Outcomes in people with eating disorders: a transdiagnostic and disorder‐specific systematic review, meta‐analysis and multivariable meta‐regression analysis

Abstract

Eating disorders (EDs) are known to be associated with high mortality and often chronic and severe course, but a recent comprehensive systematic review of their outcomes is currently missing. In the present systematic review and meta‐analysis, we examined cohort studies and clinical trials published between 1980 and 2021 that reported, for DSM/ICD‐defined EDs, overall ED outcomes (i.e., recovery, improvement and relapse, all‐cause and ED‐related hospitalization, and chronicity); the same outcomes related to purging, binge eating and body weight status; as well as mortality. We included 415 studies (N=88,372, mean age: 25.7±6.9 years, females: 72.4%, mean follow‐up: 38.3±76.5 months), conducted in persons with anorexia nervosa (AN), bulimia nervosa (BN), binge eating disorder (BED), other specified feeding and eating disorders (OSFED), and/or mixed EDs, from all continents except Africa. In all EDs pooled together, overall recovery occurred in 46% of patients (95% CI: 44‐49, n=283, mean follow‐up: 44.9±62.8 months, no significant ED‐group difference). The recovery rate was 42% at <2 years, 43% at 2 to <4 years, 54% at 4 to <6 years, 59% at 6 to <8 years, 64% at 8 to <10 years, and 67% at ≥10 years. Overall chronicity occurred in 25% of patients (95% CI: 23‐29, n=170, mean follow‐up: 59.3±71.2 months, no significant ED‐group difference). The chronicity rate was 33% at <2 years, 40% at 2 to <4 years, 23% at 4 to <6 years, 25% at 6 to <8 years, 12% at 8 to <10 years, and 18% at ≥10 years. Mortality occurred in 0.4% of patients (95% CI: 0.2‐0.7, n=214, mean follow‐up: 72.2±117.7 months, no significant ED‐group difference). Considering observational studies, the mortality rate was 5.2 deaths/1,000 person‐years (95% CI: 4.4‐6.1, n=167, mean follow‐up: 88.7±120.5 months; significant difference among EDs: p<0.01, range: from 8.2 for mixed ED to 3.4 for BN). Hospitalization occurred in 26% of patients (95% CI: 18‐36, n=18, mean follow‐up: 43.2±41.6 months; significant difference among EDs: p<0.001, range: from 32% for AN to 4% for BN). Regarding diagnostic migration, 8% of patients with AN migrated to BN and 16% to OSFED; 2% of patients with BN migrated to AN, 5% to BED, and 19% to OSFED; 9% of patients with BED migrated to BN and 19% to OSFED; 7% of patients with OSFED migrated to AN and 10% to BN. Children/adolescents had more favorable outcomes across and within EDs than adults. Self‐injurious behaviors were associated with lower recovery rates in pooled EDs. A higher socio‐demographic index moderated lower recovery and higher chronicity in AN across countries. Specific treatments associated with higher recovery rates were family‐based therapy, cognitive‐behavioral therapy (CBT), psychodynamic therapy, and nutritional interventions for AN; self‐help, CBT, dialectical behavioral therapy (DBT), psychodynamic therapy, nutritional and pharmacological treatments for BN; CBT, nutritional and pharmacological interventions, and DBT for BED; and CBT and psychodynamic therapy for OSFED. In AN, pharmacological treatment was associated with lower recovery, and waiting list with higher mortality. These results should inform future research, clinical practice and health service organization for persons with EDs.

Eating disorders (EDs) are severe psychiatric conditions characterized by altered eating behavior, that can lead to severe weight loss and underweight, or to weight gain and obesity ¹ . They have been recently reclassified as “feeding and eating disorders” in the DSM‐5 ² , ³ and ICD‐11 ⁴ , ⁵ , ⁶ . According to these classifications, they encompass anorexia nervosa (AN), bulimia nervosa (BN), binge eating disorder (BED), and other specified feeding and eating disorders (OSFED) as the most common and studied conditions.

The most important psychopathological feature of AN and BN is the overvaluation of body shape and weight ⁷ , ⁸ , ⁹ , ¹⁰ . Individuals with AN are underweight, refuse to gain weight and/or deny the severity of underweight status with/without engaging in binge eating and compensatory behaviors. Binge episodes are defined as introducing an amount of food that is larger than what an average person would have eaten, in a short period of time, with sensations of loss of control over eating ¹¹ . Compensatory behaviors consist of purging behaviors (self‐induced vomiting, or using laxatives or diuretics) or excessive exercise to prevent weight gain or lose weight. People with BN are not underweight and engage in recurring episodes of binge eating and compensatory behaviors and/or fasting or compulsive exercise. Individuals with BED engage in recurring episodes of binge eating that are not followed by compensatory behaviors. The overvaluation of shape and weight is not a characteristic feature of BED, and this condition is often associated with, or leads to, obesity ¹² . OSFED is a residual category including individuals who do not meet full threshold criteria for the main EDs, and encompassing atypical AN, purging disorder, subthreshold BN and BED, and night eating syndrome. A proportion of these patients move to a diagnosis of a main ED over time ¹³ .

All EDs are marked by frequent psychiatric and physical comorbidity ¹⁴ , ¹⁵ , ¹⁶ , and impaired physical, social and work functioning ¹⁷ , ¹⁸ , ¹⁹ . People of all ages, ethnicities and socio‐economic conditions ²⁰ can be affected by EDs, although adolescents and young adults are particularly at risk, and the mean age of onset is decreasing ²¹ . AN is more common in women and starts earlier than BN and BED ²² . BN and BED show less gender differences and a higher prevalence in ethnic minorities than AN ²³ .

