We identified 27 studies meeting eligibility criteria (see Fig. 1). An overview of pertinent study characteristics and definitions of recovery/remission and relapse in AN are listed in Tables 1 and 2. Definitions of relapse were fundamental to understanding the reported rates in these studies. Our review revealed widely varied definitions of relapse and recovery/remission in AN. Definitions of recovery and remission are reviewed first since relapse is predicated upon them.
Definitions of recovery and remission
Recovery typically requires an extended period of time during which minimal or no criteria for the disorder are met, whereas remission requires a shorter duration . The literature can roughly be divided into articles that (1) define remission/recovery based solely on weight measurement, (2) define remission/recovery based solely on symptom reports, (3) define remission/recovery based solely on weight and symptom reports, i.e., diagnostic criteria available at the time. We briefly review these studies next (Table 1 lists studies providing definitions of partial remission, full remission, and recovery).
Several studies used body mass index (BMI) as the only criterion for recovery. Cutoffs included a BMI above 19  or 20 [7, 15]. In contrast, some described remission based solely on psychiatric symptoms. In one, full remission was defined as an absence of all symptoms or only “residual symptoms” for at least 12 weeks, and partial remission was defined as a reduction of symptoms to a sub-diagnostic level for at least 12 weeks . Adopted from the MacArthur guidelines for depression , Keel et al.  defined full remission as a Psychiatric Status Rating (PSR) score of ≤2 for 8 weeks. Clausen  used the same score for 12 weeks, and defined partial remission as a PSR ≤3 for 12 weeks.
Other articles described outcomes in terms of body weight and menstruation, using terminology such as “good,” “intermediate,” “poor,” or “died” [19,20,21,22]. These criteria, or modifications of them, are often referred to as the “Morgan-Russell” criteria . A later version specified remission as weight ≥85% of ideal body weight, regular menses, and no bingeing or purging behaviors . Modifying these criteria, recovery was later defined as not meeting AN DSM-IV-TR criteria for a minimum of 8 weeks .
Several proposed definitions included both weight and clinical symptoms. Pike  defined remission as ≥90% of ideal body weight, resumption of menses, absence of compensatory behaviors, and Eating Disorder Examination (EDE)  subscales within 2 standard deviations (SD) of normal. Recovery was defined as meeting remission criteria for at least 8 weeks. Strober et al.  defined full recovery as the absence of all criteria for at least 8 weeks, and partial recovery as a “good outcome” (weight within 15% of average and normal menstruation) from the Morgan-Russell criteria . Other studies did not have a duration criterion for the absence of symptoms but used the “good outcome” criteria to define recovery [20,21,22]. Stice’s Eating Disorder Diagnostic Scale defined remission as BMI ≥17.5, regular menses, and no subthreshold or full threshold eating disorder [26, 27]. Martin  defined recovered as having a global rating scale of “excellent,” meaning an individual was >90% ideal weight, had regular menstruation, and normal eating and social patterns. Eckert et al.  defined “recovered” as within 15% of ideal body weight, cyclical menses, and no significant disturbance in eating or weight control behaviors or body image disturbance. Kordy et al.  defined full recovery for restricting AN as a BMI >19 and no extreme fear of weight gain for 12 months (plus no purging and no binges for 12 months for AN-BP). They defined full remission for both subtypes as meeting the same criteria for 3 months. Partial remission was a BMI >17.5 and ≤1 binge per week and no vomiting or laxative abuse for 1 month in AN-BP. Another proposed definition of full recovery was a BMI ≥18.5, absence of binging, purging, or fasting for at least 3 months, not meeting criteria for a current eating disorder, and all EDE-Questionnaire (EDE-Q) subscales within 1 SD of normal . They defined partial recovery as the same without the EDE-Q criterion.
Definitions of relapse
Different definitions of relapse were identified (see Table 2). Some definitions were dependent on weight or BMI measures including: BMI < 16.5 for 2 weeks , and BMI < 17.5 [7, 15] or <18.5  for three consecutive months. Other definitions included 15% loss of average body weight after achieving normal body weight, either during the index hospitalization or any time during the 10-year follow-up period . Strober et al.  similarly defined relapse as <85% ideal body weight, which could occur post-discharge or post-recovery. Furthermore, relapse could be partial if the individual had recurrence of psychological symptoms but sustained 85% of ideal weight, or full relapse if both psychological symptoms returned and body weight dropped to less than 85%. Several groups [19,20,21,22, 24] defined relapse as Morgan-Russell criteria of “poor” (BMI ≤18.5).
Other definitions of relapse were dependent on psychiatric symptoms or a combination of psychiatric symptoms and weight changes. Kordy et al.  used a definition of change from DSM-IV partial or full remission to full syndrome. Clausen  defined relapse as PSR ≥ 3 or PSR ≤ 2 after 3 months remission. Relapse has also been defined as meeting full syndrome criteria (PSR ≥ 5) after 8 weeks of remission [17, 32] and after 12 weeks of remission . Pike’s  more in-depth definition of relapse includes weight loss, EDE increase, medical issues, and a return of disordered eating, whereas Martin’s  is the simplest, requiring only that an individual needs psychiatric intervention.
Rates of Relapse
Relapse rates of AN were highly variable ranging from a low of 9% to a high of 52% following treatment, with the majority of studies reporting rates greater than 25% [4, 7, 10, 14,15,16,17,18, 21, 22, 24, 28, 29, 32,33,34]. Studies suggest that adolescents [4, 20, 28] and individuals with restricting subtype AN [7, 29] have a lower likelihood of relapse. The first year is the most critical, with particular risk of relapse occurring as early as 3 months post-treatment [4, 7, 15, 32]. Not surprisingly, those who recover fully have lower relapse rates (9%) than those who only partially recover (35%) . Together, these results suggest that while most patients experience brief episodes of recovery, a large proportion relapse. Moreover, the risk is particularly high within the first year.
There was substantial variability in the literature for follow-up procedures. Initial evaluation time points ranged from 4 weeks to 17 months post-treatment [4, 7, 14, 15, 17, 20, 28, 32, 35]. Some studies utilized only a single follow-up time point [15, 28], whereas others followed patients across multiple time points [4, 7, 14, 17, 20, 32, 35]. Some studies had regular follow-up visits (e.g., every 4 weeks , 3 months ), whereas others had irregularly spaced follow-ups (e.g., 2, 6 and 12 year follow up ).
Variable follow-up intervals could complicate estimations of relapse rates, since relapse rates can vary by duration of the study follow-up. According to this view, shorter follow-up durations might be associated with lower relapse rates than longer durations. We identified articles supporting this possibility. For example, relapse in a study measuring at 6 months was lower (9% for fully recovered and 35% for partially recovered)  versus studies measuring at 1-year (27–70%) [7, 14] (see Table 3). Relapse rates also varied by remission criteria, with stricter remission criteria displaying lower relapse rates than less stringent criteria. This is evidenced by two 10-year longitudinal studies. Eckert and colleagues  reported higher relapse rates (42%) with less stringent relapse criteria and Strober and colleagues  reported lower relapse rates (29.5%) with stricter relapse criteria.