially in the absence of control conditions, few conclusions can be drawn regarding the efficacy of self-help approaches for BN. Moreover, the term self-help must be considered carefully as many of the interventions labeled self-help included considerable contact with providers.

One report yielded preliminary evidence for treating BN with light leading to some short-term decreases in binge eating.²⁹⁵ One study provided some support for guided imagery compared to journaling, although long-term maintenance of treatment effects is unknown.²⁹⁶ Crisis prevention approaches do not appear to be effective in the treatment of BN, based on one study, as patients do not avail themselves of the opportunity to contact their therapists when symptoms reemerge.²⁹⁷