The etiopathogenesis of EDs is thought to be multifactorial, with models postulating the presence of predisposing factors (genetic vulnerability ²⁴ , ²⁵ , temperamental traits, and childhood traumatic experiences ²⁶ ), precipitating factors (the environmental context at the time of onset ¹ ), and maintaining factors (secondary aspects of the illness, such as brain adaptation induced by malnutrition, social isolation, and changes in the environment ²⁷ ). However, a clear understanding of this etiopathogenesis is currently lacking, although it would be essential to improve treatment effectiveness ²⁸ .

Access to treatment for EDs is inadequate, with only 20‐25% of individuals receiving professional consultation for their symptoms ²⁹ . Barriers to treatment access include stigma, lacking insight into the illness, shame, scarce availability of evidence‐based interventions, and fragmented or underfunded health services ³⁰ , ³¹ , which contribute to low recovery rates and frequent chronicity ³² . The complexity of EDs requires a multidisciplinary treatment approach to address psychological, environmental, nutritional, behavioral and physical problems, as well as mental health comorbidities ³³ , ³⁴ . Psychological and nutritional treatments are recommended by guidelines for all EDs ³⁵ , ³⁶ . Evidence‐based psychotherapies have been developed, but their effectiveness in adults with AN does not differ from treatment as usual (TAU) ³⁷ , and there is no superiority of a specific approach ³⁸ . In contrast, family‐based interventions have shown long‐term superiority on other active treatments in adolescents and young adults with AN and in adolescents with BN³⁹. Cognitive‐behavioral therapy (CBT) is the most validated treatment in people with BN and BED ²⁰ , with some indications of long‐term effectiveness ³⁹ . Pharmacological augmentation has been effective in the short term, namely antidepressants in BN and antidepressants or lisdexamfetamine in BED ³⁹ .

Studies reporting on outcomes of EDs have been heterogeneous with respect to definitions of recovery, relapse, remission and hospitalization; sample size, study design, duration of follow‐up, and overall quality ⁴⁰ , ⁴¹ , ⁴² , ⁴³ , ⁴⁴ , ⁴⁵ . Thus, a systematic review is needed which is comprehensive enough to explore outcome moderators and explain heterogeneity of findings. The most extensive reviews of outcomes in EDs have been published over one decade ago, and reported only on the course of AN and BN ⁴⁶ , ⁴⁷ . An update and extension of their findings, and an evaluation of outcomes also in BED and OSFED, are now timely.

Although several therapeutic interventions for EDs have been validated by research and are implemented in real‐world clinical practice, they have a different impact on patients with the same diagnosis ⁴⁸ . More tailored and individualized therapies are a research and clinical priority to overcome the “therapeutic stagnation” in EDs ⁴⁹ . A variety of predictors and moderators of treatment outcomes have been reported in patients with EDs ⁵⁰ , ⁵¹ , ⁵² , but the overall picture remains unclear.

The primary aim of this review and meta‐analysis was to explore clinically relevant outcomes of specific EDs – including recovery, improvement, relapse after recovery, hospitalization, chronicity and mortality – over different follow‐up times. Additional objectives included exploring the presence of moderators and mediators of the main outcomes within and across EDs, evaluating the proportion of patients migrating between ED diagnoses, and estimating the real‐world effectiveness of different interventions.

METHODS

Search strategy and inclusion/exclusion criteria

We conducted a PRISMA 2020 ⁵³ ‐compliant systematic review searching Embase, Medline and PsycINFO from 1980 to 2021, aiming to include prospective or database cohort studies as well as trials reporting on clinical outcomes of EDs. The search key included terms related to EDs and outcomes of interest (see supplementary information). Additionally, a manual search was conducted to identify further studies not detected by the systematic search, through Medline and Google Scholar.

Inclusion criteria were: a) original peer‐reviewed articles; b) published in English; c) based on controlled or non‐controlled trials, longitudinal database studies or prospective cohort studies, including patients with EDs (i.e., AN, BN, BED, OSFED) defined according to any version of the DSM or ICD; and d) reporting frequencies of at least one of the following outcomes: recovery, improvement, chronicity, all‐cause hospitalization, relapse after recovery, and mortality.

Exclusion criteria were: a) meta‐analyses, review articles and case reports/case series; b) retrospective studies (except database studies) and case‐control studies; c) animal studies; d) studies published before 1980; e) studies that did not report any binary outcome of interest; f) studies that included patients with ED symptoms but no full ED diagnosis; and g) studies with <10 participants.

Outcomes and data extraction

The co‐primary outcomes were recovery and chronicity of the overall ED symptomatology – which we defined as “overall” recovery and chronicity – and mortality. Recovery was defined as absence of ED symptoms or “good outcome” assessed by a validated scale (e.g., the Morgan‐Russell Outcome Assessment Schedule ⁵⁴ for AN, BN and mixed EDs). Chronicity was defined as continued presence of an ED diagnosis or “poor outcome” assessed by a validated scale (see also supplementary information).

Additional outcomes were recovery and chronicity of specific ED symptoms (i.e., binge eating, purging, abnormal weight), and all‐cause hospitalization. Moreover, we considered overall as well as specific binge eating, purging and weight improvement (i.e., symptom improvement or “intermediate outcome”) and relapse (i.e., symptom relapse after recovery), as well as ED‐related hospitalization (see also supplementary information).

In addition to outcomes, pairs of independent authors extracted the following data from eligible studies: bibliographic identifiers, country, year of data collection, primary component of the intervention, mean age of included sample at baseline, study design, treatment setting, proportion of females, mean body mass index (BMI) at baseline, duration of treatment, duration of illness, total follow‐up duration, proportion of persons with individual psychiatric comorbidities (i.e., major depressive disorder, anxiety disorders, obsessive‐compulsive disorder, substance use disorders, personality disorders, and history of self‐injurious behaviors). Moreover, a socio‐demographic index (accounting for income per capita, average educational attainment, and fertility rates) was assigned to each country ⁵⁵ , to explore regional differences.

The quality of observational studies was assessed with the Newcastle‐Ottawa scale ⁵⁶ . The risk of bias of trials was evaluated using the Cochrane's risk of bias (RoB) tool ⁵⁷ .

Statistical analyses

We conducted random‐effects meta‐analyses of the frequency of clinical outcomes. We used the longest time point if more than one was available. We reported the pooled percentage of individuals with the outcome of interest as well as the average follow‐up duration, in months, within each ED and pooling all EDs together. We also calculated the frequency of outcomes by follow‐up duration, considering the following time points: <2 months, 2 to <4 years, 4 to <6 years, 6 to <8 years, 8 to <10 years, and ≥10 years. For mortality, we also calculated deaths/1,000 person‐years considering observational studies.

We conducted subgroup analyses testing whether the frequency of outcomes differed across EDs. We also conducted subgroup analyses by decade of data collection, primary ingredient of treatment, age group, study design, treatment setting, and continent where the study was conducted.

We used multivariable mixed‐effects meta‐regression adjusted for mean age, illness duration, and duration of follow‐up, testing the following potential moderators or mediators: sample size, data collection year, percentage of females, mean BMI, duration of treatment, proportion of patients with the individual psychiatric comorbidities, and socio‐demographic index of the country where the study was conducted. The meta‐regression analysis with sample size also served to measure publication bias ⁵⁸ .

We conducted sensitivity analyses by number of outcome categories reported in eligible studies, to investigate if higher granularity and specificity of outcome description would affect the frequency of the outcome.

From meta‐regression analyses we computed beta, that is the change in log of the proportion of individuals with the outcome of interest for each unit change in the moderator. R packages used in these analyses were metaprop and metareg commands from the meta package ⁵⁹ , in R 4.1.3.

RESULTS

Database search results and characteristics of included studies

From an initial 7,929 hits, we ultimately included 415 studies, reporting data on 88,372 persons with EDs. The PRISMA flow chart is reported in Figure 1. The lists of included studies and of studies excluded after full text assessment, with references and the reason for exclusion, are available in the supplementary information.

Figure 1.

PRISMA flow chart. ED – eating disorder

Overall, 55.4% of the studies had an observational design. Patients had a weighted mean age of 25.7±6.9 years, and 72.4% were females. The mean follow‐up duration across all studies was 38.3±76.5 months. Studies were conducted most frequently in Europe (50.6%), followed by North America (38.8%), Oceania (5.5%), Asia (3.1%), South America (1.2%), and across multiple continents (0.7%).

Patients were most frequently diagnosed with AN (41.7%, n=173; N=37,160; mean follow‐up: 64.1±102.8 months); and progressively less frequently with BN (35.4%, n=147; N=23,197; mean follow‐up: 30.4±72.9 months); BED (17.1%, n=71; N=5,781; mean follow‐up: 8.8±16.0 months); OSFED (5.3%, n=22; N=11,930; mean follow‐up: 98.6±205.2 months); and mixed EDs (14.7%, n=61; N=10,304; mean follow‐up: 31.1±46.9 months).

Only 12% of observational studies had high quality, and 24% of randomized controlled trials (RCTs) had low risk of bias.

Frequency of outcomes across and within eating disorders

The frequencies of primary and additional overall outcomes across EDs, and their rates at the different follow‐up time points, are visualized in Figures 2 and 3.

Figure 2.

Overall recovery, improvement, relapse, hospitalization, chronicity and mortality across persons with eating disorders (EDs). FU – follow‐up, AN – anorexia nervosa, BN – bulimia nervosa, BED – binge eating disorder, OSFED – other specified feeding and eating disorders, NA – not applicable, N/C – not calculable.

Figure 3.

Overall recovery, improvement, relapse, hospitalization and chronicity in persons with all eating disorders pooled together over follow‐up duration. NA – not applicable.

Pooling all EDs together, overall recovery occurred in 46% of patients (95% CI: 44‐49, n=283, mean follow‐up: 44.9±62.8 months), without a significant difference among EDs (p=0.17). The recovery rate was 42% at <2 years, 43% at 2 to <4 years, 54% at 4 to <6 years, 59% at 6 to <8 years, 64% at 8 to <10 years, and 67% at ≥10 years. There was an increase of the recovery rate over follow‐up in AN, BN and OSFED, whereas the rate decreased (from 57% at <2 years to 16% at 6 to <8 years) in BED (see supplementary information).

Overall chronicity occurred in 25% of patients (95% CI: 23‐29, n=170, mean follow‐up: 59.3±71.2 months), without a significant difference among EDs (p=0.23). The chronicity rate was 33% at <2 years, 40% at 2 to <4 years, 23% at 4 to <6 years, 25% at 6 to <8 years, 12% at 8 to <10 years, and 18% at ≥10 years. There was a decrease of the chronicity rate over follow‐up in AN, BN and OSFED, whereas the rate increased (from 17% at <2 years to 72% at 4 to <6 years) in BED (see supplementary information).

Mortality during follow‐up occurred in 0.4% of patients (95% CI: 0.2‐0.7, n=214, mean follow‐up: 72.2±117.7 months), without a significant difference among EDs (p=0.058), due to large confidence intervals. Mortality increased with longer follow‐up duration across and within EDs (see Figure 4). When focusing on observational studies, the mortality rate was 5.2 deaths/1,000 person‐years (95% CI: 4.4‐6.1, n=167, mean follow‐up: 88.7±120.5 months; significant difference among EDs: p<0.01, range: from 8.2 for mixed ED to 3.4 for BN) (see also supplementary information).

Figure 4.

Mortality in persons with eating disorders (EDs) over follow‐up duration. AN – anorexia nervosa, BN – bulimia nervosa, BED – binge eating disorder, OSFED – other specified feeding and eating disorders, NA – not applicable.

Hospitalization occurred in 26% of patients (95% CI: 18‐36, n=18, mean follow‐up: 43.2±41.6 months), with a significant difference among EDs (p<0.001). It was highest in AN (32%, 95% CI: 23‐43, n=14, mean follow‐up: 47.4±44.7 months) and lowest in BN (4%, 95% CI: 1‐10, n=2, mean follow‐up: 26.9±17.3 months). ED‐related hospitalization (reported in 11 studies, of which 10 focusing on AN) occurred in 34% of ED patients (95% CI: 24‐47, N=896) and 35% of AN patients (95% CI: 24‐49, N=777).

Among additional outcomes, improvement occurred in 28% of patients (95% CI: 25‐32, n=101, mean follow‐up: 54.9±69.8 months), with a significant difference among EDs (p=0.02). It was highest in BN (40%, 95% CI: 31‐50, n=31, mean follow‐up: 26.3±33.2 months) and lowest in AN (24%, 95% CI: 20‐29, n=56, mean follow‐up: 82.8±80.5 months).

Relapse after recovery occurred in 26% of patients (95% CI: 21‐31, n=45, mean follow‐up: 42.8±45.4 months), with a significant difference among specific EDs (p<0.001). It was highest in BN (31%, 95% CI: 25‐39, n=19, mean follow‐up: 37.7±40.4 months) and lowest in BED (10%, 95% CI: 6‐17, n=1, follow‐up: 45.1 months).

The results of sensitivity analyses accounting for heterogeneity of definitions and granularity in defining different outcome categories were largely consistent with the main findings (see supplementary information).

Frequency of eating disorder‐specific symptom outcomes across and within eating disorders

The frequencies of outcomes of specific ED symptoms are reported in Figure 5 and supplementary information.

Figure 5.

Purging, binge eating, and abnormal weight recovery and chronicity across persons with eating disorders (EDs). FU – follow‐up, AN – anorexia nervosa, BN – bulimia nervosa, BED – binge eating disorder, OSFED – other specified feeding and eating disorders, NA – not applicable.

Pooling all EDs, the rate of recovery was 43% for purging (95% CI: 37‐49, n=30, no ED subgroup difference). It was 43% for binge eating (95% CI: 37‐48, n=67), being highest in OSFED (83%; 95% CI: 53‐96, n=1) and lowest in AN (31%; 95% CI: 23‐41, n=3), with a significant subgroup difference (p=0.018). Recovery of weight occurred in 49% of patients (95% CI: 43‐55, n=48), being highest in OSFED (69%, 95% CI: 63‐74, n=1) and lowest in BED (17%, 95% CI: 12‐24, n=1), with a significant ED subgroup difference (p<0.001).

Pooling all EDs, the rate of chronicity was 23% for purging (95% CI: 15‐34, n=13, no ED subgroup difference); and 29% for binge eating (95% CI: 22‐37, n=22, no ED subgroup difference). It was 23% for abnormal weight (95% CI: 16‐31, n=28), being highest in AN (31%; 95% CI: 22‐41, n=20) and lowest in BN (7%; 95% CI: 6‐10, n=4), with a significant ED subgroup difference (p<0.001).

Diagnostic migration across eating disorders

Frequencies of diagnostic migration across EDs are reported in Table 1. From AN, 8% of patients migrated to BN and 16% to OSFED. From BN, 2% migrated to AN, 5% to BED, and 19% to OSFED. From BED, 9% migrated to BN and 19% to OSFED. From OSFED, 7% migrated to AN and 10% to BN.

Table 1.

Conversion rates among eating disorders

	n	N	Prevalence (95% CI)	I2
From anorexia nervosa
To bulimia nervosa	35	5,758	0.08 (0.05‐0.11)	96
To other specified feeding and eating disorders	23	3,211	0.16 (0.11‐0.21)	90
From bulimia nervosa
To anorexia nervosa	12	3,121	0.02 (0.01‐0.04)	76
To binge eating disorder	9	957	0.05 (0.02‐0.13)	89
To other specified feeding and eating disorders	13	1,561	0.19 (0.12‐0.29)	88
From binge eating disorder
To bulimia nervosa	4	225	0.09 (0.02‐0.31)	89
To other specified feeding and eating disorders	2	57	0.19 (0.11‐0.32)	59
From other specified feeding and eating disorders
To anorexia nervosa	4	236	0.07 (0.02‐0.21)	12
To bulimia nervosa	5	288	0.10 (0.05‐0.18)	46

Moderators and mediators of outcomes across and within eating disorders in subgroup analyses

A synopsis of statistically significant moderators and mediators of outcomes across and within EDs in subgroup analyses is provided in Table 2 (see supplementary information for a complete report).

Table 2.

Statistically significant moderators of recovery, chronicity and mortality across eating disorders (EDs) identified by subgroup analyses

Moderator	AN	BN	BED	OSFED	All EDs
Age group with 26‐50% recovery	Adults	Adults		Adults	Adults
Age group with 51‐75% recovery	Children/adolescents		Adults		Children/adolescents
Age group with 76‐100% recovery		Children/adolescents	Children/adolescents	Children/adolescents
Primary treatment component with 0‐25% recovery	Pharmacological
Primary treatment component with 26‐50% recovery	Nutritional, FBT, CBT, psychodynamic, multidisciplinary (specific)	Self‐help, CBT, psychodynamic, nutritional, pharmacological	CBT, pharmacological, DBT	CBT, multidisciplinary (specific), psychodynamic	FBT, self‐help, CBT, psychodynamic, DBT, pharmacological, multidisciplinary (specific)
Primary treatment component with 51‐75% recovery Primary treatment component with 76‐100% recovery		Multidisciplinary (specific), DBT	Nutritional		Nutritional
Continents with 26‐50% recovery		Europe, North America		Europe
Continents with 51‐75% recovery		Asia, Oceania		North America, Asia, Oceania
Age group with 51‐75% chronicity				Adults
Age group with 26‐50% chronicity	Adults			Children/adolescents	Adults
Age group with 0‐25% chronicity	Children/adolescents		Adults		Children/adolescents
Continents with 26‐50% chronicity	Oceania, North America, South America				South America, North America, Oceania
Continents with 0‐25% chronicity	Asia, Europe				Asia, Europe
Age group with 1‐5% mortality	Adults
Age group with <1% mortality	Children/adolescents	Adults, children/adolescents			Children/adolescents, adults
Primary treatment component with >5% mortality	Waiting list				Waiting list
Primary treatment component with 1‐5% mortality	CBT, psychodynamic			CBT	CBT, psychodynamic, FBT
Primary treatment component with <1% mortality	FBT, nutritional, multidisciplinary (specific)			Psychodynamic	Psychoeducation, nutritional, multidisciplinary (specific)

AN – anorexia nervosa, BN – bulimia nervosa, BED – binge eating disorder, OSFED – other specified feeding and eating disorders, CBT – cognitive‐behavioral therapy, DBT – dialectical behavioral therapy, FBT – family‐based therapy. Treatment‐as‐usual and interventions tested in one study only are not included. Results of additional subgroup analyses are reported in the supplementary information.

Children/adolescents had significantly higher recovery rates than adults across and within all EDs; lower chronicity rates across all EDs and in AN; and lower mortality rates across EDs and within AN and BN.

Pooling all EDs, nutritional intervention was the primary treatment component associated with the largest recovery. CBT was the only specific intervention that had recovery rates of 26% or higher across all EDs. In AN, additional specific treatments that were associated with higher rates of recovery in ≥2 studies were family‐based therapy, psychodynamic therapy, and multidisciplinary specific treatment. In BN, they were self‐help, psychodynamic therapy, pharmacological treatment, multidisciplinary specific treatment, and DBT. In BED, they were pharmacological treatment, and DBT. In OSFED, they were multidisciplinary specific treatment and psychodynamic therapy. In AN, the use of pharmacotherapy was associated with low recovery rates. Waiting list was associated with the highest mortality rate in both pooled EDs and AN.

A significant difference in rates of recovery across continents only emerged for BN (highest in Asia, lowest in North America) and OSFED (highest in North America, lowest in Europe). Differences across continents regarding chronicity only emerged for pooled EDs (lowest chronicity in Asia, highest in South America) and AN (lowest chronicity in Asia, highest in Oceania). No differences among continents emerged for mortality either across or within EDs.

Moderators of outcomes across and within specific eating disorders in meta‐regression analyses

Across all EDs, recovery increased with mean baseline BMI, and percentage of patients with obsessive‐compulsive disorder, and decreased with proportion of patients with self‐injurious behaviors. In AN, a higher recovery rate was associated with higher percentage of patients with obsessive‐compulsive disorder, less recent data collection, and lower country socio‐demographic index. In BN, higher recovery rate was associated with longer duration of treatment, higher proportion of patients with major depressive disorder, and higher number of treatment ingredients. Lower chronicity rate in AN was associated with smaller sample size and lower country socio‐demographic index (see Table 3).

Table 3.

Statistically significant moderators of recovery, chronicity and mortality across eating disorders (EDs) identified by multivariable meta‐regression analyses

Moderator	AN	BN	BED	OSFED	All EDs
		Recovery
Data collection year	Beta=–0.001 (–0.002 to 0.000), n=64, p=0.0081
Treatment duration		Beta=0.019 (0.004‐0.034), n=52, p=0.011
Mean body mass index					Beta=0.102 (0.042‐0.162), n=91, p=0.00088
% obsessive‐compulsive disorder	Beta=3.517 (1.452‐5.581), n=11, p=0.00084				Beta=3.576 (1.579‐5.574), n=12, p<0.00045
% major depressive disorder		Beta=3.928 (2.213‐5.643), n=7, p<0.0001
% self‐injurious behaviors					Beta=–4.452 (–6.832 to –2.072), n=8, p=0.00025
Country socio‐demographic index	Beta=–6.473 (–11.503 to –1.443), n=63, p=0.012
Number of treatment ingredients		Beta=0.172 (0.011‐0.332), n=52, p=0.036
		Chronicity
Sample size	Beta=0.002 (0.001‐0.003), n=49, p=0.00019
Country socio‐demographic index	Beta=6.430 (0.539‐12.321), n=49, p=0.032
		Mortality
Data collection year	Beta=–0.109 (–0.177 to –0.041), n=52, p=0.0016				Beta=–0.066 (–0.125 to –0.007), n=85, p=0.028
% females		Beta=–3.401 (–6.001 to –0.801), n=31, p=0.01
Mean body mass index	Beta=–0.799 (–1.264 to –0.335), n=34, p=0.00074				Beta=–0.290 (–0.496 to –0.083), n=57, p=0.0059
Treatment duration	Beta=–0.054 (–0.104 to –0.005), n=52, p=0.031
Country socio‐demographic index				Beta=–15.279 (–29.432 to –1.125), n=7, p=0.034

Beta values are reported with 95% CIs. The analyses were adjusted for mean age, illness duration and duration of follow‐up. AN – anorexia nervosa, BN – bulimia nervosa, BED – binge eating disorder, OSFED – other specified feeding and eating disorders. All results are available in the supplementary information.

For mortality, across all EDs pooled together, rates decreased with more recent data collection and higher baseline BMI, which was confirmed in AN, where also longer treatment duration was associated with lower mortality. In BN, a higher proportion of females was associated with lower mortality rates, while in OSFED lower mortality was associated with higher socio‐demographic index (see Table 3).

DISCUSSION

This systematic review meta‐analyzed 415 studies from all continents, except Africa, that investigated clinically relevant outcomes of specific EDs – including recovery, improvement, relapse after recovery, hospitalization, chronicity and mortality – over different follow‐up times. Additional objectives included exploring the presence of moderators and mediators of the main outcomes within and across EDs, evaluating the proportion of patients migrating between ED diagnoses, and estimating the real‐world effectiveness of different interventions.

This is the first meta‐analysis that provides a systematic atlas of the clinical outcomes of EDs – including AN, BN, BED and OSFED – over up to more than 10 years of follow‐up. Its findings can inform the clinical management of persons with EDs, the relevant research agenda, as well as appropriate health services and resource allocation aiming to improve outcomes in people with EDs.

Results indicate that recovery rates are similar among different EDs when considering overall symptoms and purging, but different regarding binge eating (with OSFED having the highest and AN the lowest recovery rate) and abnormal weight (with OSFED having the highest and BED the lowest recovery rate). BED is frequently associated with overweight or obesity, which can cause physical complications, including arthritis, diabetes, cardiovascular and respiratory conditions, that limit the ability to exercise, might contribute to further weight gain and are barriers to weight loss ⁶⁰ , ⁶¹ , ⁶² . Moreover, BED is associated with mental comorbidities, including depressive and bipolar disorder as well as borderline personality disorder ⁶⁰ , which are associated with lower physical activity levels, as well as poor physical and mental health status, which are each barriers to a healthy weight status ⁶³ , ⁶⁴ , ⁶⁵ , ⁶⁶ , ⁶⁷ . It is crucial to intervene early in BED, to avoid that the network of psychological, behavioral and physical symptoms sustaining obesity clusters and aggregates to the point of becoming refractory to treatment ⁶⁸ , ⁶⁹ , ⁷⁰ , ⁷¹ .

AN had the most severe outcomes among EDs, including highest relapse and chronicity of weight loss, and highest rates of hospitalization. These outcomes are consistent with the numerous physical health complications that are associated with underweight and malnourishment ⁷² , ⁷³ , ⁷⁴ , ⁷⁵ , ⁷⁶ , ⁷⁷ , and with the importance of early weight gain for a positive disease course ⁷⁸ , ⁷⁹ . BN, instead, had the highest rates of overall improvement and relapse, suggesting a more episodic course compared to other EDs.

Recovery is not achieved by more than half of patients with EDs. Moreover, pooled across EDs, 26% experience relapses after recovery, and 26% require hospitalization during follow‐up. The overall mortality risk in EDs is 0.4% (range: 0‐1.4%), which is a relatively high rate for individuals at a mean age of 25.7±6.9 years. Moreover, the mean follow‐up period of 38.3±76.5 months was likely too short to capture the full mortality risk, and differential attrition rate may have affected the capturing of mortality in sicker patients, who may have been less likely to remain in longer‐term follow‐up. Also, since studies with longer duration reported higher mortality risk, future representative studies following individuals with EDs for sufficient periods of time will likely report higher lifetime mortality rates.

Recovery rates increased and chronicity rates decreased over follow‐up across all EDs, except for BED. The significant effect of time indicates that short‐term RCTs might be prone to type I error, with shorter trials overestimating the efficacy of experimental interventions versus TAU, which might not hold true at longer follow‐up. In BED, an opposite trend of outcomes emerged over follow‐up, with recovery decreasing and chronicity increasing over time, confirming that early remission is crucial, before ED symptoms, overweight or obesity, and physical health implications cluster tightly.

The most common diagnostic migration from AN, BN and BED is to OSFED (16% to 19%), with lower rates in the opposite direction (7% to 10%), or among other EDs (2% to 9%). Diagnostic migration from AN, BN or BED to OSFED implies that some symptoms of a specific ED have improved, while other symptoms have persisted, so that not all diagnostic criteria are met. OSFED may evolve in three different directions: relapse of AN (7%) or BN (10%), a chronic course (34%), or transition to recovery (57%).

As OSFED is a less well‐defined and more heterogeneous diagnosis, it may pose a challenge to clinicians when deciding on the best course of treatment ⁸⁰ . Indeed, it may be implicitly considered as a residual category and a less severe disorder ⁴ . However, results from this meta‐analysis indicate that recovery and chronicity rates are not different from those of AN or BN, and that the risk of relapse of OSFED actually increases with follow‐up time, from 8% within 2 years to 52% after over 10 years.

Children and adolescents had the highest recovery rate in pooled EDs, and the lowest chronicity rate in AN. These results are in line with the staging model of EDs ²⁷ , which suggests that, while illness progresses, neurobiological and psychosocial maintaining factors develop and make persons suffering from EDs more resistant to treatment ⁸¹ . These data further indicate the importance of providing sufficient resources to enable early diagnosis and treatment, and of reducing barriers and delays to treatment access in young people ⁸² , ⁸³ . For this purpose, services and policy makers should look at family and general practitioner education ³¹ , and the strict division still existing in many settings between child and adult mental health services should be overcome ³⁴ , ⁸⁴ , ⁸⁵ .

Among interventions for EDs included in more than one study, treatments with a primary nutritional component were associated with the highest rates of overall recovery. Prescription of a healthy meal plan, psychoeducation on physiologic nutritional needs, and supervised meals in intensive or family‐based settings are crucial to normalize the eating pattern, and interrupt some of the behavioral symptoms. Moreover, psychopathological symptoms can indirectly benefit from the improvement of eating behaviors.

However, a treatment exclusively focused on nutrition would not offer the necessary insight on the personalized cycle and network of symptoms of each person's ED, nor would it train coping skills to address environmental (interpersonal) triggers, target body checking, address mental comorbidities, and ultimately facilitate full recovery and prevent relapse.

Our findings support the role of family‐based therapy and CBT in AN ³⁹ . All main guidelines ³⁵ indicate the effectiveness of family‐based interventions in adolescents with AN. Additionally, according to a recent umbrella review ³⁹ , also young adults with AN may benefit from these interventions. Most guidelines also suggest the effectiveness of CBT, and two of them point to CBT as the first‐line individual psychotherapy for people with AN ³⁵ .

Importantly, waiting list was associated with the highest mortality in AN. This finding is of crucial clinical relevance, and should encourage ED services offering treatment for AN to avoid passive waiting list, and to offer psychoeducational elements and active monitoring to those awaiting admission, to capture worsening of behavioral and clinical symptoms early.

Our data suggest that self‐help, CBT and DBT are effective in BN. Most guidelines outline CBT as the first‐line psychotherapy for this condition ³⁵ , while the National Institute for Health and Care Excellence (NICE) guidelines ⁸⁶ point to guided self‐help as first‐line treatment, probably in light of cost‐effectiveness considerations. Compared with guidelines ³⁵ , our finding concerning DBT is novel. DBT is one of the “third wave” psychotherapies, which include CBT elements (i.e., skills training, exposure, self‐monitoring) but focus more on the context and on interpersonal functioning, which are maintaining factors of EDs ⁸⁷ , ⁸⁸ . DBT can also address comorbid borderline personality symptoms, which do have an impact on outcomes of EDs, especially in BN and AN binge‐purge type.

In individuals with BED, CBT and pharmacotherapy are supported by the current findings, which is in line with guidelines ³⁵ that suggest the use of CBT, selective serotonin reuptake inhibitors, anticonvulsants (topiramate) and anti‐obesity (orlistat) drugs. Lisdexamfetamine has been approved only in some countries, and long‐term effectiveness of medications has not been proven ⁸⁹ , ⁹⁰ . The efficacy of stimulants or topiramate in BED might be partially mediated by appetite suppression. Improvement in cognition as well as in impulse control and comorbid anxiety and depressive symptoms is also likely to mediate the efficacy of pharmacological options to treat BED ⁹¹ .

Beyond age group and specific treatment components, other factors moderated outcomes of EDs in this meta‐analysis. For example, comorbid mental disorders or symptoms were associated with better or worse outcomes across EDs, and had the largest effect sizes among outcome moderators. This finding reflects the central role of non‐ED psychopathology in maintaining the symptoms of EDs, and the importance of targeting comorbid mental conditions when treating EDs ⁷⁰ , ⁹² . OCD and major depressive disorder moderated higher rates of recovery in AN and BN, respectively, possibly because these conditions are pharmacologically addressable. Self‐injurious behaviors were associated with lower recovery rates in pooled EDs. They can be considered a proxy of borderline personality disorder traits ⁹³ , which do not respond to most pharmacological treatments ⁹⁴ , but which do respond to DBT ⁹⁵ .

In BN, a higher number of treatment components and a longer duration of treatment were associated with higher rates of recovery. These results reflect the multidimensional nature of symptoms in BN, the frequent presence of comorbid mental disorders and symptoms, and the need to provide multidisciplinary care that is not too limited in time ³⁹ , ⁷⁰ , ⁹⁶ , ⁹⁷ .

Pooling all EDs, mortality decreased in more recent studies, although this finding was driven by studies in AN. While education and training on AN vary across countries, and are frequently suboptimal ⁹⁸ , ⁹⁹ , our data suggest that physical health of subjects with AN has improved over decades, possibly reflecting decreased stigma ¹⁰⁰ and increasing knowledge and attention to physical comorbidities ¹⁰¹ .

However, as reflected by the decrease in recovery rates of AN over the last decades, decreasing mortality is not enough. AN symptoms can evolve to a state of severe and enduring illness, in which patients describe their life as permeated by the disease, that assumes an identity role ¹⁰² . On the other hand, residual eating‐related cognitions may persist throughout the life of a substantial group of patients, without being necessarily incompatible with a reasonable degree of psychological well‐being ¹⁰³ . The issue of personal recovery, implying an emphasis on the affected persons’ perceptions and values, has not been explored sufficiently in people with EDs, representing an important potential focus for future research.

Higher BMI was associated with decreased mortality in AN, supporting the current classification of severity in the DSM‐5.

A higher proportion of males predicted higher mortality in BN. The prevalence of EDs in males is increasing at a faster rate than in females ¹⁰⁴ . Although some data point to similarities with females, some qualitative differences have been detected, such as a more frequent history of overweight and drive for leanness and muscularity, which can promote the use of specific drugs (i.e., anabolic steroids) ¹⁰⁵ . The sensitivity of most commonly used tools to quantify ED symptoms may be suboptimal in males ¹⁰⁶ , and EDs may be under‐detected in these individuals ¹⁰⁶ , ¹⁰⁷ , ¹⁰⁸ . Our findings call for additional research examining sex and gender as moderators of different clinical characteristics and outcomes in EDs ¹⁰⁵ , ¹⁰⁹ .

A higher socio‐demographic index moderated lower recovery and higher chronicity in AN. Countries with a higher index might culturally be more impregnated with thinness pressure, while other sociocultural systems might be less thinness‐ or body image‐centred ¹¹⁰ , ¹¹¹ .

This meta‐analysis pools quantitative results from over twice as many studies as the previous available reviews ⁴⁴ , ⁴⁶ , ⁴⁷ , which were conducted over 20 to almost 15 years ago, and did not meta‐analyze the data. Our work confirms prior findings and adds novel results. For instance, in AN, this work confirms the rates of chronicity being above 20%, and at the same time the need to communicate hope even to persons with severe clinical presentations, as rates of recovery increased and rates of chronicity decreased over the patients’ follow‐up time. However, the outcome of EDs in general has not improved over decades, apart from lower mortality in AN.

This meta‐analysis adds evidence from observational and interventional studies about the impact of several interventions on clinical outcomes in EDs, accounting for longer follow‐up and quantitatively testing moderators and mediators of outcomes in subgroup and multivariable analyses, thus informing clinical care and organization of services. Results from these additional approaches indicate, for example, the need to offer nutritional interventions and avoid unmonitored waiting list status.

This work has several strengths. First, it covers over 40 years of observational and interventional studies, providing a comprehensive overview of the current knowledge on EDs. Second, the inclusion of a large number of studies allowed the exploration and identification of several moderators and mediators, providing insights that can inform research, clinical practice and policy‐making. Third, this work is unique in that it pooled data from studies conducted across all continents, except for Africa, providing a global perspective on EDs. Fourth, including observational studies allowed the measurement of long‐term outcomes, enabling a more generalizable assessment of the effectiveness of different interventions and the frequency of relevant outcomes in people with EDs.

The meta‐analysis also has some limitations. First, not limiting ourselves to RCTs, we could not compare specific treatments or rank them. These meta‐analyses already exist ³⁷ , ³⁸ , but generally have a short time frame and cannot capture all of the clinically relevant outcomes this work focused on. Moreover, compared to RCTs, the inclusion of observational studies allowed us to capture and comment on a population that is more representative of real‐world patients, which supports the generalizability of the findings. Second, including both RCTs and observational studies might have increased heterogeneity of the samples to some degree. Third, the diagnostic criteria for EDs in DSM and ICD have changed over time, and this may have influenced the results of this study.

Fourth, we did not conduct a meta‐analysis of diagnostic migration from EDs to other mental disorders, because this was beyond the scope of this work, and future systematic reviews should quantify rates of this migration. Fifth, psychosocial determinants of health (i.e., early life trauma or emotion regulation), that have more recently emerged as outcome predictors ¹¹² , were not widely available in the meta‐analyzed studies, and could thus not be taken into account. Sixth, findings do not include outcomes of EDs in Africa, where research in this field should be promoted. Seventh, we did not investigate the impact of ED subtypes – i.e., restricting versus binge‐purge AN, or atypical AN – as data were too limited concerning the outcomes this meta‐analysis focused on. Eighth, the definition of outcomes was heterogeneous in the included studies. In order to account for this heterogeneity of definitions and granularity in defining different outcome categories, we have conducted several sensitivity analyses by the number of outcome categories, whose results were largely consistent with the main findings. Finally, despite the large number of eligible studies, some outcomes were based on only few studies, and the relative findings should be considered preliminary.

In conclusion, this systematic review and meta‐analysis contributes to the understanding of EDs and provides insights into their treatment and real‐world outcomes, both overall and regarding specific EDs. The results highlight the severe course of EDs, although there has been a decrease in mortality rates over time. It is imperative that patients with EDs are identified early and moved from waitlists to active care as quickly as possible, with a particular focus on those with a low BMI and comorbid self‐injurious behaviors. Comorbid depression and OCD are treatable and should be monitored and addressed.

It is also important to identify and treat EDs in children and adolescents as early as possible, to improve outcomes and prevent delays in managing EDs into adulthood. Management plans should include a nutritional intervention across all EDs. For AN, a multi‐component intervention should include family‐based therapy for children/adolescents and young adults, as well as CBT. Self‐help, CBT and DBT are effective interventions for BN, while CBT and pharmacotherapy are recommended for BED.

There is a need for long‐term observational studies to fully capture outcomes of individuals with EDs. International consensus should be reached in defining recovery, improvement, chronicity and relapse in this population. The issue of personal recovery in people with EDs should be explicitly addressed by research. Future studies should also aim to improve the detection of moderators and mediators that can help stratify patient subgroups, in order to allow a more personalized treatment approach to persons with EDs.